diff --git a/.github/scripts/i18n/README.md b/.github/scripts/i18n/README.md
index f29b05dd5..8956725eb 100644
--- a/.github/scripts/i18n/README.md
+++ b/.github/scripts/i18n/README.md
@@ -42,6 +42,11 @@ Configure explicit paths in `translation-config.json` → `chunked_files`, or re
 on `auto_chunk` (default: body ≥ 10k chars and ≥ 4 `##` sections) to auto-enable
 `heading_sections`.
 
+Changelog `<Update description="…">` dates are **derived from English** and
+localized automatically (`ja`/`zh`: `YYYY年M月D日`, `ko`: `YYYY년 M월 D일`) after
+each block is translated or re-serialized — English month names should not
+remain in translated files.
+
 During a chunked run the script:
 
 1. Parses English into blocks (intro + each `##` section).
@@ -49,6 +54,10 @@ During a chunked run the script:
 3. Translates only pending blocks (plus frontmatter when needed).
 4. Checkpoints after every block so a failed run can resume.
 
+`translationBlockHashes` keys are written in **descending semver order** for
+changelog (`v0.25.1` before `v0.25.0`), matching the canonical `<Update>`
+sequence. Long pages use **English document order** for section hashes.
+
 Example:
 
 ```bash
diff --git a/.github/scripts/i18n/chunked-translate.ts b/.github/scripts/i18n/chunked-translate.ts
index 7f2b176b0..6298f77fb 100644
--- a/.github/scripts/i18n/chunked-translate.ts
+++ b/.github/scripts/i18n/chunked-translate.ts
@@ -93,6 +93,20 @@ export function parseFrontmatterAndBody(content: string): { frontmatter: string;
   return { frontmatter: `${match[1]}\n`, body: match[2] };
 }
 
+function parseBlockHashEntry(
+  line: string
+): { label: string; hash: string } | null {
+  const quoted = line.match(/^\s{2}("(?:\\.|[^"\\])*"):\s*"?([a-f0-9]{8})"?\s*$/);
+  if (quoted) {
+    return { label: JSON.parse(quoted[1]!) as string, hash: quoted[2]! };
+  }
+  const plain = line.match(/^\s{2}([^:]+):\s*"?([a-f0-9]{8})"?\s*$/);
+  if (plain) {
+    return { label: plain[1]!.trim(), hash: plain[2]! };
+  }
+  return null;
+}
+
 /** Parse `translationBlockHashes` YAML map from frontmatter body. */
 export function parseBlockHashesFromFrontmatter(fmBody: string): Record<string, string> {
   const out: Record<string, string> = {};
@@ -104,14 +118,33 @@ export function parseBlockHashesFromFrontmatter(fmBody: string): Record<string,
       continue;
     }
     if (inMap) {
-      const quoted = line.match(/^\s{2}("(?:\\.|[^"\\])*"):\s*"?([a-f0-9]{8})"?\s*$/);
-      if (quoted) {
-        out[JSON.parse(quoted[1]!)] = quoted[2]!;
+      const entry = parseBlockHashEntry(line);
+      if (entry) {
+        out[entry.label] = entry.hash;
         continue;
       }
-      const plain = line.match(/^\s{2}([^:]+):\s*"?([a-f0-9]{8})"?\s*$/);
-      if (plain) {
-        out[plain[1]!.trim()] = plain[2]!;
+      if (/^[A-Za-z_][\w-]*:/.test(line)) {
+        inMap = false;
+      }
+    }
+  }
+  return out;
+}
+
+/** Block labels in frontmatter file order (not sorted). */
+export function parseBlockHashLabelOrderFromFrontmatter(fmBody: string): string[] {
+  const labels: string[] = [];
+  const lines = fmBody.split("\n");
+  let inMap = false;
+  for (const line of lines) {
+    if (/^translationBlockHashes:\s*$/.test(line)) {
+      inMap = true;
+      continue;
+    }
+    if (inMap) {
+      const entry = parseBlockHashEntry(line);
+      if (entry) {
+        labels.push(entry.label);
         continue;
       }
       if (/^[A-Za-z_][\w-]*:/.test(line)) {
@@ -119,12 +152,30 @@ export function parseBlockHashesFromFrontmatter(fmBody: string): Record<string,
       }
     }
   }
-  return out;
+  return labels;
 }
 
-export function formatBlockHashesYaml(blockHashes: Record<string, string>): string {
+export function blockHashLabelOrderDrifts(
+  enLabels: string[],
+  storedLabels: string[]
+): boolean {
+  if (enLabels.length !== storedLabels.length) return true;
+  return enLabels.some((label, i) => label !== storedLabels[i]);
+}
+
+export function formatBlockHashesYaml(
+  blockHashes: Record<string, string>,
+  labelOrder: string[]
+): string {
   const lines = ["translationBlockHashes:"];
-  for (const label of Object.keys(blockHashes).sort()) {
+  const seen = new Set<string>();
+  for (const label of labelOrder) {
+    if (!(label in blockHashes) || seen.has(label)) continue;
+    lines.push(`  ${JSON.stringify(label)}: ${blockHashes[label]}`);
+    seen.add(label);
+  }
+  for (const label of Object.keys(blockHashes)) {
+    if (seen.has(label)) continue;
     lines.push(`  ${JSON.stringify(label)}: ${blockHashes[label]}`);
   }
   return lines.join("\n");
@@ -177,12 +228,13 @@ export function setChunkedTranslationMeta(
   content: string,
   fileHash: string,
   enPath: string,
-  blockHashes: Record<string, string>
+  blockHashes: Record<string, string>,
+  labelOrder: string[]
 ): string {
   const metaLines = [
     `translationSourceHash: ${fileHash}`,
     `translationFrom: ${enPath}`,
-    formatBlockHashesYaml(blockHashes),
+    formatBlockHashesYaml(blockHashes, labelOrder),
   ];
   const metaBlock = metaLines.join("\n");
 
@@ -332,8 +384,159 @@ export function parseUpdateBlocks(body: string): ContentBlock[] {
   return blocks;
 }
 
+/** Parse ComfyUI changelog label like `v0.25.1` into numeric segments. */
+export function parseChangelogVersion(label: string): number[] {
+  const m = label.match(/^v?(\d+(?:\.\d+)*)/i);
+  if (!m) return [];
+  return m[1]!.split(".").map((n) => Number(n));
+}
+
+/** Descending semver compare for changelog `<Update label="…">` values (newest first). */
+export function compareChangelogVersionsDesc(a: string, b: string): number {
+  const pa = parseChangelogVersion(a);
+  const pb = parseChangelogVersion(b);
+  const len = Math.max(pa.length, pb.length);
+  for (let i = 0; i < len; i++) {
+    const da = pa[i] ?? 0;
+    const db = pb[i] ?? 0;
+    if (da !== db) return db - da;
+  }
+  return b.localeCompare(a);
+}
+
+export function canonicalChangelogLabelOrder(labels: string[]): string[] {
+  return [...labels].sort(compareChangelogVersionsDesc);
+}
+
+export function sortUpdateBlocksByVersion(blocks: ContentBlock[]): ContentBlock[] {
+  return [...blocks].sort((a, b) => compareChangelogVersionsDesc(a.label, b.label));
+}
+
+export function orderSlotsForStrategy(strategy: ChunkStrategy, slots: BlockSlot[]): BlockSlot[] {
+  if (strategy !== "update_blocks") return slots;
+  return [...slots].sort((a, b) => compareChangelogVersionsDesc(a.label, b.label));
+}
+
+export function canonicalBlockLabelOrder(
+  strategy: ChunkStrategy,
+  labels: string[]
+): string[] {
+  return strategy === "update_blocks" ? canonicalChangelogLabelOrder(labels) : labels;
+}
+
 export function changelogLabelHash(blocks: ContentBlock[]): string {
-  return blockHash(blocks.map((b) => b.label).join("|"));
+  const labels = canonicalChangelogLabelOrder(blocks.map((b) => b.label));
+  return blockHash(labels.join("|"));
+}
+
+const EN_MONTH_NAMES = [
+  "January",
+  "February",
+  "March",
+  "April",
+  "May",
+  "June",
+  "July",
+  "August",
+  "September",
+  "October",
+  "November",
+  "December",
+] as const;
+
+const EN_DATE_DESC_RE =
+  /^(January|February|March|April|May|June|July|August|September|October|November|December)\s+(\d{1,2}),\s+(\d{4})$/;
+
+export function parseEnglishChangelogDate(
+  date: string
+): { year: number; month: number; day: number } | null {
+  const m = date.trim().match(EN_DATE_DESC_RE);
+  if (!m) return null;
+  const month = EN_MONTH_NAMES.indexOf(m[1] as (typeof EN_MONTH_NAMES)[number]) + 1;
+  if (month <= 0) return null;
+  return { year: Number(m[3]), month, day: Number(m[2]) };
+}
+
+/** Localized `<Update description="…">` date for ja / ko / zh. */
+export function formatChangelogDateForLang(
+  year: number,
+  month: number,
+  day: number,
+  lang: string
+): string {
+  switch (lang) {
+    case "ja":
+    case "zh":
+      return `${year}年${month}月${day}日`;
+    case "ko":
+      return `${year}년 ${month}월 ${day}일`;
+    default:
+      return `${EN_MONTH_NAMES[month - 1]} ${day}, ${year}`;
+  }
+}
+
+export function localizeEnglishChangelogDate(date: string, lang: string): string {
+  const parsed = parseEnglishChangelogDate(date);
+  if (!parsed || lang === "en") return date;
+  return formatChangelogDateForLang(parsed.year, parsed.month, parsed.day, lang);
+}
+
+export function extractUpdateDescription(updateBlock: string): string | null {
+  return updateBlock.match(/<Update\s+[^>]*description="([^"]+)"/)?.[1] ?? null;
+}
+
+/** Replace description with locale date derived from the English source block. */
+export function syncUpdateBlockDescription(
+  translatedBlock: string,
+  enBlock: ContentBlock,
+  langCode: string
+): string {
+  const enDesc = extractUpdateDescription(enBlock.content);
+  if (!enDesc || langCode === "en") return translatedBlock;
+  const localized = localizeEnglishChangelogDate(enDesc, langCode);
+  if (localized === enDesc) return translatedBlock;
+  return translatedBlock.replace(
+    /(<Update\s+[^>]*description=")([^"]+)(")/,
+    `$1${localized}$3`
+  );
+}
+
+export function applyChangelogBlockLocalizations(
+  slots: BlockSlot[],
+  enBlocks: ContentBlock[],
+  langCode: string
+): BlockSlot[] {
+  if (langCode === "en") return slots;
+  const enByLabel = new Map(enBlocks.map((b) => [b.label, b]));
+  return slots.map((s) => {
+    const enBlock = enByLabel.get(s.label);
+    if (!s.content || !enBlock) return s;
+    return {
+      label: s.label,
+      content: syncUpdateBlockDescription(s.content, enBlock, langCode),
+    };
+  });
+}
+
+export function hasChangelogDateDrift(
+  enContent: string,
+  targetContent: string,
+  langCode: string
+): boolean {
+  if (langCode === "en") return false;
+  const enDoc = parseDocument(enContent, "update_blocks");
+  const targetDoc = parseDocument(targetContent, "update_blocks");
+  const targetByLabel = new Map(targetDoc.blocks.map((b) => [b.label, b]));
+  for (const enBlock of enDoc.blocks) {
+    const enDesc = extractUpdateDescription(enBlock.content);
+    if (!enDesc) continue;
+    const expected = localizeEnglishChangelogDate(enDesc, langCode);
+    const targetBlock = targetByLabel.get(enBlock.label);
+    if (!targetBlock) continue;
+    const actual = extractUpdateDescription(targetBlock.content) ?? "";
+    if (actual !== expected) return true;
+  }
+  return false;
 }
 
 // ---------------------------------------------------------------------------
@@ -352,21 +555,36 @@ export function serializeChunkedDocument(
   slots: BlockSlot[],
   fileHash: string,
   enRel: string,
-  blockHashes: Record<string, string>
+  blockHashes: Record<string, string>,
+  strategy: ChunkStrategy
 ): string {
-  const body = slots
+  const ordered = orderSlotsForStrategy(strategy, slots);
+  const body = ordered
     .map((s) => s.content)
     .filter((c): c is string => c !== null)
     .join("\n\n");
+  const labelOrder = ordered.map((s) => s.label);
   const raw = `${frontmatter}\n${body}\n`;
-  return setChunkedTranslationMeta(raw, fileHash, enRel, blockHashes);
+  return setChunkedTranslationMeta(raw, fileHash, enRel, blockHashes, labelOrder);
+}
+
+/** Rebuild changelog body with `<Update>` blocks in descending semver order. */
+export function serializeUpdateBlocksDocument(
+  frontmatter: string,
+  blocks: ContentBlock[]
+): string {
+  const body = sortUpdateBlocksByVersion(blocks)
+    .map((b) => b.content)
+    .join("\n\n");
+  return `${frontmatter}\n${body}\n`;
 }
 
 export function getSectionSyncStatus(
   enContent: string,
   existingContent: string,
   strategy: ChunkStrategy,
-  force: boolean
+  force: boolean,
+  langCode?: string
 ): SectionSyncStatus {
   const enDoc = parseDocument(enContent, strategy);
   const enHashes = documentBlockHashes(enDoc.blocks);
@@ -389,22 +607,33 @@ export function getSectionSyncStatus(
 
   const existingDoc = parseDocument(existingContent, strategy);
 
+  const existingFmBody = parseFrontmatterAndBody(existingContent).frontmatter
+    .replace(/^---\n/, "")
+    .replace(/\n---$/, "");
+  const storedLabels = parseBlockHashLabelOrderFromFrontmatter(existingFmBody);
+  const enLabels = canonicalBlockLabelOrder(
+    strategy,
+    enDoc.blocks.map((b) => b.label)
+  );
+  const hasOrderDrift = blockHashLabelOrderDrifts(enLabels, storedLabels);
+
   if (strategy === "update_blocks") {
     const existingLabels = new Set(existingDoc.blocks.map((b) => b.label));
     const pendingBlocks = enDoc.blocks
       .filter((b) => !existingLabels.has(b.label))
       .map((b) => b.label);
+    const hasDateDrift = langCode
+      ? hasChangelogDateDrift(enContent, existingContent, langCode)
+      : false;
     return {
-      upToDate: pendingBlocks.length === 0,
+      upToDate: pendingBlocks.length === 0 && !hasOrderDrift && !hasDateDrift,
       pendingBlocks,
       needsFrontmatter: existingDoc.blocks.length === 0,
+      needsReserialize: pendingBlocks.length === 0 && (hasOrderDrift || hasDateDrift),
     };
   }
 
-  const existingFm = parseFrontmatterAndBody(existingContent).frontmatter;
-  const storedHashes = parseBlockHashesFromFrontmatter(
-    existingFm.replace(/^---\n/, "").replace(/\n---$/, "")
-  );
+  const storedHashes = parseBlockHashesFromFrontmatter(existingFmBody);
 
   const pendingBlocks = enDoc.blocks
     .filter((b) => storedHashes[b.label] !== enHashes[b.label])
@@ -413,10 +642,11 @@ export function getSectionSyncStatus(
   const hasStructureDrift = Object.keys(storedHashes).some((k) => !(k in enHashes));
 
   return {
-    upToDate: pendingBlocks.length === 0 && !hasStructureDrift,
+    upToDate: pendingBlocks.length === 0 && !hasStructureDrift && !hasOrderDrift,
     pendingBlocks,
     needsFrontmatter: Object.keys(storedHashes).length === 0,
-    needsReserialize: hasStructureDrift && pendingBlocks.length === 0,
+    needsReserialize:
+      pendingBlocks.length === 0 && (hasStructureDrift || hasOrderDrift),
   };
 }
 
diff --git a/.github/scripts/i18n/translate-i18n.ts b/.github/scripts/i18n/translate-i18n.ts
index c415cceb3..0d9f77fcb 100644
--- a/.github/scripts/i18n/translate-i18n.ts
+++ b/.github/scripts/i18n/translate-i18n.ts
@@ -74,6 +74,7 @@ import {
   type ChunkedFileConfig,
   type ChunkStrategy,
   aggregateDocumentHash,
+  applyChangelogBlockLocalizations,
   changelogLabelHash,
   documentBlockHashes,
   getSectionSyncStatus,
@@ -83,6 +84,7 @@ import {
   resolveChunkStrategy,
   serializeChunkedDocument,
   stripTranslationMetaFromFrontmatter,
+  syncUpdateBlockDescription,
   validateTranslatedBlock,
 } from "./chunked-translate.ts";
 
@@ -545,12 +547,13 @@ async function writeChunkedCheckpoint(
   fileHash: string,
   enRel: string,
   blockHashes: Record<string, string>,
+  strategy: ChunkStrategy,
   label?: string
 ): Promise<void> {
   await mkdir(dirname(targetPath), { recursive: true });
   await writeFile(
     targetPath,
-    serializeChunkedDocument(frontmatter, slots, fileHash, enRel, blockHashes)
+    serializeChunkedDocument(frontmatter, slots, fileHash, enRel, blockHashes, strategy)
   );
   if (label) {
     const done = slots.filter((s) => s.content !== null).length;
@@ -573,7 +576,7 @@ async function translateChunkedFile(
   blocksTranslated: number;
   output: string | null;
 }> {
-  const status = getSectionSyncStatus(enContent, existingContent, strategy, force);
+  const status = getSectionSyncStatus(enContent, existingContent, strategy, force, lang.code);
   if (status.upToDate) {
     return { mismatches: [], status: "up-to-date", blocksTranslated: 0, output: null };
   }
@@ -608,20 +611,22 @@ async function translateChunkedFile(
   if (status.needsReserialize && status.pendingBlocks.length === 0) {
     if (slots.every((s) => s.content?.trim())) {
       const translatedFrontmatter = existingDoc?.frontmatter ?? enDoc.frontmatter;
-      const filledSlots: BlockSlot[] = slots.map((s) => ({
-        label: s.label,
-        content: s.content!,
-      }));
+      const filledSlots = applyChangelogBlockLocalizations(
+        slots.map((s) => ({ label: s.label, content: s.content! })),
+        enDoc.blocks,
+        lang.code
+      );
       const output = serializeChunkedDocument(
         translatedFrontmatter,
         filledSlots,
         fileHash,
         enRel,
-        enBlockHashes
+        enBlockHashes,
+        strategy
       );
       await mkdir(dirname(targetPath), { recursive: true });
       await writeFile(targetPath, output);
-      console.log(`    Re-serialized after EN section removal (no re-translation)`);
+      console.log(`    Re-serialized (order/date fix, no re-translation)`);
       return {
         mismatches: [],
         status: "translated",
@@ -659,6 +664,7 @@ async function translateChunkedFile(
       fileHash,
       enRel,
       enBlockHashes,
+      strategy,
     );
   } else {
     translatedFrontmatter = existingDoc!.frontmatter;
@@ -688,6 +694,9 @@ async function translateChunkedFile(
 
     let translatedBlock = cleanModelOutput(blockResult.content);
     translatedBlock = localizeMdxPaths(translatedBlock, lang, config.languages);
+    if (strategy === "update_blocks") {
+      translatedBlock = syncUpdateBlockDescription(translatedBlock, enBlock, lang.code);
+    }
 
     if (!validateTranslatedBlock(strategy, enBlock, translatedBlock)) {
       console.log(`    [WARN] Block ${blockTag}: invalid output, keeping existing`);
@@ -705,18 +714,21 @@ async function translateChunkedFile(
       fileHash,
       enRel,
       enBlockHashes,
+      strategy,
       blockTag
     );
   }
 
   const output = serializeChunkedDocument(
     translatedFrontmatter,
-    slots,
+    applyChangelogBlockLocalizations(slots, enDoc.blocks, lang.code),
     fileHash,
     enRel,
-    enBlockHashes
+    enBlockHashes,
+    strategy
   );
-  const didWork = blocksTranslated > 0 || frontmatterDirty || status.needsFrontmatter;
+  const didWork =
+    blocksTranslated > 0 || frontmatterDirty || status.needsFrontmatter || status.needsReserialize;
 
   return {
     mismatches: allMismatches,
@@ -974,7 +986,8 @@ async function runTranslatePhase(options: {
           enContent,
           existing,
           chunkStrategy,
-          false
+          false,
+          lang.code
         );
         if (chunkedStatus.upToDate) upToDate.push(job);
         else pending.push(job);
@@ -1004,9 +1017,10 @@ async function runTranslatePhase(options: {
       if (chunkStrategy) {
         const enContent = await readFileOr(makeMapping(lang, relPath, false).enPath);
         const existing = await readFileOr(makeMapping(lang, relPath, false).targetPath);
-        const cs = getSectionSyncStatus(enContent, existing, chunkStrategy, false);
+        const cs = getSectionSyncStatus(enContent, existing, chunkStrategy, false, lang.code);
         const parts: string[] = [];
         if (cs.needsFrontmatter) parts.push("frontmatter");
+        if (cs.needsReserialize) parts.push("reserialize");
         if (cs.pendingBlocks.length > 0) {
           const unit =
             chunkStrategy === "heading_sections" ? "section(s)" : "version(s)";
diff --git a/changelog/index.mdx b/changelog/index.mdx
index bbcd97ce5..d2bb11493 100644
--- a/changelog/index.mdx
+++ b/changelog/index.mdx
@@ -4,6 +4,65 @@ description: "Track ComfyUI's latest features, improvements, and bug fixes. For
 icon: "clock-rotate-left"
 ---
 
+<Update label="v0.25.1" description="June 16, 2026">
+
+**Partner Node Updates**
+* [**Kling V3-Turbo**](https://github.com/Comfy-Org/ComfyUI/pull/14528): Added support for Kling V3-Turbo model
+
+</Update>
+
+<Update label="v0.25.0" description="June 16, 2026">
+
+**New Open-Source Model Support**
+* [**SeedVR2**](https://github.com/Comfy-Org/ComfyUI/pull/14110): Support for SeedVR2 video super-resolution model
+* [**Depth Anything 3**](https://github.com/Comfy-Org/ComfyUI/pull/13853): Monocular depth estimation model from LiheYoung
+* [**SCAIL-2**](https://github.com/Comfy-Org/ComfyUI/pull/14373): Character replacement model for enhanced video editing
+* [**Bernini-R**](https://github.com/Comfy-Org/ComfyUI/pull/14216): Wan video model support
+* **Ideogram 4**: Fixed fp8 dtype issues for improved stability
+* **10-bit Video Support**: Added core node support for 10-bit video input/output
+
+**New Nodes**
+* [**PreviewGaussianSplat + PreviewPointCloud**](https://github.com/Comfy-Org/ComfyUI/pull/14194): New 3D preview nodes for Gaussian splat and point cloud visualization
+* [**Color Primitive**](https://github.com/Comfy-Org/ComfyUI/pull/14260): Color type and utilities for workflows
+* **Improved ResolutionSelector** ([#14309](https://github.com/Comfy-Org/ComfyUI/pull/14309)): Enhanced resolution selection with better defaults
+
+**Partner Node Updates**
+* **Bria Transparent Video Background + Green Background + Replace Background** nodes
+* **Gemini Text Node** ([#14299](https://github.com/Comfy-Org/ComfyUI/pull/14299)): New LLM partner node
+* **Runway Aleph2** ([#14306](https://github.com/Comfy-Org/ComfyUI/pull/14306)): New video generation node
+* **Krea 2 Medium Turbo** model support
+* **Tripo3D Import 3D** node
+* **Flux Erase**: Added seed input
+* **NanoBanana**: Added temperature and top_p controls
+* **Flux KV cache**: Fixed crash with split batches
+
+**Performance & Stability**
+* **Aimdo 0.4.10** + `--reserve-vram` and `--vram-headroom` options
+* Comfy Aimdo 0.4.9 with reliability improvements
+* Media loading: Generalized `--high-ram` option
+* **Assets API**: Cursor-based pagination, image dimension extraction at ingest, asset ID in WebSocket messages
+* Always enable CUDA malloc on CUDA 130+
+* Force cudnn.benchmark to false for deterministic results
+* Fixed odd-height crash and edge bleed in unaligned-width image/video decode
+* Fixed nondeterministic video decode at unaligned widths
+* **Ideogram 4**: Fixed don't reset cast buffers in cleanup_models_gc()
+* **Trainer**: Fixed conditions become trainable in training node
+
+</Update>
+
+<Update label="v0.24.1" description="June 3, 2026">
+
+**Partner Node Updates**
+* Added Bria Green Background node
+* Added Bria Transparent Video Background node with alpha channel support in SaveWEBM
+* Added Krea 2 Medium Turbo model
+* Added seed input to Flux Erase node
+
+**Bug Fixes**
+* Fixed Seedance 2.0 1080p first/last-frame stretch jump
+
+</Update>
+
 <Update label="v0.24.0" description="June 3, 2026">
 
 **New Open-Source Model Support**
@@ -89,6 +148,32 @@ icon: "clock-rotate-left"
 
 </Update>
 
+<Update label="v0.22.3" description="May 20, 2026">
+
+**Partner Node Updates**
+* Added Krea2 Image nodes
+* Improved video reference uploading for SeeDance 2 using memoryview
+
+</Update>
+
+<Update label="v0.22.2" description="May 20, 2026">
+
+**Partner Node Updates**
+* Added Rodin2.5 nodes with quality mesh options (usdz export disabled)
+* Fixed seed parameter always being passed to server in Rodin2.5
+
+</Update>
+
+<Update label="v0.22.1" description="May 20, 2026">
+
+**Partner Node Updates**
+* Added OpenRouter LLM node
+* Added reasoning widget to Anthropic node
+* Added widget for automatic upscaling for ByteDance2Reference node
+* Fixed passing images to Grok LLM
+
+</Update>
+
 <Update label="v0.22.0" description="May 20, 2026">
 
 **Open-Source Model Support**
@@ -199,6 +284,35 @@ icon: "clock-rotate-left"
 
 </Update>
 
+<Update label="v0.20.3" description="April 27, 2026">
+
+**Partner Node Updates**
+* Added Luma UNI-1 models
+* Added GPT 5.5 and 5.5-pro LLM models
+* Added grok-imagine-image-quality model
+* New NanoBanana2 node with DynamicCombo/Autogrow
+
+**Bug Fixes**
+* Fixed ImageBlend/ImageCompositeMasked handling images with different channel counts (alpha channel)
+* Fixed Kling V3 price badge in Motion Control node
+
+</Update>
+
+<Update label="v0.20.2" description="April 27, 2026">
+
+**Partner Node Updates**
+* Added Topaz Astra 2 model (now default upscaler)
+* ByteDance virtual portrait library for regular images
+* Added custom resolution support for GPTImage2 node
+* Removed Moonvalley API nodes
+* Increased default timeout for partner API node tasks
+
+**Model Support**
+* SDPose resize fix
+* OneTainer ERNIE LoRA support
+
+</Update>
+
 <Update label="v0.20.1" description="April 27, 2026">
 
 **New Model Support**
@@ -237,6 +351,7 @@ icon: "clock-rotate-left"
 * Added range type support for improved node parameter handling
 
 </Update>
+
 <Update label="v0.20.0" description="April 27, 2026">
 
 **New Model Support**
@@ -275,6 +390,34 @@ icon: "clock-rotate-left"
 
 </Update>
 
+<Update label="v0.19.5" description="April 17, 2026">
+
+**Partner Node Updates**
+* Added SD2 Real Human support for Stable Diffusion 2
+* Added 4K resolution to Kling nodes
+* Added GPTImage price badge fixes and new resolutions
+
+**Bug Fixes**
+* Fixed 4K resolution restriction for veo-3.0 models
+* Fixed Stable_Zero123 cc_projection weight assignment on Windows (aimdo-enabled)
+
+</Update>
+
+<Update label="v0.19.4" description="April 17, 2026">
+
+**Partner Node Updates**
+* Added 4K resolution support for Veo models + Veo 3 Lite model
+* Added gpt-image-2 as a version option
+* Added automatic video downscaling for ByteDance 2 nodes
+* Added SD2 Real Human support
+
+**Bug Fixes**
+* Fixed Stable_Zero123 cc_projection weight assignment on Windows (aimdo-enabled)
+* Fixed 4K resolution restriction for veo-3.0 models
+* Fixed GPTImage price badges
+
+</Update>
+
 <Update label="v0.19.3" description="April 17, 2026">
 
 **LTX text generation**
@@ -286,27 +429,28 @@ icon: "clock-rotate-left"
 * Corrected StabilityAI price badges for accurate API node pricing information
 
 </Update>
+
 <Update label="v0.19.2" description="April 17, 2026">
 
 **Performance & memory**
-- Fixed OOM regression in quantized models during inference, improving memory usage for _apply() operations
+* Fixed OOM regression in quantized models during inference, improving memory usage for _apply() operations
 
 **Text generation**
-- **LTX:** Added option to disable the default template in text generation nodes for more flexible prompt handling
-- Introduced JsonExtractString node for JSON data manipulation in workflows
+* **LTX:** Added option to disable the default template in text generation nodes for more flexible prompt handling
+* Introduced JsonExtractString node for JSON data manipulation in workflows
 
 **Bug fixes**
-- **Ernie Image:** Fixed incorrect class name — use `ErnieTEModel_` instead of `ErnieTEModel`
+* **Ernie Image:** Fixed incorrect class name — use `ErnieTEModel_` instead of `ErnieTEModel`
 
 **Partner nodes**
-- Added 1080p resolution support for SeeDance 2.0 model, enabling higher quality video generation
+* Added 1080p resolution support for SeeDance 2.0 model, enabling higher quality video generation
 
 </Update>
 
 <Update label="v0.19.1" description="April 16, 2026">
 
 **New Model Support**
-* [Erine Image  Text to Image](https://blog.comfy.org/p/ernie-image-day-0-support)
+* [Ernie Image Text to Image](https://blog.comfy.org/p/ernie-image-day-0-support)
 * LLM: Ministral model
 
 **Partner Node**
@@ -316,6 +460,7 @@ icon: "clock-rotate-left"
 * Preview as Text node update: Support any value to string conversion
 
 </Update>
+
 <Update label="v0.19.0" description="April 13, 2026">
 
 **Model & Generation Support**
@@ -363,6 +508,32 @@ icon: "clock-rotate-left"
 
 </Update>
 
+<Update label="v0.18.5" description="March 25, 2026">
+
+**Stability**
+* Additional testing after fp8 checkpoint fix revert
+
+</Update>
+
+<Update label="v0.18.4" description="March 25, 2026">
+
+**Bug Fixes**
+* Fixed pinned memory accounting for Windows, reducing risk of OOM errors
+* Fixed fp8 scaled checkpoints not loading correctly
+* Reverted fp8 checkpoint fix causing regressions
+
+**Partner Node Updates**
+* Added new Wan2.7 partner nodes
+
+</Update>
+
+<Update label="v0.18.3" description="March 25, 2026">
+
+**Partner Node Updates**
+* Added new Topaz model in API nodes
+
+</Update>
+
 <Update label="v0.18.2" description="March 25, 2026">
 
 **Partner Node(API)**
@@ -370,6 +541,7 @@ icon: "clock-rotate-left"
 - Grok Video Extend 
 - [Blog](https://blog.comfy.org/p/grok-imagine-model-feature-updates)
 </Update>
+
 <Update label="v0.18.1" description="March 23, 2026">
 
 **FP16 Support Fixes**
@@ -381,6 +553,7 @@ icon: "clock-rotate-left"
 - Fixed light and color issues in WAN VAE processing, improving output quality and color accuracy
 
 </Update>
+
 <Update label="v0.18.0" description="March 21, 2026">
 
 **Memory & Performance Optimizations**
@@ -417,16 +590,19 @@ icon: "clock-rotate-left"
 - Fixed various memory leaks and corruption issues in rare edge cases
 
 </Update>
+
 <Update label="v0.17.2" description="March 15, 2026">
 
 This release focuses on stability and bug fixes. For complete technical details, see the [full changelog](https://github.com/Comfy-Org/ComfyUI/compare/v0.17.1...v0.17.2) on GitHub.
 
 </Update>
+
 <Update label="v0.17.1" description="March 13, 2026">
 
 This is a patch release that includes various bug fixes and stability improvements. For full details on changes included in this release, see the complete changelog comparison between v0.17.0 and v0.17.1.
 
 </Update>
+
 <Update label="v0.17.0" description="March 13, 2026">
 
 **Architecture & Performance Improvements**
@@ -459,6 +635,7 @@ This is a patch release that includes various bug fixes and stability improvemen
 - Updated comfy-kitchen to version 0.2.8 and comfy-aimdo to version 0.2.10
 
 </Update>
+
 <Update label="v0.16.4" description="March 7, 2026">
 
 **New Nodes & Features**
@@ -485,6 +662,7 @@ This is a patch release that includes various bug fixes and stability improvemen
 - Added causal_fix parameter to add_keyframe_index and append_keyframe functions, enhancing keyframe handling precision in video workflows
 
 </Update>
+
 <Update label="v0.16.2" description="March 5, 2026">
 
 **Core Updates**
@@ -572,6 +750,7 @@ This is a patch release that includes various bug fixes and stability improvemen
 - Updated workflow templates to version 0.9.4
 
 </Update>
+
 <Update label="v0.15.0" description="February 24, 2026">
 
 **New Nodes & Features**
@@ -614,6 +793,7 @@ This is a patch release that includes various bug fixes and stability improvemen
 **Bug Fixes**
 - Fixed issue with Gemini/Nano banana API nodes that sometimes returned blank images, improving reliability for AI-powered image generation workflows.
 </Update>
+
 <Update label="v0.14.1" description="February 17, 2026">
 
 **Bug Fixes**
@@ -628,6 +808,231 @@ This is a patch release that includes various bug fixes and stability improvemen
 
 
 </Update>
+
+<Update label="v0.14.0" description="February 17, 2026">
+
+**New Open-Source Model Support**
+* [**Qwen 2512 ControlNet / Fun ControlNet**](https://github.com/Comfy-Org/ComfyUI/pull/12359): Support for Qwen-based ControlNet models for enhanced image control
+* **NAG Implementation**: Nesterov accelerated gradient applied to all Flux-based models
+
+**New Nodes & APIs**
+* [**VideoSlice Node**](https://github.com/Comfy-Org/ComfyUI/pull/12107): Video slicing utility for precise temporal control
+* [**Node Replacement API**](https://github.com/Comfy-Org/ComfyUI/pull/12014): Significant architecture change allowing node replacement in workflows
+* **Create List Node** ([#12173](https://github.com/Comfy-Org/ComfyUI/pull/12173)): List creation utility node
+
+**Performance & Technical Improvements**
+* Left-padded text encoder attention mask support
+* Torch RMSNorm for Flux models + Hunyuan video refactor
+* Dynamic VRAM: training fixes, vanilla-fp8 LoRA quality fix
+* ModelPatcherDynamic: force load non-leaf weights
+* Fix LoRA extraction in offload + dynamic VRAM mode
+* Use torch RMSNorm for flux models and refactor hunyuan video
+
+**Partner Node Updates**
+* **Magnific Upscalers** API nodes enabled
+* **Tencent 3D nodes**: ModelTo3DUV, 3DTextureEdit, 3DParts
+* **Bria RMBG** API nodes
+* **Kling API nodes**: V3, O3 models
+
+**Trainer Improvements**
+* Training with proper offloading (KohakuBlueleaf)
+* Built-in LoRA training now works on Anima
+
+</Update>
+
+<Update label="v0.13.0" description="February 10, 2026">
+
+**New Model Support & Improvements**
+* [**EasyCache: LTX2 Support**](https://github.com/Comfy-Org/ComfyUI/pull/12231): EasyCache now works with LTX2 for faster video generation
+* **ACE Step 1.5** ([#12311](https://github.com/Comfy-Org/ComfyUI/pull/12311)): Now works without LLM, base model default workflow fix, text encoding improvements
+* **FP16 for Cosmos-Predict2 and Anima** ([#12249](https://github.com/Comfy-Org/ComfyUI/pull/12249)): Reduced memory usage for supported models
+* **Make tonemap latent work on any dim latents** ([#12363](https://github.com/Comfy-Org/ComfyUI/pull/12363)): More flexible tone mapping
+
+**Performance & Fixes**
+* Dynamic VRAM fixes for ACE 1.5 performance + VRAM leak fix
+* LazyCache fix ([#12344](https://github.com/Comfy-Org/ComfyUI/pull/12344))
+* Disable prompt weights for LTXV2
+* Frontend bumped to 1.38.13
+
+**Partner Node Updates**
+* **Kling API nodes**: New V3, O3 models
+* Moonvalley API nodes fixes
+
+</Update>
+
+<Update label="v0.12.3" description="February 5, 2026">
+
+**ACE Step 1.5 Improvements**
+* [**LLM sampling options**](https://github.com/Comfy-Org/ComfyUI/pull/12295): Added sampling options and reference audio support
+* **Text encoding improvements** ([#12283](https://github.com/Comfy-Org/ComfyUI/pull/12283)): Better text processing for ACE
+* **VAE tiled decode for audio** ([#12299](https://github.com/Comfy-Org/ComfyUI/pull/12299))
+* Sage attention disabled for ACE Step 1.5
+* Fix NaN issue on some hardware/pytorch configs
+
+**3D Improvements**
+* File3DAny output on Load3D node; SaveGLB accepts File3DAny input
+
+</Update>
+
+<Update label="v0.12.2" description="February 4, 2026">
+
+**ACE Step 1.5**
+* [**4B LM model support**](https://github.com/Comfy-Org/ComfyUI/pull/12257): Support for the 4B ACE step 1.5 language model
+* VRAM leak fix in dynamic VRAM mode
+
+</Update>
+
+<Update label="v0.12.1" description="February 3, 2026">
+
+**ACE Step 1.5 Fixes**
+* [**Progress bar**](https://github.com/Comfy-Org/ComfyUI/pull/12242): Added progress bar for ACE Step
+* **Base model LoRAs** support
+* **Tiled VAE fix** for ACE
+* Mac compatibility fixes
+
+**3D & API**
+* Basic 3D model file types in API
+* LLM logits cast as comfy-weight
+
+</Update>
+
+<Update label="v0.12.0" description="February 3, 2026">
+
+**New Model Support**
+* [**ACE Step 1.5**](https://github.com/Comfy-Org/ComfyUI/pull/12237): Basic support for the ACE Step 1.5 audio/speech generation model — a major new model family
+* [**KV Cache for LLM text generation**](https://github.com/Comfy-Org/ComfyUI/pull/12195): Performance boost for running llama models with cached key-value pairs
+* [**Adaptive Model Loading**](https://github.com/Comfy-Org/ComfyUI/pull/11845): Reduced RAM usage, fixed VRAM OOMs, and fixed Windows shared memory spilling
+
+**New Nodes**
+* **Color type + Color to RGB Int node** ([#12145](https://github.com/Comfy-Org/ComfyUI/pull/12145))
+
+**API Nodes**
+* **HitPaw API nodes** ([#12117](https://github.com/Comfy-Org/ComfyUI/pull/12117))
+* **Recraft** style node
+* **Vidu**: Q3 model, Extend and MultiFrame nodes
+
+**Developer**
+* Assets Part 2 - more endpoints
+* send is_input_list on v1/v3 schema to frontend
+
+</Update>
+
+<Update label="v0.11.1" description="January 29, 2026">
+
+**Partner Node Updates**
+* [**Grok Imagine API nodes**](https://github.com/Comfy-Org/ComfyUI/pull/12136): Image generation via Grok
+* Manager bumped to 4.1b1
+
+**Developer**
+* Dev-only nodes support
+* Python 3.14 compat notes
+
+</Update>
+
+<Update label="v0.11.0" description="January 27, 2026">
+
+**New Open-Source Model Support**
+* [**Anima Model**](https://github.com/Comfy-Org/ComfyUI/pull/12012): Support for the Anima model family with fp16 optimization
+* [**Multi/InfiniteTalk**](https://github.com/Comfy-Org/ComfyUI/pull/10179): Large-scale multi-speaker support for LTXV
+* [**Flux2 LyCORIS LoKr**](https://github.com/Comfy-Org/ComfyUI/pull/11997): LoKr (KronA-based) LoRA support for Flux2
+* **LTX2 Tiny VAE** (taeltx_2) support
+* **Z-Image Omni** base model support
+* **ModelScope LoRA** format for Flux2 Klein
+
+**New Nodes & Features**
+* **Magnific API nodes** ([#11986](https://github.com/Comfy-Org/ComfyUI/pull/11986)): Upscaling via Magnific
+* **Tencent Hunyuan3D API nodes** ([#12026](https://github.com/Comfy-Org/ComfyUI/pull/12026))
+* **WaveSpeed API nodes** ([#11945](https://github.com/Comfy-Org/ComfyUI/pull/11945))
+* **Search aliases** field in node schema for better node discoverability
+* **kwargs inputs** for arbitrary inputs from frontend
+* **RoPE Position Adjustment** for small grid IC-LoRA
+
+**Performance**
+* QWEN VAE and WAN speed/VRAM optimizations
+* LTX2 VAE VRAM reduction
+* LTX2 refactored forward for better VRAM efficiency
+
+**Model Enhancements**
+* Flux2 Klein checkpoints saved with SaveCheckpoint now load properly
+* Config for Qwen 3 0.6B model
+* Dynamically detect chroma radiance patch size
+
+</Update>
+
+<Update label="v0.10.0" description="January 21, 2026">
+
+**New Features**
+* [**API Node Auto-Discovery**](https://github.com/Comfy-Org/ComfyUI/pull/11943): All `nodes_*.py` files are now auto-discovered, simplifying node development
+* **Advanced Parameter support** ([#11939](https://github.com/Comfy-Org/ComfyUI/pull/11939)): Advanced widget support for input classes
+* **Image sizes added to CLIP vision outputs** ([#11923](https://github.com/Comfy-Org/ComfyUI/pull/11923))
+
+**Partner Node Updates**
+* **ByteDance Seedance 1.5 Pro** API node
+* **Bria Edit** node ([#11978](https://github.com/Comfy-Org/ComfyUI/pull/11978))
+* Autogrow validation improvements
+
+**Bug Fixes**
+* Mono audio to fake stereo for LTXV VAE encoding ([#11965](https://github.com/Comfy-Org/ComfyUI/pull/11965))
+* Flux2 Klein memory estimation adjustments
+* AMD GPU tensor type fix
+
+</Update>
+
+<Update label="v0.9.2" description="January 15, 2026">
+
+**New Model Support**
+* [**Flux2 Klein**](https://github.com/Comfy-Org/ComfyUI/pull/11890): Major new model family support — Flux2 Klein with expanded model capabilities
+* [**Z-Image ControlNet Lite**](https://github.com/Comfy-Org/ComfyUI/pull/11849): Lightweight version of Alibaba-Pai Z-Image ControlNet
+
+**New Nodes**
+* **Meshy 3D API nodes** ([#11843](https://github.com/Comfy-Org/ComfyUI/pull/11843))
+* **ResizeImageMaskNode**: Added crop-to-multiple mode
+* **Blueprints directory** for built-in templates
+
+**Performance & Technical**
+* NVFP4 LoRA optimization (multiple rounds)
+* Throttled ProgressBar WebSocket updates
+* Lanczos grayscale upscaling fix
+* VAELoader metadata loading
+* Repository URL updated to Comfy-Org/ComfyUI
+
+</Update>
+
+<Update label="v0.9.1" description="January 13, 2026">
+
+**Bug Fix**
+* Bumped LTXAV memory estimation
+
+</Update>
+
+<Update label="v0.9.0" description="January 13, 2026">
+
+**New Features**
+* [**Basic Asset Support**](https://github.com/Comfy-Org/ComfyUI/pull/11315): Foundation for model asset management — endpoints for model CRUD operations
+* **NVFP4 Model LoRA** ([#11837](https://github.com/Comfy-Org/ComfyUI/pull/11837)): LoRA support now works on NVFP4 quantized models
+* [**SigLIP 2 NAFlex**](https://github.com/Comfy-Org/ComfyUI/pull/11831): Support for SigLIP 2 NAFlex CLIP vision model
+* [**ImageCompare Node**](https://github.com/Comfy-Org/ComfyUI/pull/11343): Compare two images visually
+* [**JoinAudioChannels**](https://github.com/Comfy-Org/ComfyUI/pull/11728): Audio channel joining utility
+
+**Model Support**
+* ModelScope LoRA for Z-Image models ([#11805](https://github.com/Comfy-Org/ComfyUI/pull/11805))
+* FP4/FP8 fixes for Chroma and text encoders
+* VAEEncodeForInpaint now supports WAN VAE tuple downscale_ratio
+* LTX2 VRAM reduction via efficient timestep embed handling
+
+**API Nodes**
+* **Vidu2** nodes ([#11760](https://github.com/Comfy-Org/ComfyUI/pull/11760))
+* **Topaz Enhance** fixes
+* **Gemini** safety block handling
+
+**Technical**
+* AMD gfx1200: PyTorch attention enabled by default
+* CI container version bump automation
+* Refactored model loading to lower memory usage
+* Frontend patched to 1.36.14
+
+</Update>
+
 <Update label="v0.8.1" description="January 8, 2026">
 
 **Model Support**
@@ -646,6 +1051,7 @@ This is a patch release that includes various bug fixes and stability improvemen
 - Updated workflow templates to v0.7.69
 
 </Update>
+
 <Update label="v0.8.0" description="January 7, 2026">
 
 **New Models Support**
@@ -672,6 +1078,7 @@ This is a patch release that includes various bug fixes and stability improvemen
 - Workflow templates updated to v0.7.67
 - Code refactoring and cleanup (CLIP preprocessing, comfy-kitchen integration)
 </Update>
+
 <Update label="v0.7.0" description="December 31, 2025">
 **System Requirements**
 - PyTorch 2.4+ minimum version requirement
@@ -701,6 +1108,7 @@ This is a patch release that includes various bug fixes and stability improvemen
 - Converted image processing nodes to V3 schema
 - Optimized Lumina/Z image model (removed unused components)
 </Update>
+
 <Update label="v0.6.0" description="December 24, 2025">
 
 **New Model Support**
@@ -721,6 +1129,7 @@ This is a patch release that includes various bug fixes and stability improvemen
 - Fixed mask concatenation issues in ZImageFunControlNet in --gpu-only mode
 
 </Update>
+
 <Update label="v0.5.1" description="December 18, 2025">
 
 - Added GPT-Image-1.5 API nodes
@@ -747,6 +1156,7 @@ This is a patch release that includes various bug fixes and stability improvemen
 - Frontend updated to v1.34.9
 
 </Update>
+
 <Update label="v0.4.0" description="December 10, 2025">
 
 **Model Support**
@@ -795,6 +1205,7 @@ This is a patch release that includes various bug fixes and stability improvemen
 - If you are using the Desktop and **if you also turn off the auto-update**, the Desktop might not be able to auto-update for the latest version this time.
 - Please download the latest Desktop version from [here](https://www.comfy.org/download) to get the latest features.
 </Update>
+
 <Update label="v0.3.76" description="December 2, 2025">
 
 **Frontend UI/UX**
@@ -843,6 +1254,7 @@ This is a patch release that includes various bug fixes and stability improvemen
 - Optimized Flux 2 text encoder VRAM usage for better performance
 
 </Update>
+
 <Update label="v0.3.72" description="November 25, 2025">
 
 **New Model Support**
@@ -868,6 +1280,7 @@ This is a patch release that includes various bug fixes and stability improvemen
 - Bumped Transformers version
 
 </Update>
+
 <Update label="v0.3.71" description="November 21, 2025">
 
 **Model Compatibility & Enhancements**
@@ -932,6 +1345,7 @@ This is a patch release that includes various bug fixes and stability improvemen
 - Unified RoPE function implementation across models
 
 </Update>
+
 <Update label="v0.3.68" description="November 5, 2025">
 
 **Performance & Memory Optimizations**
@@ -958,6 +1372,7 @@ This is a patch release that includes various bug fixes and stability improvemen
 - Fixed Windows pinned memory allocation issues
 
 </Update>
+
 <Update label="v0.3.67" description="October 28, 2025">
 
 **API Nodes**
@@ -979,6 +1394,7 @@ This is a patch release that includes various bug fixes and stability improvemen
 - Template updates to version 0.2.4
 
 </Update>
+
 <Update label="v0.3.66" description="October 21, 2025">
 
 **Frontend Updates**
@@ -1004,6 +1420,7 @@ This is a patch release that includes various bug fixes and stability improvemen
 - Improved merge_nested_dicts functionality with proper input ordering
 - Added deprecation warnings for unused files
 </Update>
+
 <Update label="v0.3.65" description="October 14, 2025">
 
 **Node Schema Migration (V3)**
@@ -1028,12 +1445,14 @@ This is a patch release that includes various bug fixes and stability improvemen
 - Fixed WAN2.2 cache VRAM leak
 
 </Update>
+
 <Update label="v0.3.64" description="October 8, 2025">
 
 **API Nodes**
 - Added Sora2 API node for OpenAI's video generation API
 
 </Update>
+
 <Update label="v0.3.63" description="October 6, 2025">
 
 **Model Compatibility Enhancements**
@@ -1059,6 +1478,7 @@ This is a patch release that includes various bug fixes and stability improvemen
 - **Subgraph Publish**: Allows publishing subgraphs to the node library
 - **Node Selection Toolbox Redesign**: Redesigned the node selection toolbox
 </Update>
+
 <Update label="v0.3.62" description="September 30, 2025">
 
 **API Node**
@@ -1108,12 +1528,14 @@ This is a patch release that includes various bug fixes and stability improvemen
 - **Frontend Version Update**: Updated to version 1.26.13
 
 </Update>
+
 <Update label="v0.3.59" description="September 10, 2025">
 
 **ByteDance Seedream 4.0 Integration**
 - **New Seedream Node**: Added ByteDanceSeedream (4.0) node.
 
 </Update>
+
 <Update label="v0.3.58" description="September 6, 2025">
 
 **New Model Support**
@@ -1124,6 +1546,7 @@ This is a patch release that includes various bug fixes and stability improvemen
 - Stable Audio 2.5 API
 - Seedance Video API
 </Update>
+
 <Update label="v0.3.57" description="September 4, 2025">
 
 **ByteDance USO Model Support**
@@ -1143,6 +1566,7 @@ This is a patch release that includes various bug fixes and stability improvemen
 - **ByteDance Image Nodes**: Added support for ByteDance image generation services
 - **Ideogram Character Reference**: Ideogram v3 API now supports character reference
 </Update>
+
 <Update label="v0.3.56" description="August 30, 2025">
 
 **Performance Enhancement**
@@ -1168,6 +1592,7 @@ This release focuses on Wan2.2 S2V related video workflow capabilities and model
 - **Fun Inpaint Model Support**: Integrated Wan2.2 5B fun inpaint model.
 
 </Update>
+
 <Update label="v0.3.54" description="August 28, 2025">
 
 **Node Model Patch Improvements**
@@ -1181,6 +1606,7 @@ This focused update improves the core node model patching system that underpins
 - **Enhanced Stability**: Core model patching improvements contribute to more reliable node execution and model handling across different workflow configurations
 
 </Update>
+
 <Update label="v0.3.53" description="August 28, 2025">
 
 **Audio Workflow Integration & Enhanced Performance Optimizations**
@@ -1216,6 +1642,7 @@ This release adds ComfyUI audio processing capabilities and includes performance
 
 
 </Update>
+
 <Update label="v0.3.52" description="August 23, 2025">
 
 **Enhanced Model Support & Qwen Image ControlNet Integration**
@@ -1321,6 +1748,7 @@ This release expands ComfyUI's model ecosystem with enhanced Qwen support, async
 - **Video Control**: WAN 2.2 fun control features provide advanced creative control for video generation workflows
 
 </Update>
+
 <Update label="v0.3.49" description="August 5, 2025">
 
 **UI Experience & Model Support**
@@ -1356,6 +1784,7 @@ This release brings user experience improvements and model support that enhance
 - **Performance**: Optimized VRAM usage allows for more ambitious projects on systems with limited resources
 
 </Update>
+
 <Update label="v0.3.48" description="August 2, 2025">
 
 **API Enhancement & Performance Optimizations**
@@ -1385,6 +1814,7 @@ This release introduces backend improvements and performance optimizations that
 - **Execution Flexibility**: New partial execution capabilities allow for more efficient debugging and development
 
 </Update>
+
 <Update label="v0.3.47" description="July 30, 2025">
 
 **Memory Optimization & Large Model Performance**
diff --git a/ja/changelog/index.mdx b/ja/changelog/index.mdx
index ea6cfd96f..2fa108727 100644
--- a/ja/changelog/index.mdx
+++ b/ja/changelog/index.mdx
@@ -2,10 +2,163 @@
 title: "変更履歴"
 description: "ComfyUI の最新機能、改善点、およびバグ修正を追跡します。詳細なリリースノートについては、[GitHub リリースページ](https://github.com/Comfy-Org/ComfyUI/releases) をご覧ください。"
 icon: "clock-rotate-left"
-translationSourceHash: 20ee5837
+translationSourceHash: 19292673
 translationFrom: changelog/index.mdx
+translationBlockHashes:
+  "v0.25.1": b23dd37e
+  "v0.25.0": 4b295314
+  "v0.24.1": 3091a9fd
+  "v0.24.0": 6df070aa
+  "v0.23.0": af9616ad
+  "v0.22.3": 5a6904a9
+  "v0.22.2": 81d85e22
+  "v0.22.1": fe7e08b9
+  "v0.22.0": b5e583a9
+  "v0.21.1": 4476798b
+  "v0.21.0": cce71d0e
+  "v0.20.3": ea4a7c00
+  "v0.20.2": 73294410
+  "v0.20.1": 095a7300
+  "v0.20.0": faf4da7d
+  "v0.19.5": 25b8c75f
+  "v0.19.4": 0095bf05
+  "v0.19.3": 1f2576e1
+  "v0.19.2": f27cb1e0
+  "v0.19.1": a7a6886e
+  "v0.19.0": 3798375e
+  "v0.18.5": b0686463
+  "v0.18.4": def92141
+  "v0.18.3": 5db865c4
+  "v0.18.2": e6c4cf63
+  "v0.18.1": 54074011
+  "v0.18.0": 3b11162d
+  "v0.17.2": 7a9ad1ca
+  "v0.17.1": 72f9ea2e
+  "v0.17.0": bb797f70
+  "v0.16.4": 74471b8a
+  "v0.16.3": ca5f3d51
+  "v0.16.2": 5ac5e4c2
+  "v0.16.1": 48041a88
+  "v0.16.0": 60df9f99
+  "v0.15.1": f44fc2d6
+  "v0.15.0": 42aec6c0
+  "v0.14.2": 592d10c3
+  "v0.14.1": 85b7622e
+  "v0.14.0": a7129c05
+  "v0.13.0": 50c2d102
+  "v0.12.3": 030409d4
+  "v0.12.2": 15956101
+  "v0.12.1": 559dfce0
+  "v0.12.0": eb7a13f4
+  "v0.11.1": b4698f4c
+  "v0.11.0": e7fb2c06
+  "v0.10.0": 23ce7740
+  "v0.9.2": 908463d4
+  "v0.9.1": 940bbc85
+  "v0.9.0": 080a8205
+  "v0.8.1": 483a4e74
+  "v0.8.0": 91302dc4
+  "v0.7.0": 5883c71c
+  "v0.6.0": 94f8fcaa
+  "v0.5.1": 5b496342
+  "v0.5.0": f986ef06
+  "v0.4.0": eab6e372
+  "v0.3.76": d4bb7314
+  "v0.3.75": f4cbbd4b
+  "v0.3.73": ac329747
+  "v0.3.72": d9758f36
+  "v0.3.71": 3a711845
+  "v0.3.70": 047f4543
+  "v0.3.69": 822a8f53
+  "v0.3.68": 323f6dd8
+  "v0.3.67": 4f8bfdf1
+  "v0.3.66": cf95523c
+  "v0.3.65": e0295dff
+  "v0.3.64": b93b2252
+  "v0.3.63": 91bd0e39
+  "v0.3.62": ff99b31c
+  "v0.3.61": 3ca48550
+  "v0.3.60": a7ab9646
+  "v0.3.59": 66253ba1
+  "v0.3.58": 10704eaa
+  "v0.3.57": 67a7bceb
+  "v0.3.56": 342318cd
+  "v0.3.55": bf738910
+  "v0.3.54": 0224d936
+  "v0.3.53": b10ddac0
+  "v0.3.52": 2a8e0096
+  "v0.3.51": 2e360b2f
+  "v0.3.50": 198f4ac6
+  "v0.3.49": 037d5d8c
+  "v0.3.48": a92aa58b
+  "v0.3.47": 295aceff
+  "v0.3.46": 46ef61dd
+  "v0.3.45": a8c769b6
+  "v0.3.44": 3899ae9e
+  "v0.3.43": 2526e22f
+  "v0.3.41": f15a2902
+  "v0.3.40": 24608eaf
 ---
 
+<Update label="v0.25.1" description="2026年6月16日">
+
+**パートナーノードの更新**
+* [**Kling V3-Turbo**](https://github.com/Comfy-Org/ComfyUI/pull/14528): Kling V3-Turboモデルのサポートを追加
+
+</Update>
+
+<Update label="v0.25.0" description="2026年6月16日">
+
+**新しいオープンソースモデルのサポート**
+* [**SeedVR2**](https://github.com/Comfy-Org/ComfyUI/pull/14110): SeedVR2 ビデオ超解像度モデルのサポート
+* [**Depth Anything 3**](https://github.com/Comfy-Org/ComfyUI/pull/13853): LiheYoung による単眼深度推定モデル
+* [**SCAIL-2**](https://github.com/Comfy-Org/ComfyUI/pull/14373): 強化されたビデオ編集のためのキャラクター置換モデル
+* [**Bernini-R**](https://github.com/Comfy-Org/ComfyUI/pull/14216): Wan ビデオモデルのサポート
+* **Ideogram 4**: 安定性向上のためのfp8 dtype問題の修正
+* **10ビットビデオサポート**: 10ビットビデオの入出力用コアノードサポートを追加
+
+**新ノード**
+* [**PreviewGaussianSplat + PreviewPointCloud**](https://github.com/Comfy-Org/ComfyUI/pull/14194): ガウススプラットとポイントクラウド可視化のための新しい3Dプレビューノード
+* [**Color Primitive**](https://github.com/Comfy-Org/ComfyUI/pull/14260): ワークフロー向けの色タイプとユーティリティ
+* **改良された ResolutionSelector** ([#14309](https://github.com/Comfy-Org/ComfyUI/pull/14309)): より良いデフォルト値による解像度選択の強化
+
+**パートナーノードの更新**
+* **Bria Transparent Video Background + Green Background + Replace Background** ノード
+* **Gemini Text Node** ([#14299](https://github.com/Comfy-Org/ComfyUI/pull/14299)): 新しいLLMパートナーノード
+* **Runway Aleph2** ([#14306](https://github.com/Comfy-Org/ComfyUI/pull/14306)): 新しいビデオ生成ノード
+* **Krea 2 Medium Turbo** モデルサポート
+* **Tripo3D Import 3D** ノード
+* **Flux Erase**: seed入力を追加
+* **NanoBanana**: temperatureとtop_pの制御を追加
+* **Flux KVキャッシュ**: 分割バッチでのクラッシュを修正
+
+**パフォーマンスと安定性**
+* **Aimdo 0.4.10** + `--reserve-vram` と `--vram-headroom` オプション
+* Comfy Aimdo 0.4.9（信頼性の向上）
+* メディア読み込み: `--high-ram` オプションの一般化
+* **Assets API**: カーソルベースのページネーション、取り込み時の画像寸法抽出、WebSocketメッセージ内のアセットID
+* CUDA 130+で常にCUDA mallocを有効化
+* 決定論的な結果のためにcudnn.benchmarkを強制的にfalseに
+* 奇数高さのクラッシュと、幅が整列されていない画像/ビデオのデコード時のエッジブリードを修正
+* 幅が整列されていない場合の非決定論的なビデオデコードを修正
+* **Ideogram 4**: cleanup_models_gc() でキャストバッファがリセットされない問題を修正
+* **Trainer**: トレーニングノードで条件がトレーニング可能になる問題を修正
+
+</Update>
+
+<Update label="v0.24.1" description="2026年6月3日">
+
+**パートナーノードの更新**
+* Bria Green Backgroundノードを追加
+* SaveWEBMでアルファチャンネルをサポートするBria Transparent Video Backgroundノードを追加
+* Krea 2 Medium Turboモデルを追加
+* Flux Eraseノードにシード入力を追加
+
+**バグ修正**
+* Seedance 2.0 1080pの最初/最後のフレームのストレッチジャンプを修正
+
+</Update>
+
 <Update label="v0.24.0" description="2026年6月3日">
 
 **新しいオープンソースモデル対応**
@@ -92,6 +245,32 @@ translationFrom: changelog/index.mdx
 
 </Update>
 
+<Update label="v0.22.3" description="2026年5月20日">
+
+**パートナーノードの更新**
+* Krea2 Image ノードを追加
+* memoryview を使用して、SeeDance 2 のビデオ参照アップロードを改善
+
+</Update>
+
+<Update label="v0.22.2" description="2026年5月20日">
+
+**パートナーノード更新**
+* 品質メッシュオプション付きの Rodin2.5 ノードを追加 (usdz エクスポートは無効)
+* Rodin2.5 で seed パラメータが常にサーバーに渡される問題を修正
+
+</Update>
+
+<Update label="v0.22.1" description="2026年5月20日">
+
+**パートナーノードの更新**
+* OpenRouter LLM ノードを追加
+* Anthropic ノードに推論ウィジェットを追加
+* ByteDance2Reference ノードに自動アップスケーリング用ウィジェットを追加
+* Grok LLM への画像の受け渡しを修正
+
+</Update>
+
 <Update label="v0.22.0" description="2026年5月20日">
 
 **オープンソースモデル対応**
@@ -202,6 +381,35 @@ translationFrom: changelog/index.mdx
 
 </Update>
 
+<Update label="v0.20.3" description="2026年4月27日">
+
+**パートナーノードの更新**
+* Luma UNI-1モデルを追加
+* GPT 5.5および5.5-pro LLMモデルを追加
+* grok-imagine-image-qualityモデルを追加
+* DynamicCombo/Autogrowを備えた新しいNanoBanana2ノード
+
+**バグ修正**
+* ImageBlend/ImageCompositeMaskedが異なるチャンネル数（アルファチャンネル）の画像を処理する際の問題を修正
+* Motion ControlノードのKling V3の価格バッジを修正
+
+</Update>
+
+<Update label="v0.20.2" description="2026年4月27日">
+
+**パートナーノードの更新**
+* Topaz Astra 2モデルを追加 (デフォルトのアップスケーラーになりました)
+* 一般画像向けByteDanceバーチャルポートレートライブラリ
+* GPTImage2ノードにカスタム解像度サポートを追加
+* Moonvalley APIノードを削除
+* パートナーAPIノードタスクのデフォルトタイムアウトを延長
+
+**モデルサポート**
+* SDPoseリサイズ修正
+* OneTainer ERNIE LoRAサポート
+
+</Update>
+
 <Update label="v0.20.1" description="2026年4月27日">
 
 **新しいモデル対応**
@@ -279,6 +487,34 @@ translationFrom: changelog/index.mdx
 
 </Update>
 
+<Update label="v0.19.5" description="2026年4月17日">
+
+**パートナーノードの更新**
+* Stable Diffusion 2向けSD2 Real Humanサポートを追加
+* Klingノードに4K解像度を追加
+* GPTImageの価格バッジ修正と新たな解像度を追加
+
+**バグ修正**
+* veo-3.0モデルの4K解像度制限を修正
+* Windowsでaimdo有効時のStable_Zero123のcc_projection重み割り当てを修正
+
+</Update>
+
+<Update label="v0.19.4" description="2026年4月17日">
+
+**パートナーノードの更新**
+* Veoモデルの4K解像度サポートを追加、Veo 3 Liteモデルも追加
+* gpt-image-2をバージョンオプションとして追加
+* ByteDance 2ノードの自動ビデオダウンスケーリングを追加
+* SD2 Real Humanサポートを追加
+
+**バグ修正**
+* WindowsでStable_Zero123のcc_projection重み割り当てを修正（aimdo有効時）
+* veo-3.0モデルの4K解像度制限を修正
+* GPTImageの価格バッジを修正
+
+</Update>
+
 <Update label="v0.19.3" description="2026年4月17日">
 
 **LTX テキスト生成**
@@ -369,6 +605,32 @@ translationFrom: changelog/index.mdx
 
 </Update>
 
+<Update label="v0.18.5" description="2026年3月25日">
+
+**安定性**
+* fp8チェックポイント修正取り消し後の追加テスト
+
+</Update>
+
+<Update label="v0.18.4" description="2026年3月25日">
+
+**バグ修正**
+* Windowsにおけるピン留めメモリの管理を修正し、OOMエラーのリスクを軽減
+* fp8スケールチェックポイントが正しく読み込まれない問題を修正
+* リグレッションを引き起こしていたfp8チェックポイントの修正を差し戻しました
+
+**パートナーノードの更新**
+* 新しいWan2.7パートナーノードを追加
+
+</Update>
+
+<Update label="v0.18.3" description="2026年3月25日">
+
+**パートナーノードの更新**
+* APIノードに新しい Topaz モデルが追加されました
+
+</Update>
+
 <Update label="v0.18.2" description="2026年3月25日">
 
 **パートナー・ノード (API)**
@@ -644,6 +906,230 @@ translationFrom: changelog/index.mdx
 
 </Update>
 
+<Update label="v0.14.0" description="2026年2月17日">
+
+**新しいオープンソースモデルのサポート**
+* [**Qwen 2512 ControlNet / Fun ControlNet**](https://github.com/Comfy-Org/ComfyUI/pull/12359): QwenベースのControlNetモデルのサポートによる画像制御の強化
+* **NAG実装**: すべてのFluxベースモデルにNesterovの加速勾配法を適用
+
+**新しいノードとAPI**
+* [**VideoSliceノード**](https://github.com/Comfy-Org/ComfyUI/pull/12107): 正確な時間制御のためのビデオスライスユーティリティ
+* [**ノード置換API**](https://github.com/Comfy-Org/ComfyUI/pull/12014): ワークフローでノード置換を可能にする大幅なアーキテクチャ変更
+* **リスト作成ノード** ([#12173](https://github.com/Comfy-Org/ComfyUI/pull/12173)): リスト作成ユーティリティノード
+
+**パフォーマンスと技術的改善**
+* 左パディングされたテキストエンコーダのアテンションマスクサポート
+* Fluxモデル向けTorch RMSNorm + Hunyuanビデオリファクタリング
+* 動的VRAM: トレーニング修正、vanilla-fp8 LoRA品質修正
+* ModelPatcherDynamic: 非リーフウェイトの強制ロード
+* オフロード＋動的VRAMモードでのLoRA抽出修正
+* Fluxモデルにtorch RMSNormを使用し、Hunyuanビデオをリファクタリング
+
+**パートナーノードの更新**
+* **Magnific Upscalers** APIノードが有効化
+* **Tencent 3Dノード**: ModelTo3DUV, 3DTextureEdit, 3DParts
+* **Bria RMBG** APIノード
+* **Kling APIノード**: V3, O3モデル
+
+**トレーナーの改善**
+* 適切なオフロードによるトレーニング (KohakuBlueleaf)
+* 内蔵LoRAトレーニングがAnimaで動作するようになりました
+
+</Update>
+
+<Update label="v0.13.0" description="2026年2月10日">
+
+**新モデルサポートと改善**
+* [**EasyCache: LTX2サポート**](https://github.com/Comfy-Org/ComfyUI/pull/12231): EasyCacheがLTX2で動作し、より高速なビデオ生成が可能になりました。
+* **ACE Step 1.5** ([#12311](https://github.com/Comfy-Org/ComfyUI/pull/12311)): LLMなしで動作、ベースモデルのデフォルトワークフローの修正、テキストエンコーディングの改善。
+* **Cosmos-Predict2とAnima向けFP16** ([#12249](https://github.com/Comfy-Org/ComfyUI/pull/12249)): サポート対象モデルのメモリ使用量を削減。
+* **任意の次元の潜在でトーンマップ潜在を動作可能に** ([#12363](https://github.com/Comfy-Org/ComfyUI/pull/12363)): より柔軟なトーンマッピング。
+
+**パフォーマンスと修正**
+* ACE 1.5のパフォーマンス向上のための動的VRAM修正とVRAMリーク修正
+* LazyCacheの修正 ([#12344](https://github.com/Comfy-Org/ComfyUI/pull/12344))
+* LTXV2のプロンプトの重みを無効化
+* フロントエンドを1.38.13に更新
+
+**パートナーノードの更新**
+* **Kling APIノード**: 新モデルV3、O3
+* Moonvalley APIノードの修正
+
+</Update>
+
+<Update label="v0.12.3" description="2026年2月5日">
+
+**ACE Step 1.5の改善**
+* [**LLM サンプリングオプション**](https://github.com/Comfy-Org/ComfyUI/pull/12295): サンプリングオプションとReferenceTimbreAudio サポートが追加されました
+* **テキストエンコーディングの改善** ([#12283](https://github.com/Comfy-Org/ComfyUI/pull/12283)): ACEでのテキスト処理が改善されました
+* **VAEのタイルデコード（オーディオ）** ([#12299](https://github.com/Comfy-Org/ComfyUI/pull/12299))
+* ACE Step 1.5ではSageアテンションが無効化されました
+* 一部のハードウェア／PyTorch設定でのNaN問題を修正
+
+**3Dの改善**
+* Load3DノードにFile3DAny出力が追加され、SaveGLBはFile3DAny入力を受け付けるようになりました
+
+</Update>
+
+<Update label="v0.12.2" description="2026年2月4日">
+
+**ACE Step 1.5**
+* [**4B LM モデルサポート**](https://github.com/Comfy-Org/ComfyUI/pull/12257): 4B ACE step 1.5 言語モデルのサポート
+* ダイナミックVRAMモードにおけるVRAMリーク修正
+
+</Update>
+
+<Update label="v0.12.1" description="2026年2月3日">
+
+**ACE Step 1.5 修正**
+* [**進捗バー**](https://github.com/Comfy-Org/ComfyUI/pull/12242): ACE Step の進捗バーを追加
+* **ベースモデル LoRA** サポート
+* ACE 向け **タイル VAE 修正**
+* Mac 互換性の修正
+
+**3D & API**
+* API での基本的な 3D モデル ファイル形式
+* LLM ロジットを comfy-weight としてキャスト
+
+</Update>
+
+<Update label="v0.12.0" description="2026年2月3日">
+
+**新モデルサポート**
+* [**ACE Step 1.5**](https://github.com/Comfy-Org/ComfyUI/pull/12237): ACE Step 1.5オーディオ/音声生成モデルの基本サポート — 主要な新しいモデルファミリー
+* [**KVキャッシュによるLLMテキスト生成**](https://github.com/Comfy-Org/ComfyUI/pull/12195): キャッシュされたキー・バリューペアを用いたllamaモデルの実行パフォーマンス向上
+* [**アダプティブモデル読み込み**](https://github.com/Comfy-Org/ComfyUI/pull/11845): RAM使用量の削減、VRAM OOMの修正、Windows共有メモリのスピリング修正
+
+**新ノード**
+* **色タイプ + 色→RGB整数ノード** ([#12145](https://github.com/Comfy-Org/ComfyUI/pull/12145))
+
+**APIノード**
+* **HitPaw APIノード** ([#12117](https://github.com/Comfy-Org/ComfyUI/pull/12117))
+* **Recraft** スタイルノード
+* **Vidu**: Q3モデル、Extendノード、MultiFrameノード
+
+**開発者向け**
+* アセット Part 2 - エンドポイント追加
+* v1/v3スキーマでis_input_listをフロントエンドに送信
+
+</Update>
+
+<Update label="v0.11.1" description="2026年1月29日">
+
+**パートナーノード更新**
+* [**Grok Imagine APIノード**](https://github.com/Comfy-Org/ComfyUI/pull/12136): Grok 経由の画像生成
+* マネージャーが 4.1b1 に更新
+
+**開発者**
+* 開発専用ノードのサポート
+* Python 3.14 互換性に関する注意事項
+
+</Update>
+
+<Update label="v0.11.0" description="2026年1月27日">
+
+**新しいオープンソースモデルのサポート**
+* [**Anima Model**](https://github.com/Comfy-Org/ComfyUI/pull/12012): Animaモデルファミリーのサポート（fp16最適化）
+* [**Multi/InfiniteTalk**](https://github.com/Comfy-Org/ComfyUI/pull/10179): LTXV向けの大規模マルチスピーカーサポート
+* [**Flux2 LyCORIS LoKr**](https://github.com/Comfy-Org/ComfyUI/pull/11997): Flux2向けLoKr（KronAベース）LoRAサポート
+* **LTX2 Tiny VAE** (taeltx_2) のサポート
+* **Z-Image Omni** ベースモデルサポート
+* Flux2 Klein用**ModelScope LoRA**フォーマット
+
+**新しいノードと機能**
+* **Magnific APIノード** ([#11986](https://github.com/Comfy-Org/ComfyUI/pull/11986)): Magnificによるアップスケーリング
+* **Tencent Hunyuan3D APIノード** ([#12026](https://github.com/Comfy-Org/ComfyUI/pull/12026))
+* **WaveSpeed APIノード** ([#11945](https://github.com/Comfy-Org/ComfyUI/pull/11945))
+* ノード発見性を向上させる、ノードスキーマ内の**エイリアスを検索**フィールド
+* フロントエンドからの任意の入力に対応する**kwargs入力**
+* 小グリッドIC-LoRA向け**RoPE位置調整**
+
+**パフォーマンス**
+* QWEN VAEおよびWANの速度・VRAM最適化
+* LTX2 VAEのVRAM削減
+* LTX2の forward 処理をリファクタリングし、VRAM効率を向上
+
+**モデル強化**
+* SaveCheckpointで保存されたFlux2 Kleinチェックポイントが正しく読み込まれるようになりました
+* Qwen 3 0.6Bモデル用設定
+* クロマ輝度パッチサイズを動的に検出
+
+</Update>
+
+<Update label="v0.10.0" description="2026年1月21日">
+
+**新機能**
+* [**APIノード自動検出**](https://github.com/Comfy-Org/ComfyUI/pull/11943): すべての `nodes_*.py` ファイルが自動的に検出されるようになり、ノード開発が簡素化されました
+* **高度なパラメーターサポート** ([#11939](https://github.com/Comfy-Org/ComfyUI/pull/11939)): 入力クラスに対する高度なウィジェットサポート
+* **CLIPビジョン出力に画像サイズが追加されました** ([#11923](https://github.com/Comfy-Org/ComfyUI/pull/11923))
+
+**パートナーノード更新**
+* **ByteDance Seedance 1.5 Pro** APIノード
+* **Bria Edit** ノード ([#11978](https://github.com/Comfy-Org/ComfyUI/pull/11978))
+* Autogrow検証の改善
+
+**バグ修正**
+* LTXV VAEエンコーディングでモノラルオーディオを疑似ステレオに変換 ([#11965](https://github.com/Comfy-Org/ComfyUI/pull/11965))
+* Flux2 Kleinメモリ推定の調整
+* AMD GPUテンソルタイプの修正
+
+</Update>
+
+<Update label="v0.9.2" description="2026年1月15日">
+
+**新モデルサポート**
+* [**Flux2 Klein**](https://github.com/Comfy-Org/ComfyUI/pull/11890): 主要な新しいモデルファミリーのサポート — 拡張されたモデル性能を持つ Flux2 Klein
+* [**Z-Image ControlNet Lite**](https://github.com/Comfy-Org/ComfyUI/pull/11849): Alibaba-Pai Z-Image ControlNet の軽量版
+
+**新ノード**
+* **Meshy 3D APIノード** ([#11843](https://github.com/Comfy-Org/ComfyUI/pull/11843))
+* **ResizeImageMaskNode**: クロップして倍数にするモードを追加
+* **ブループリントディレクトリ**: 組み込みテンプレート用
+
+**パフォーマンスと技術的な改善**
+* NVFP4 LoRA 最適化 (複数回実施)
+* スロットル制御された ProgressBar WebSocket 更新
+* Lanczos グレースケールアップスケーリング修正
+* VAELoader メタデータ読み込み
+* リポジトリの URL を Comfy-Org/ComfyUI に更新
+
+</Update>
+
+<Update label="v0.9.1" description="2026年1月13日">
+
+**バグ修正**
+- LTXAVのメモリ推定値を引き上げました
+
+</Update>
+
+<Update label="v0.9.0" description="2026年1月13日">
+
+**新機能**
+* [**Basic Asset Support**](https://github.com/Comfy-Org/ComfyUI/pull/11315): モデルアセット管理の基盤 — モデルCRUD操作用のエンドポイント
+* **NVFP4 Model LoRA** ([#11837](https://github.com/Comfy-Org/ComfyUI/pull/11837)): LoRAサポートがNVFP4量子化モデルで動作するようになりました
+* [**SigLIP 2 NAFlex**](https://github.com/Comfy-Org/ComfyUI/pull/11831): SigLIP 2 NAFlex CLIPビジョンモデルのサポート
+* [**ImageCompareノード**](https://github.com/Comfy-Org/ComfyUI/pull/11343): 2つの画像を視覚的に比較
+* [**JoinAudioChannels**](https://github.com/Comfy-Org/ComfyUI/pull/11728): オーディオチャンネル結合ユーティリティ
+
+**モデルサポート**
+* Z-Imageモデル向けModelScope LoRA ([#11805](https://github.com/Comfy-Org/ComfyUI/pull/11805))
+* Chromaおよびテキストエンコーダー向けFP4/FP8修正
+* VAEEncodeForInpaintがWAN VAEのタプルdownscale_ratioをサポート
+* 効率的なtimestep embed処理によるLTX2のVRAM削減
+
+**APIノード**
+* **Vidu2** ノード ([#11760](https://github.com/Comfy-Org/ComfyUI/pull/11760))
+* **Topaz Enhance** 修正
+* **Gemini** セーフティブロック対応
+
+**技術**
+* AMD gfx1200: PyTorchのアテンションがデフォルトで有効に
+* CIコンテナバージョン更新の自動化
+* モデル読み込みをリファクタリングし、メモリ使用量を低減
+* フロントエンドが1.36.14にパッチ適用
+
+</Update>
+
 <Update label="v0.8.1" description="2026年1月8日">
 
 **モデル対応**
@@ -1182,13 +1668,13 @@ translationFrom: changelog/index.mdx
 - **Ideogram Character Reference**: Ideogram v3 API がキャラクター参照をサポートするようになりました
 </Update>
 
-<Update label="v0.3.56" description="August 30, 2025">
+<Update label="v0.3.56" description="2025年8月30日">
 
 **パフォーマンス向上**
 - **WindowsでのRAM使用量の削減**
 </Update>
 
-<Update label="v0.3.55" description="August 29, 2025">
+<Update label="v0.3.55" description="2025年8月29日">
 
 **Wan2.2 S2V ワークフロー強化 & モデルサポート拡充**
 
@@ -1222,7 +1708,7 @@ translationFrom: changelog/index.mdx
 
 </Update>
 
-<Update label="v0.3.53" description="August 28, 2025">
+<Update label="v0.3.53" description="2025年8月28日">
 
 **音声ワークフロー統合とパフォーマンス最適化の強化**
 
@@ -1328,7 +1814,7 @@ translationFrom: changelog/index.mdx
 
 </Update>
 
-<Update label="v0.3.50" description="August 13, 2025">
+<Update label="v0.3.50" description="2025年8月13日">
 
 **モデル統合とパフォーマンス強化**
 
@@ -1429,7 +1915,7 @@ translationFrom: changelog/index.mdx
 
 </Update>
 
-<Update label="v0.3.47" description="July 30, 2025">
+<Update label="v0.3.47" description="2025年7月30日">
 
 **メモリ最適化と大規模モデルのパフォーマンス向上**
 
@@ -1533,7 +2019,7 @@ translationFrom: changelog/index.mdx
 
 </Update>
 
-<Update label="v0.3.44" description="July 8, 2025">
+<Update label="v0.3.44" description="2025年7月8日">
 
 **サンプリングとモデル制御の強化**
 
@@ -1587,7 +2073,7 @@ translationFrom: changelog/index.mdx
 
 </Update>
 
-<Update label="v0.3.41" description="June 17, 2025">
+<Update label="v0.3.41" description="2025年6月17日">
 
 **モデルサポートの追加**
 - **Cosmos Predict2 サポート**: テキストから画像 (2B および 14B モデル) と画像から動画の生成ワークフローの実装
diff --git a/ko/changelog/index.mdx b/ko/changelog/index.mdx
index 79d70c303..88fb593cb 100644
--- a/ko/changelog/index.mdx
+++ b/ko/changelog/index.mdx
@@ -2,10 +2,163 @@
 title: "변경 로그"
 description: "ComfyUI의 최신 기능, 개선 사항 및 버그 수정을 추적하세요. 자세한 릴리스 노트는 [Github 릴리스](https://github.com/Comfy-Org/ComfyUI/releases) 페이지를 참조하세요."
 icon: "clock-rotate-left"
-translationSourceHash: 20ee5837
+translationSourceHash: 19292673
 translationFrom: changelog/index.mdx
+translationBlockHashes:
+  "v0.25.1": b23dd37e
+  "v0.25.0": 4b295314
+  "v0.24.1": 3091a9fd
+  "v0.24.0": 6df070aa
+  "v0.23.0": af9616ad
+  "v0.22.3": 5a6904a9
+  "v0.22.2": 81d85e22
+  "v0.22.1": fe7e08b9
+  "v0.22.0": b5e583a9
+  "v0.21.1": 4476798b
+  "v0.21.0": cce71d0e
+  "v0.20.3": ea4a7c00
+  "v0.20.2": 73294410
+  "v0.20.1": 095a7300
+  "v0.20.0": faf4da7d
+  "v0.19.5": 25b8c75f
+  "v0.19.4": 0095bf05
+  "v0.19.3": 1f2576e1
+  "v0.19.2": f27cb1e0
+  "v0.19.1": a7a6886e
+  "v0.19.0": 3798375e
+  "v0.18.5": b0686463
+  "v0.18.4": def92141
+  "v0.18.3": 5db865c4
+  "v0.18.2": e6c4cf63
+  "v0.18.1": 54074011
+  "v0.18.0": 3b11162d
+  "v0.17.2": 7a9ad1ca
+  "v0.17.1": 72f9ea2e
+  "v0.17.0": bb797f70
+  "v0.16.4": 74471b8a
+  "v0.16.3": ca5f3d51
+  "v0.16.2": 5ac5e4c2
+  "v0.16.1": 48041a88
+  "v0.16.0": 60df9f99
+  "v0.15.1": f44fc2d6
+  "v0.15.0": 42aec6c0
+  "v0.14.2": 592d10c3
+  "v0.14.1": 85b7622e
+  "v0.14.0": a7129c05
+  "v0.13.0": 50c2d102
+  "v0.12.3": 030409d4
+  "v0.12.2": 15956101
+  "v0.12.1": 559dfce0
+  "v0.12.0": eb7a13f4
+  "v0.11.1": b4698f4c
+  "v0.11.0": e7fb2c06
+  "v0.10.0": 23ce7740
+  "v0.9.2": 908463d4
+  "v0.9.1": 940bbc85
+  "v0.9.0": 080a8205
+  "v0.8.1": 483a4e74
+  "v0.8.0": 91302dc4
+  "v0.7.0": 5883c71c
+  "v0.6.0": 94f8fcaa
+  "v0.5.1": 5b496342
+  "v0.5.0": f986ef06
+  "v0.4.0": eab6e372
+  "v0.3.76": d4bb7314
+  "v0.3.75": f4cbbd4b
+  "v0.3.73": ac329747
+  "v0.3.72": d9758f36
+  "v0.3.71": 3a711845
+  "v0.3.70": 047f4543
+  "v0.3.69": 822a8f53
+  "v0.3.68": 323f6dd8
+  "v0.3.67": 4f8bfdf1
+  "v0.3.66": cf95523c
+  "v0.3.65": e0295dff
+  "v0.3.64": b93b2252
+  "v0.3.63": 91bd0e39
+  "v0.3.62": ff99b31c
+  "v0.3.61": 3ca48550
+  "v0.3.60": a7ab9646
+  "v0.3.59": 66253ba1
+  "v0.3.58": 10704eaa
+  "v0.3.57": 67a7bceb
+  "v0.3.56": 342318cd
+  "v0.3.55": bf738910
+  "v0.3.54": 0224d936
+  "v0.3.53": b10ddac0
+  "v0.3.52": 2a8e0096
+  "v0.3.51": 2e360b2f
+  "v0.3.50": 198f4ac6
+  "v0.3.49": 037d5d8c
+  "v0.3.48": a92aa58b
+  "v0.3.47": 295aceff
+  "v0.3.46": 46ef61dd
+  "v0.3.45": a8c769b6
+  "v0.3.44": 3899ae9e
+  "v0.3.43": 2526e22f
+  "v0.3.41": f15a2902
+  "v0.3.40": 24608eaf
 ---
 
+<Update label="v0.25.1" description="2026년 6월 16일">
+
+**파트너 노드 업데이트**
+* [**Kling V3-Turbo**](https://github.com/Comfy-Org/ComfyUI/pull/14528): Kling V3-Turbo 모델 지원 추가
+
+</Update>
+
+<Update label="v0.25.0" description="2026년 6월 16일">
+
+**새로운 오픈소스 모델 지원**
+* [**SeedVR2**](https://github.com/Comfy-Org/ComfyUI/pull/14110): SeedVR2 비디오 초해상도 모델 지원
+* [**Depth Anything 3**](https://github.com/Comfy-Org/ComfyUI/pull/13853): LiheYoung의 단안 깊이 추정 모델
+* [**SCAIL-2**](https://github.com/Comfy-Org/ComfyUI/pull/14373): 향상된 비디오 편집을 위한 캐릭터 교체 모델
+* [**Bernini-R**](https://github.com/Comfy-Org/ComfyUI/pull/14216): Wan 비디오 모델 지원
+* **Ideogram 4**: 안정성 향상을 위한 fp8 dtype 문제 수정
+* **10비트 비디오 지원**: 10비트 비디오 입출력을 위한 코어 노드 지원 추가
+
+**새로운 노드**
+* [**PreviewGaussianSplat + PreviewPointCloud**](https://github.com/Comfy-Org/ComfyUI/pull/14194): 가우시안 스플랫 및 포인트 클라우드 시각화를 위한 새로운 3D 미리보기 노드
+* [**Color Primitive**](https://github.com/Comfy-Org/ComfyUI/pull/14260): 워크플로용 색상 유형 및 유틸리티
+* **개선된 ResolutionSelector** ([#14309](https://github.com/Comfy-Org/ComfyUI/pull/14309)): 더 나은 기본값으로 해상도 선택 향상
+
+**파트너 노드 업데이트**
+* **Bria Transparent Video Background + Green Background + Replace Background** 노드
+* **Gemini Text Node** ([#14299](https://github.com/Comfy-Org/ComfyUI/pull/14299)): 새로운 LLM 파트너 노드
+* **Runway Aleph2** ([#14306](https://github.com/Comfy-Org/ComfyUI/pull/14306)): 새로운 비디오 생성 노드
+* **Krea 2 Medium Turbo** 모델 지원
+* **Tripo3D Import 3D** 노드
+* **Flux Erase**: 시드 입력 추가
+* **NanoBanana**: temperature 및 top_p 제어 추가
+* **Flux KV 캐시**: 분할 배치에서의 충돌 수정
+
+**성능 및 안정성**
+* **Aimdo 0.4.10** + `--reserve-vram` 및 `--vram-headroom` 옵션
+* Comfy Aimdo 0.4.9, 안정성 개선 사항 포함
+* 미디어 로딩: 일반화된 `--high-ram` 옵션
+* **Assets API**: 커서 기반 페이지네이션, 수집 시 이미지 크기 추출, WebSocket 메시지의 애셋 ID
+* CUDA 130+에서 항상 CUDA malloc 활성화
+* 결정적 결과를 위해 `cudnn.benchmark`를 `false`로 강제 설정
+* 정렬되지 않은 너비의 이미지/비디오 디코드에서 홀수 높이 충돌 및 테두리 번짐 수정
+* 정렬되지 않은 너비에서 비결정적 비디오 디코드 수정
+* **Ideogram 4**: `cleanup_models_gc()`에서 cast 버퍼를 재설정하지 않도록 수정
+* **Trainer**: 학습 노드에서 조건이 학습 가능해지는 문제 수정
+
+</Update>
+
+<Update label="v0.24.1" description="2026년 6월 3일">
+
+**파트너 노드 업데이트**
+* Bria 초록색 배경 노드 추가
+* SaveWEBM에 알파 채널 지원이 포함된 Bria 투명 비디오 배경 노드 추가
+* Krea 2 Medium Turbo 모델 추가
+* Flux Erase 노드에 시드 입력 추가
+
+**버그 수정**
+* Seedance 2.0 1080p 첫 번째/마지막 프레임 늘어나는 현상 수정
+
+</Update>
+
 <Update label="v0.24.0" description="2026년 6월 3일">
 
 **새로운 오픈소스 모델 지원**
@@ -91,6 +244,32 @@ translationFrom: changelog/index.mdx
 
 </Update>
 
+<Update label="v0.22.3" description="2026년 5월 20일">
+
+**파트너 노드 업데이트**
+* Krea2 이미지 노드 추가
+* memoryview를 사용한 SeeDance 2의 비디오 레퍼런스 업로드 개선
+
+</Update>
+
+<Update label="v0.22.2" description="2026년 5월 20일">
+
+**파트너 노드 업데이트**
+* 품질 메시 옵션이 포함된 Rodin2.5 노드 추가 (usdz 내보내기 비활성화)
+* Rodin2.5에서 시드 파라미터가 항상 서버로 전달되던 문제 수정
+
+</Update>
+
+<Update label="v0.22.1" description="2026년 5월 20일">
+
+**파트너 노드 업데이트**
+* OpenRouter LLM 노드 추가됨
+* Anthropic 노드에 추론 위젯 추가됨
+* ByteDance2Reference 노드에 자동 업스케일링 위젯 추가됨
+* Grok LLM에 이미지 전달 수정됨
+
+</Update>
+
 <Update label="v0.22.0" description="2026년 5월 20일">
 
 **오픈소스 모델 지원**
@@ -200,6 +379,35 @@ translationFrom: changelog/index.mdx
 
 </Update>
 
+<Update label="v0.20.3" description="2026년 4월 27일">
+
+**파트너 노드 업데이트**
+* Luma UNI-1 모델 추가
+* GPT 5.5 및 5.5-pro LLM 모델 추가
+* grok-imagine-image-quality 모델 추가
+* DynamicCombo/Autogrow를 지원하는 새로운 NanoBanana2 노드
+
+**버그 수정**
+* ImageBlend/ImageCompositeMasked에서 채널 수가 다른 이미지(알파 채널) 처리 문제 수정
+* Motion Control 노드 내 Kling V3 가격 배지 수정
+
+</Update>
+
+<Update label="v0.20.2" description="2026년 4월 27일">
+
+**파트너 노드 업데이트**
+* Topaz Astra 2 모델 추가 (이제 기본 업스케일러)
+* 일반 이미지용 ByteDance 가상 인물 라이브러리
+* GPTImage2 노드에 사용자 지정 해상도 지원 추가
+* Moonvalley API 노드 제거
+* 파트너 API 노드 작업의 기본 타임아웃 증가
+
+**모델 지원**
+* SDPose 크기 조정 수정
+* OneTainer ERNIE LoRA 지원
+
+</Update>
+
 <Update label="v0.20.1" description="2026년 4월 27일">
 
 **새로운 모델 지원**
@@ -277,6 +485,34 @@ translationFrom: changelog/index.mdx
 
 </Update>
 
+<Update label="v0.19.5" description="2026년 4월 17일">
+
+**파트너 노드 업데이트**
+* Stable Diffusion 2용 SD2 Real Human 지원 추가
+* Kling 노드에 4K 해상도 추가
+* GPTImage 가격 배지 수정 및 새 해상도 추가
+
+**버그 수정**
+* veo-3.0 모델의 4K 해상도 제한 수정
+* Windows(aimdo 활성화)에서 Stable_Zero123 cc_projection 가중치 할당 수정
+
+</Update>
+
+<Update label="v0.19.4" description="2026년 4월 17일">
+
+**파트너 노드 업데이트**
+* Veo 모델의 4K 해상도 지원 및 Veo 3 Lite 모델 추가
+* gpt-image-2를 버전 옵션으로 추가
+* ByteDance 2 노드에 자동 비디오 다운스케일링 추가
+* SD2 Real Human 지원 추가
+
+**버그 수정**
+* Windows에서 Stable_Zero123 cc_projection 가중치 할당 수정 (aimdo 활성화 시)
+* veo-3.0 모델의 4K 해상도 제한 수정
+* GPTImage 가격 배지 수정
+
+</Update>
+
 <Update label="v0.19.3" description="2026년 4월 17일">
 
 **LTX 텍스트 생성**
@@ -367,6 +603,32 @@ translationFrom: changelog/index.mdx
 
 </Update>
 
+<Update label="v0.18.5" description="2026년 3월 25일">
+
+**안정성**
+* fp8 체크포인트 수정 되돌리기 이후 추가 테스트
+
+</Update>
+
+<Update label="v0.18.4" description="2026년 3월 25일">
+
+**버그 수정**
+* Windows에서 고정 메모리 처리 방식을 수정하여 OOM 오류 위험 감소
+* fp8 스케일 체크포인트가 올바르게 로드되지 않는 문제 수정
+* 회귀를 유발하는 fp8 체크포인트 수정 되돌림
+
+**파트너 노드 업데이트**
+* 새로운 Wan2.7 파트너 노드 추가
+
+</Update>
+
+<Update label="v0.18.3" description="2026년 3월 25일">
+
+**파트너 노드 업데이트**
+* API 노드에 새로운 Topaz 모델 추가
+
+</Update>
+
 <Update label="v0.18.2" description="2026년 3월 25일">
 
 **파트너 노드(API)**
@@ -640,6 +902,230 @@ translationFrom: changelog/index.mdx
 - 워크플로우 템플릿을 v0.8.43으로 업데이트함
 
 
+</Update>
+
+<Update label="v0.14.0" description="2026년 2월 17일">
+
+**새로운 오픈소스 모델 지원**
+* [**Qwen 2512 ControlNet / Fun ControlNet**](https://github.com/Comfy-Org/ComfyUI/pull/12359): 향상된 이미지 제어를 위한 Qwen 기반 ControlNet 모델 지원
+* **NAG 구현**: 모든 Flux 기반 모델에 적용되는 네스테로프 가속 그래디언트
+
+**새로운 노드 및 API**
+* [**VideoSlice Node**](https://github.com/Comfy-Org/ComfyUI/pull/12107): 정밀한 시간 제어를 위한 비디오 슬라이싱 유틸리티
+* [**Node Replacement API**](https://github.com/Comfy-Org/ComfyUI/pull/12014): 워크플로에서 노드 교체를 허용하는 중대한 아키텍처 변경
+* **목록 만들기 노드** ([#12173](https://github.com/Comfy-Org/ComfyUI/pull/12173)): 목록 생성 유틸리티 노드
+
+**성능 및 기술 개선 사항**
+* 왼쪽 패딩 텍스트 인코더 어텐션 마스크 지원
+* Flux 모델을 위한 Torch RMSNorm + Hunyuan 비디오 리팩터링
+* 동적 VRAM: 학습 수정 사항, vanilla-fp8 LoRA 품질 수정
+* ModelPatcherDynamic: 비단말 가중치 강제 로드
+* 오프로드 및 동적 VRAM 모드에서 LoRA 추출 수정
+* Flux 모델에 torch RMSNorm 사용 및 Hunyuan 비디오 리팩터링
+
+**파트너 노드 업데이트**
+* **Magnific Upscalers** API 노드 활성화
+* **Tencent 3D 노드**: ModelTo3DUV, 3DTextureEdit, 3DParts
+* **Bria RMBG** API 노드
+* **Kling API 노드**: V3, O3 모델
+
+**트레이너 개선 사항**
+* 적절한 오프로딩을 통한 학습 (KohakuBlueleaf)
+* 내장 LoRA 학습이 이제 Anima에서 작동
+
+</Update>
+
+<Update label="v0.13.0" description="2026년 2월 10일">
+
+**새로운 모델 지원 및 개선 사항**
+* [**EasyCache: LTX2 지원**](https://github.com/Comfy-Org/ComfyUI/pull/12231): 더 빠른 비디오 생성을 위해 EasyCache가 LTX2에서 작동합니다.
+* **ACE Step 1.5** ([#12311](https://github.com/Comfy-Org/ComfyUI/pull/12311)): 이제 LLM 없이 작동하며, 베이스 모델 기본 워크플로 수정, 텍스트 인코딩 개선
+* **Cosmos-Predict2 및 Anima를 위한 FP16** ([#12249](https://github.com/Comfy-Org/ComfyUI/pull/12249)): 지원 모델의 메모리 사용 감소
+* **모든 차원의 잠재 데이터에서 톤맵 잠재 작동하도록 함** ([#12363](https://github.com/Comfy-Org/ComfyUI/pull/12363)): 보다 유연한 톤 매핑
+
+**성능 및 수정 사항**
+* ACE 1.5 성능을 위한 동적 VRAM 수정 + VRAM 누수 수정
+* LazyCache 수정 ([#12344](https://github.com/Comfy-Org/ComfyUI/pull/12344))
+* LTXV2에 대한 프롬프트 가중치 비활성화
+* 프런트엔드 1.38.13으로 업데이트됨
+
+**파트너 노드 업데이트**
+* **Kling API 노드**: 새로운 V3, O3 모델
+* Moonvalley API 노드 수정
+
+</Update>
+
+<Update label="v0.12.3" description="2026년 2월 5일">
+
+**ACE Step 1.5 개선 사항**
+* [**LLM 샘플링 옵션**](https://github.com/Comfy-Org/ComfyUI/pull/12295): 샘플링 옵션 및 ReferenceTimbreAudio 지원 추가
+* **텍스트 인코딩 개선** ([#12283](https://github.com/Comfy-Org/ComfyUI/pull/12283)): ACE를 위한 더 나은 텍스트 처리
+* **오디오용 VAE 타일드 디코딩** ([#12299](https://github.com/Comfy-Org/ComfyUI/pull/12299))
+* ACE Step 1.5에서 Sage 어텐션 비활성화
+* 일부 하드웨어/PyTorch 구성에서 NaN 문제 수정
+
+**3D 개선 사항**
+* Load3D 노드에서 File3DAny 출력; SaveGLB가 File3DAny 입력을 허용합니다.
+
+</Update>
+
+<Update label="v0.12.2" description="2026년 2월 4일">
+
+**ACE Step 1.5**
+* [**4B LM 모델 지원**](https://github.com/Comfy-Org/ComfyUI/pull/12257): 4B ACE step 1.5 언어 모델 지원
+* 동적 VRAM 모드에서 VRAM 누수 수정
+
+</Update>
+
+<Update label="v0.12.1" description="2026년 2월 3일">
+
+**ACE Step 1.5 수정 사항**
+* [**진행 표시줄**](https://github.com/Comfy-Org/ComfyUI/pull/12242): ACE Step용 진행 표시줄 추가
+* **베이스 모델 LoRAs** 지원
+* **ACE용 타일드 VAE 수정**
+* Mac 호환성 수정
+
+**3D & API**
+* API에서의 기본 3D 모델 파일 형식
+* LLM 로짓을 comfy-weight으로 캐스트
+
+</Update>
+
+<Update label="v0.12.0" description="2026년 2월 3일">
+
+**새로운 모델 지원**
+* [**ACE Step 1.5**](https://github.com/Comfy-Org/ComfyUI/pull/12237): ACE Step 1.5 오디오/음성 생성 모델에 대한 기본 지원 — 중요한 새로운 모델 계열
+* [**LLM 텍스트 생성을 위한 KV 캐시**](https://github.com/Comfy-Org/ComfyUI/pull/12195): 캐시된 키-값 쌍을 사용하여 llama 모델을 실행할 때 성능 향상
+* [**적응형 모델 로딩**](https://github.com/Comfy-Org/ComfyUI/pull/11845): RAM 사용량 감소, VRAM OOM 문제 해결, Windows 공유 메모리 스필링 수정
+
+**새로운 노드**
+* **Color type + Color to RGB Int 노드** ([#12145](https://github.com/Comfy-Org/ComfyUI/pull/12145))
+
+**API 노드**
+* **HitPaw API 노드** ([#12117](https://github.com/Comfy-Org/ComfyUI/pull/12117))
+* **Recraft** 스타일 노드
+* **Vidu**: Q3 모델, Extend 및 MultiFrame 노드
+
+**개발자**
+* 에셋 파트 2 - 더 많은 엔드포인트
+* v1/v3 스키마에서 is_input_list를 프론트엔드로 전송
+
+</Update>
+
+<Update label="v0.11.1" description="2026년 1월 29일">
+
+**파트너 노드 업데이트**
+* [**Grok Imagine API 노드**](https://github.com/Comfy-Org/ComfyUI/pull/12136): Grok을 통한 이미지 생성
+* 매니저가 4.1b1로 업데이트됨
+
+**개발자**
+* 개발 전용 노드 지원
+* Python 3.14 호환성 노트
+
+</Update>
+
+<Update label="v0.11.0" description="2026년 1월 27일">
+
+**새로운 오픈소스 모델 지원**
+* [**Anima 모델**](https://github.com/Comfy-Org/ComfyUI/pull/12012): fp16 최적화를 포함한 Anima 모델 제품군 지원
+* [**Multi/InfiniteTalk**](https://github.com/Comfy-Org/ComfyUI/pull/10179): LTXV를 위한 대규모 다중 화자 지원
+* [**Flux2 LyCORIS LoKr**](https://github.com/Comfy-Org/ComfyUI/pull/11997): Flux2용 LoKr(KronA 기반) LoRA 지원
+* **LTX2 Tiny VAE**(taeltx_2) 지원
+* **Z-Image Omni** 베이스 모델 지원
+* **Flux2 Klein**용 ModelScope LoRA 형식
+
+**새로운 노드 및 기능**
+* **Magnific API 노드** ([#11986](https://github.com/Comfy-Org/ComfyUI/pull/11986)): Magnific을 통한 업스케일링
+* **Tencent Hunyuan3D API 노드** ([#12026](https://github.com/Comfy-Org/ComfyUI/pull/12026))
+* **WaveSpeed API 노드** ([#11945](https://github.com/Comfy-Org/ComfyUI/pull/11945))
+* 노드 검색 가능성 향상을 위한 노드 스키마의 **별칭 검색** 필드
+* 프론트엔드에서 임의 입력을 위한 **kwargs 입력**
+* 작은 그리드 IC-LoRA를 위한 **RoPE 위치 조정**
+
+**성능**
+* QWEN VAE 및 WAN 속도/VRAM 최적화
+* LTX2 VAE VRAM 감소
+* VRAM 효율 향상을 위한 LTX2 순전파(forward) 리팩토링
+
+**모델 개선 사항**
+* SaveCheckpoint로 저장된 Flux2 Klein 체크포인트가 이제 올바르게 로드됩니다.
+* Qwen 3 0.6B 모델용 설정
+* 크로마 래디언스 패치 크기 동적 감지
+
+</Update>
+
+<Update label="v0.10.0" description="2026년 1월 21일">
+
+**새로운 기능**
+* [**API 노드 자동 검색**](https://github.com/Comfy-Org/ComfyUI/pull/11943): 모든 `nodes_*.py` 파일이 이제 자동으로 검색되어 노드 개발이 간소화되었습니다.
+* **고급 매개변수 지원** ([#11939](https://github.com/Comfy-Org/ComfyUI/pull/11939)): 입력 클래스를 위한 고급 위젯 지원
+* **CLIP 비전 출력에 이미지 크기 추가** ([#11923](https://github.com/Comfy-Org/ComfyUI/pull/11923))
+
+**파트너 노드 업데이트**
+* **ByteDance Seedance 1.5 Pro** API 노드
+* **Bria Edit** 노드 ([#11978](https://github.com/Comfy-Org/ComfyUI/pull/11978))
+* Autogrow 검증 개선
+
+**버그 수정**
+* LTXV VAE 인코딩을 위한 모노 오디오를 가짜 스테레오로 변환 ([#11965](https://github.com/Comfy-Org/ComfyUI/pull/11965))
+* Flux2 Klein 메모리 추정 조정
+* AMD GPU tensor 유형 수정
+
+</Update>
+
+<Update label="v0.9.2" description="2026년 1월 15일">
+
+**새로운 모델 지원**
+* [**Flux2 Klein**](https://github.com/Comfy-Org/ComfyUI/pull/11890): 주요 신규 모델 계열 지원 — Flux2 Klein (확장된 모델 기능 제공)
+* [**Z-Image ControlNet Lite**](https://github.com/Comfy-Org/ComfyUI/pull/11849): Alibaba-Pai Z-Image ControlNet의 경량화 버전
+
+**새로운 노드**
+* **Meshy 3D API 노드** ([#11843](https://github.com/Comfy-Org/ComfyUI/pull/11843))
+* **ResizeImageMaskNode**: crop-to-multiple 모드 추가
+* **블루프린트 디렉토리** (내장 템플릿용)
+
+**성능 및 기술**
+* NVFP4 LoRA 최적화 (여러 차례)
+* ProgressBar WebSocket 업데이트 제한
+* Lanczos 그레이스케일 업스케일링 수정
+* VAELoader 메타데이터 로딩
+* 저장소 URL이 Comfy-Org/ComfyUI로 업데이트됨
+
+</Update>
+
+<Update label="v0.9.1" description="2026년 1월 13일">
+
+**버그 수정**
+- LTXAV 메모리 추정치를 상향 조정했습니다.
+
+</Update>
+
+<Update label="v0.9.0" description="2026년 1월 13일">
+
+**새로운 기능**
+* [**기본 에셋 지원**](https://github.com/Comfy-Org/ComfyUI/pull/11315): 모델 에셋 관리를 위한 기반 — 모델 CRUD 연산 엔드포인트
+* **NVFP4 모델 LoRA** ([#11837](https://github.com/Comfy-Org/ComfyUI/pull/11837)): LoRA 지원이 이제 NVFP4 양자화 모델에서 작동합니다
+* [**SigLIP 2 NAFlex**](https://github.com/Comfy-Org/ComfyUI/pull/11831): SigLIP 2 NAFlex CLIP 비전 모델 지원
+* [**ImageCompare 노드**](https://github.com/Comfy-Org/ComfyUI/pull/11343): 두 이미지를 시각적으로 비교합니다
+* [**JoinAudioChannels**](https://github.com/Comfy-Org/ComfyUI/pull/11728): 오디오 채널 결합 유틸리티
+
+**모델 지원**
+* Z-Image 모델을 위한 ModelScope LoRA ([#11805](https://github.com/Comfy-Org/ComfyUI/pull/11805))
+* Chroma 및 텍스트 인코더에 대한 FP4/FP8 수정
+* VAEEncodeForInpaint가 WAN VAE 튜플 downscale_ratio를 지원합니다
+* 효율적인 timestep embed 처리를 통한 LTX2 VRAM 감소
+
+**API 노드**
+* **Vidu2** 노드 ([#11760](https://github.com/Comfy-Org/ComfyUI/pull/11760))
+* **Topaz Enhance** 수정
+* **Gemini** 안전 차단 처리
+
+**기술**
+* AMD gfx1200: PyTorch 어텐션 기본 활성화
+* CI 컨테이너 버전 자동 증가
+* 메모리 사용량 감소를 위한 모델 로딩 리팩터링
+* 프론트엔드가 1.36.14로 패치됨
+
 </Update>
 
 <Update label="v0.8.1" description="2026년 1월 8일">
diff --git a/zh/changelog/index.mdx b/zh/changelog/index.mdx
index 55ce5773c..837bee7d6 100644
--- a/zh/changelog/index.mdx
+++ b/zh/changelog/index.mdx
@@ -2,10 +2,163 @@
 title: "更新日志"
 description: "跟踪 ComfyUI 的最新功能、改进和错误修复。详细的发布说明请查看 [Github releases](https://github.com/Comfy-Org/ComfyUI/releases) 页面。"
 icon: "clock-rotate-left"
-translationSourceHash: 20ee5837
+translationSourceHash: 19292673
 translationFrom: changelog/index.mdx
+translationBlockHashes:
+  "v0.25.1": b23dd37e
+  "v0.25.0": 4b295314
+  "v0.24.1": 3091a9fd
+  "v0.24.0": 6df070aa
+  "v0.23.0": af9616ad
+  "v0.22.3": 5a6904a9
+  "v0.22.2": 81d85e22
+  "v0.22.1": fe7e08b9
+  "v0.22.0": b5e583a9
+  "v0.21.1": 4476798b
+  "v0.21.0": cce71d0e
+  "v0.20.3": ea4a7c00
+  "v0.20.2": 73294410
+  "v0.20.1": 095a7300
+  "v0.20.0": faf4da7d
+  "v0.19.5": 25b8c75f
+  "v0.19.4": 0095bf05
+  "v0.19.3": 1f2576e1
+  "v0.19.2": f27cb1e0
+  "v0.19.1": a7a6886e
+  "v0.19.0": 3798375e
+  "v0.18.5": b0686463
+  "v0.18.4": def92141
+  "v0.18.3": 5db865c4
+  "v0.18.2": e6c4cf63
+  "v0.18.1": 54074011
+  "v0.18.0": 3b11162d
+  "v0.17.2": 7a9ad1ca
+  "v0.17.1": 72f9ea2e
+  "v0.17.0": bb797f70
+  "v0.16.4": 74471b8a
+  "v0.16.3": ca5f3d51
+  "v0.16.2": 5ac5e4c2
+  "v0.16.1": 48041a88
+  "v0.16.0": 60df9f99
+  "v0.15.1": f44fc2d6
+  "v0.15.0": 42aec6c0
+  "v0.14.2": 592d10c3
+  "v0.14.1": 85b7622e
+  "v0.14.0": a7129c05
+  "v0.13.0": 50c2d102
+  "v0.12.3": 030409d4
+  "v0.12.2": 15956101
+  "v0.12.1": 559dfce0
+  "v0.12.0": eb7a13f4
+  "v0.11.1": b4698f4c
+  "v0.11.0": e7fb2c06
+  "v0.10.0": 23ce7740
+  "v0.9.2": 908463d4
+  "v0.9.1": 940bbc85
+  "v0.9.0": 080a8205
+  "v0.8.1": 483a4e74
+  "v0.8.0": 91302dc4
+  "v0.7.0": 5883c71c
+  "v0.6.0": 94f8fcaa
+  "v0.5.1": 5b496342
+  "v0.5.0": f986ef06
+  "v0.4.0": eab6e372
+  "v0.3.76": d4bb7314
+  "v0.3.75": f4cbbd4b
+  "v0.3.73": ac329747
+  "v0.3.72": d9758f36
+  "v0.3.71": 3a711845
+  "v0.3.70": 047f4543
+  "v0.3.69": 822a8f53
+  "v0.3.68": 323f6dd8
+  "v0.3.67": 4f8bfdf1
+  "v0.3.66": cf95523c
+  "v0.3.65": e0295dff
+  "v0.3.64": b93b2252
+  "v0.3.63": 91bd0e39
+  "v0.3.62": ff99b31c
+  "v0.3.61": 3ca48550
+  "v0.3.60": a7ab9646
+  "v0.3.59": 66253ba1
+  "v0.3.58": 10704eaa
+  "v0.3.57": 67a7bceb
+  "v0.3.56": 342318cd
+  "v0.3.55": bf738910
+  "v0.3.54": 0224d936
+  "v0.3.53": b10ddac0
+  "v0.3.52": 2a8e0096
+  "v0.3.51": 2e360b2f
+  "v0.3.50": 198f4ac6
+  "v0.3.49": 037d5d8c
+  "v0.3.48": a92aa58b
+  "v0.3.47": 295aceff
+  "v0.3.46": 46ef61dd
+  "v0.3.45": a8c769b6
+  "v0.3.44": 3899ae9e
+  "v0.3.43": 2526e22f
+  "v0.3.41": f15a2902
+  "v0.3.40": 24608eaf
 ---
 
+<Update label="v0.25.1" description="2026年6月16日">
+
+**合作伙伴节点更新**
+* [**Kling V3-Turbo**](https://github.com/Comfy-Org/ComfyUI/pull/14528): 添加了对 Kling V3-Turbo 模型的支持
+
+</Update>
+
+<Update label="v0.25.0" description="2026年6月16日">
+
+**新的开源模型支持**
+* [**SeedVR2**](https://github.com/Comfy-Org/ComfyUI/pull/14110)：SeedVR2视频超分辨率模型支持
+* [**Depth Anything 3**](https://github.com/Comfy-Org/ComfyUI/pull/13853)：来自LiheYoung的单目深度估计模型
+* [**SCAIL-2**](https://github.com/Comfy-Org/ComfyUI/pull/14373)：用于增强视频编辑的角色替换模型
+* [**Bernini-R**](https://github.com/Comfy-Org/ComfyUI/pull/14216)：Wan视频模型支持
+* **Ideogram 4**：修复了fp8数据类型问题以提高稳定性
+* **10位视频支持**：为核心节点添加了10位视频输入/输出支持
+
+**新节点**
+* [**PreviewGaussianSplat + PreviewPointCloud**](https://github.com/Comfy-Org/ComfyUI/pull/14194)：用于高斯泼溅和点云可视化的新3D预览节点
+* [**Color Primitive**](https://github.com/Comfy-Org/ComfyUI/pull/14260)：工作流中的颜色类型和实用工具
+* **改进的ResolutionSelector** ([#14309](https://github.com/Comfy-Org/ComfyUI/pull/14309))：增强的分辨率选择，具有更好的默认值
+
+**合作伙伴节点更新**
+* **Bria透明视频背景 + 绿色背景 + 替换背景** 节点
+* **Gemini文本节点** ([#14299](https://github.com/Comfy-Org/ComfyUI/pull/14299))：新的LLM合作伙伴节点
+* **Runway Aleph2** ([#14306](https://github.com/Comfy-Org/ComfyUI/pull/14306))：新的视频生成节点
+* **Krea 2 Medium Turbo** 模型支持
+* **Tripo3D导入3D** 节点
+* **Flux擦除**：添加了种子输入
+* **NanoBanana**：添加了温度（temperature）和top_p控制
+* **Flux KV缓存**：修复了分割批次时的崩溃
+
+**性能与稳定性**
+* **Aimdo 0.4.10** + `--reserve-vram` 和 `--vram-headroom` 选项
+* Comfy Aimdo 0.4.9，具有可靠性改进
+* 媒体加载：泛化的 `--high-ram` 选项
+* **资产API**：基于游标的分页、摄取时提取图像尺寸、WebSocket消息中的资产ID
+* 在CUDA 130+上始终启用CUDA malloc
+* 强制将cudnn.benchmark设为否以获得确定性结果
+* 修复了未对齐宽度图像/视频解码中的奇数高度崩溃和边缘溢出
+* 修复了未对齐宽度下的非确定性视频解码
+* **Ideogram 4**：修复了在cleanup_models_gc()中不要重置转换缓冲区的问题
+* **训练器**：修复了训练节点中条件变为可训练的问题
+
+</Update>
+
+<Update label="v0.24.1" description="2026年6月3日">
+
+**合作伙伴节点更新**
+* 新增 Bria Green Background 节点
+* 新增 Bria Transparent Video Background 节点，支持 SaveWEBM 中的透明度通道
+* 新增 Krea 2 Medium Turbo 模型
+* 为 Flux Erase 节点新增种子输入
+
+**问题修复**
+* 修复了 Seedance 2.0 1080p 首帧/末帧拉伸跳跃问题
+
+</Update>
+
 <Update label="v0.24.0" description="2026年6月3日">
 
 **新增开源模型支持**
@@ -92,6 +245,32 @@ translationFrom: changelog/index.mdx
 
 </Update>
 
+<Update label="v0.22.3" description="2026年5月20日">
+
+**合作伙伴节点更新**
+* 新增 Krea2 图像节点
+* 改进了使用 memoryview 的 SeeDance 2 视频参考上传
+
+</Update>
+
+<Update label="v0.22.2" description="2026年5月20日">
+
+**合作伙伴节点更新**
+* 新增 Rodin2.5 节点，并包含高质量网格选项（usdz 导出已禁用）
+* 修复了 Rodin2.5 中种子参数始终会被传递至服务器的问题
+
+</Update>
+
+<Update label="v0.22.1" description="2026年5月20日">
+
+**合作伙伴节点更新**
+* 新增 OpenRouter LLM 节点
+* 为 Anthropic 节点新增推理小部件
+* 为 ByteDance2Reference 节点新增自动图像放大小部件
+* 修复了向 Grok LLM 传递图像的问题
+
+</Update>
+
 <Update label="v0.22.0" description="2026年5月20日">
 
 **开源模型支持**
@@ -202,6 +381,35 @@ translationFrom: changelog/index.mdx
 
 </Update>
 
+<Update label="v0.20.3" description="2026年4月27日">
+
+**合作伙伴节点更新**
+* 新增 Luma UNI-1 模型
+* 新增 GPT 5.5 和 5.5-pro LLM 模型
+* 新增 grok-imagine-image-quality 模型
+* 全新 NanoBanana2 节点，支持 DynamicCombo/Autogrow
+
+**错误修复**
+* 修复了 ImageBlend/ImageCompositeMasked 处理具有不同通道数（透明度通道）的图像时的问题
+* 修复了 Motion Control 节点中 Kling V3 的价格标签
+
+</Update>
+
+<Update label="v0.20.2" description="2026年4月27日">
+
+**合作伙伴节点更新**
+* 添加了 Topaz Astra 2 模型（现已成为默认上采样器）
+* 面向常规图像的字节跳动虚拟肖像库
+* 为 GPTImage2 节点添加了自定义分辨率支持
+* 移除了 Moonvalley API 节点
+* 增加了合作伙伴 API 节点任务的默认超时时间
+
+**模型支持**
+* SDPose 调整大小修复
+* OneTainer ERNIE LoRA 支持
+
+</Update>
+
 <Update label="v0.20.1" description="2026年4月27日">
 
 **新增模型支持**
@@ -279,6 +487,34 @@ translationFrom: changelog/index.mdx
 
 </Update>
 
+<Update label="v0.19.5" description="2026年4月17日">
+
+**合作伙伴节点更新**
+* 为 Stable Diffusion 2 添加了 SD2 Real Human 支持
+* 在 Kling 节点中增加了 4K 分辨率
+* 添加了 GPTImage 价格徽章修复和新分辨率
+
+**错误修复**
+* 修复了 veo-3.0 模型的 4K 分辨率限制
+* 修复了 Windows 上 Stable_Zero123 cc_projection 权重赋值问题（aimdo-enabled）
+
+</Update>
+
+<Update label="v0.19.4" description="2026年4月17日">
+
+**合作伙伴节点更新**
+* 为 Veo 模型和 Veo 3 Lite 模型添加了 4K 分辨率支持
+* 添加了 gpt-image-2 作为版本选项
+* 为字节跳动 2 节点添加了自动视频降级
+* 添加了 SD2 Real Human 支持
+
+**问题修复**
+* 修复了在 Windows 上（启用 aimdo 时）Stable_Zero123 cc_projection 权重分配的问题
+* 修复了 veo-3.0 模型的 4K 分辨率限制
+* 修复了 GPTImage 价格徽章
+
+</Update>
+
 <Update label="v0.19.3" description="2026年4月17日">
 
 **LTX 文本生成**
@@ -369,6 +605,32 @@ translationFrom: changelog/index.mdx
 
 </Update>
 
+<Update label="v0.18.5" description="2026年3月25日">
+
+**稳定性**
+* fp8 checkpoint 修复回滚后的额外测试
+
+</Update>
+
+<Update label="v0.18.4" description="2026年3月25日">
+
+**错误修复**
+* 修复了 Windows 固定内存核算的问题，降低了 OOM 错误的风险
+* 修复了 fp8 缩放的模型加载不正确的问题
+* 回退了造成性能退化的 fp8 模型修复
+
+**合作节点更新**
+* 新增 Wan2.7 合作节点
+
+</Update>
+
+<Update label="v0.18.3" description="2026年3月25日">
+
+**合作伙伴节点更新**
+* 在 API 节点中添加了新的 Topaz 模型
+
+</Update>
+
 <Update label="v0.18.2" description="2026年3月25日">
 
 **合作伙伴节点 (API) **
@@ -376,6 +638,7 @@ translationFrom: changelog/index.mdx
 - Grok 视频扩展
 - [相关博客](https://blog.comfy.org/p/grok-imagine-model-feature-updates)
 </Update>
+
 <Update label="v0.18.1" description="2026年3月23日">
 
 **FP16 支持修复**
@@ -387,6 +650,7 @@ translationFrom: changelog/index.mdx
 - 修复了 WAN VAE 处理中的光照和颜色问题，提升了输出质量和色彩准确性
 
 </Update>
+
 <Update label="v0.18.0" description="2026年3月21日">
 
 **内存与性能优化**
@@ -423,6 +687,7 @@ translationFrom: changelog/index.mdx
 - 修复了罕见边缘情况下的各种内存泄露和损坏问题
 
 </Update>
+
 <Update label="v0.17.2" description="2026年3月15日">
 
 此版本专注于稳定性和错误修复。完整的技术详细信息请查看 GitHub 上的[完整更新日志](https://github.com/Comfy-Org/ComfyUI/compare/v0.17.1...v0.17.2)。
@@ -558,6 +823,7 @@ translationFrom: changelog/index.mdx
 - 工作流模板更新至 0.9.7 版本
 
 </Update>
+
 <Update label="v0.15.1" description="2026年2月26日">
 
 **错误修复与稳定性**
@@ -581,6 +847,7 @@ translationFrom: changelog/index.mdx
 - 将工作流模板更新至 0.9.4 版本
 
 </Update>
+
 <Update label="v0.15.0" description="2026年2月24日">
 
 **新节点和功能**
@@ -617,6 +884,7 @@ translationFrom: changelog/index.mdx
 - 在迁移期间为 LTX 2.0 工作流提供临时兼容性修复
 
 </Update>
+
 <Update label="v0.14.2" description="2026年2月18日">
 
 **错误修复**
@@ -636,6 +904,231 @@ translationFrom: changelog/index.mdx
 - 将工作流模板更新至v0.8.43
 
 </Update>
+
+<Update label="v0.14.0" description="2026年2月17日">
+
+**新增开源模型支持**
+* [**Qwen 2512 ControlNet / Fun ControlNet**](https://github.com/Comfy-Org/ComfyUI/pull/12359)：支持基于 Qwen 的 ControlNet 模型，以增强图像控制
+* **NAG 实现**：应用于所有基于 Flux 模型的 Nesterov 加速梯度
+
+**新增节点与 API**
+* [**VideoSlice 节点**](https://github.com/Comfy-Org/ComfyUI/pull/12107)：用于精确时间控制的视频切片工具
+* [**节点替换 API**](https://github.com/Comfy-Org/ComfyUI/pull/12014)：重大架构变更，允许在工作流中替换节点
+* **创建列表节点** ([#12173](https://github.com/Comfy-Org/ComfyUI/pull/12173))：列表创建工具节点
+
+**性能与技术改进**
+* 左填充文本编码器注意力遮罩支持
+* 用于 Flux 模型的 Torch RMSNorm 及混元视频重构
+* 动态显存：训练修复、原始 fp8 LoRA 质量修复
+* ModelPatcherDynamic：强制加载非叶子权重
+* 修复卸载 + 动态显存模式下的 LoRA 提取
+* 为 Flux 模型使用 torch RMSNorm 并重构混元视频
+
+**合作伙伴节点更新**
+* **Magnific Upscalers** API 节点已启用
+* **腾讯 3D 节点**：ModelTo3DUV、3DTextureEdit、3DParts
+* **Bria RMBG** API 节点
+* **Kling API 节点**：V3、O3 模型
+
+**训练器改进**
+* 具备适当卸载功能的训练（KohakuBlueleaf）
+* 内建 LoRA 训练现可在 Anima 上运行
+
+</Update>
+
+<Update label="v0.13.0" description="2026年2月10日">
+
+**新模型支持与改进**
+* [**EasyCache: LTX2 支持**](https://github.com/Comfy-Org/ComfyUI/pull/12231)：EasyCache 现在可与 LTX2 一起使用，加快视频生成速度
+* **ACE Step 1.5** ([#12311](https://github.com/Comfy-Org/ComfyUI/pull/12311))：现在无需 LLM 即可工作，修复了基础模型默认工作流，改进了文本编码
+* **Cosmos-Predict2 和 Anima 的 FP16** ([#12249](https://github.com/Comfy-Org/ComfyUI/pull/12249))：减少支持模型的内存占用
+* **让色调映射 Latent 适配任意维度的 Latent** ([#12363](https://github.com/Comfy-Org/ComfyUI/pull/12363))：实现了更灵活的色调映射
+
+**性能与修复**
+* 针对 ACE 1.5 性能的动态 VRAM 修复 + VRAM 泄漏修复
+* 惰性缓存修复 ([#12344](https://github.com/Comfy-Org/ComfyUI/pull/12344))
+* 禁用 LTXV2 的提示权重
+* 前端升级至 1.38.13
+
+**合作伙伴节点更新**
+* **Kling API 节点**：新增 V3、O3 模型
+* Moonvalley API 节点修复
+
+</Update>
+
+<Update label="v0.12.3" description="2026年2月5日">
+
+**ACE 第1.5步改进**
+* [**LLM采样选项**](https://github.com/Comfy-Org/ComfyUI/pull/12295)：添加了采样选项和参考音频支持
+* **文本编码改进** ([#12283](https://github.com/Comfy-Org/ComfyUI/pull/12283))：为ACE提供了更好的文本处理
+* **VAE分块解码用于音频** ([#12299](https://github.com/Comfy-Org/ComfyUI/pull/12299))
+* 禁用了ACE第1.5步的Sage注意力
+* 修复了某些硬件/PyTorch配置上的NaN问题
+
+**3D改进**
+* 在Load3D节点上增加了File3DAny输出；SaveGLB接受File3DAny输入
+
+</Update>
+
+<Update label="v0.12.2" description="2026年2月4日">
+
+**ACE Step 1.5**
+* [**4B LM model support**](https://github.com/Comfy-Org/ComfyUI/pull/12257)：支持 4B ACE step 1.5 语言模型
+* 修复动态 VRAM 模式下的 VRAM 泄漏
+
+</Update>
+
+<Update label="v0.12.1" description="2026年2月3日">
+
+**ACE步骤1.5修复**
+* [**进度条**](https://github.com/Comfy-Org/ComfyUI/pull/12242)：为ACE步骤添加了进度条
+* **基础模型 LoRAs** 支持
+* **分块VAE修复** 适用于ACE
+* Mac兼容性修复
+
+**3D与API**
+* API中的基础3D模型文件类型
+* LLM logits 转换为 comfy-weight
+
+</Update>
+
+<Update label="v0.12.0" description="2026年2月3日">
+
+**新模型支持**
+* [**ACE Step 1.5**](https://github.com/Comfy-Org/ComfyUI/pull/12237): 对 ACE Step 1.5 音频/语音生成模型的基本支持 — 一个重要的新模型系列
+* [**用于 LLM 文本生成的 KV 缓存**](https://github.com/Comfy-Org/ComfyUI/pull/12195): 通过缓存键值对提升运行 llama 模型的性能
+* [**自适应模型加载**](https://github.com/Comfy-Org/ComfyUI/pull/11845): 减少内存使用，修复显存溢出，并修复 Windows 共享内存溢出
+
+**新节点**
+* **颜色类型 + 颜色转 RGB 整数节点** ([#12145](https://github.com/Comfy-Org/ComfyUI/pull/12145))
+
+**API 节点**
+* **HitPaw API 节点** ([#12117](https://github.com/Comfy-Org/ComfyUI/pull/12117))
+* **Recraft** 风格节点
+* **Vidu**: Q3 模型、扩展和多帧节点
+
+**开发者**
+* 资产第二部分 - 更多端点
+* 通过 v1/v3 schema 向发送 is_input_list 到前端
+
+</Update>
+
+<Update label="v0.11.1" description="2026年1月29日">
+
+**合作伙伴节点更新**
+* [**Grok Imagine API nodes**](https://github.com/Comfy-Org/ComfyUI/pull/12136): 通过 Grok 生成图像
+* Manager 已更新至 4.1b1
+
+**开发者**
+* 开发专用节点支持
+* Python 3.14 兼容性说明
+
+</Update>
+
+<Update label="v0.11.0" description="2026年1月27日">
+
+**新的开源模型支持**
+* [**Anima 模型**](https://github.com/Comfy-Org/ComfyUI/pull/12012)：对 Anima 模型系列的支持，包含 fp16 优化
+* [**Multi/InfiniteTalk**](https://github.com/Comfy-Org/ComfyUI/pull/10179)：针对 LTXV 的大规模多说话人支持
+* [**Flux2 LyCORIS LoKr**](https://github.com/Comfy-Org/ComfyUI/pull/11997)：针对 Flux2 的 LoKr（基于 KronA）LoRA 支持
+* **LTX2 Tiny VAE** (taeltx_2) 支持
+* **Z-Image Omni** 基础模型支持
+* Flux2 Klein 的 **ModelScope LoRA** 格式
+
+**新节点与功能**
+* **Magnific API 节点** ([#11986](https://github.com/Comfy-Org/ComfyUI/pull/11986))：通过 Magnific 进行图像放大
+* **Tencent Hunyuan3D API 节点** ([#12026](https://github.com/Comfy-Org/ComfyUI/pull/12026))
+* **WaveSpeed API 节点** ([#11945](https://github.com/Comfy-Org/ComfyUI/pull/11945))
+* 节点架构中的**搜索别名**字段，以便更好地发现节点
+* 针对前端任意输入的 **kwargs 输入**
+* 针对小网格 IC-LoRA 的 **RoPE 位置调整**
+
+**性能**
+* Qwen VAE 和 Wan 的速度/显存优化
+* LTX2 VAE 显存减少
+* 为提升显存效率而重构的 LTX2 前向传播
+
+**模型增强**
+* 使用 SaveCheckpoint 保存的 Flux2 Klein 模型现在能正常加载
+* Qwen 3 0.6B 模型的配置
+* 动态检测色度辐射补丁大小
+
+</Update>
+
+<Update label="v0.10.0" description="2026年1月21日">
+
+**新功能**
+* [**API节点自动发现**](https://github.com/Comfy-Org/ComfyUI/pull/11943)：所有 `nodes_*.py` 文件现在会自动被发现，简化了节点开发
+* **高级参数支持**（[#11939](https://github.com/Comfy-Org/ComfyUI/pull/11939)）：输入类现在支持高级控件
+* **CLIP视觉输出中添加了图像尺寸**（[#11923](https://github.com/Comfy-Org/ComfyUI/pull/11923)）
+
+**合作伙伴节点更新**
+* **字节跳动 Seedance 1.5 Pro** API节点
+* **Bria Edit** 节点（[#11978](https://github.com/Comfy-Org/ComfyUI/pull/11978)）
+* Autogrow 验证改进
+
+**错误修复**
+* 为LTXV VAE编码将单声道音频转换为伪立体声（[#11965](https://github.com/Comfy-Org/ComfyUI/pull/11965)）
+* Flux2 Klein 内存估算调整
+* AMD GPU张量类型修复
+
+</Update>
+
+<Update label="v0.9.2" description="2026年1月15日">
+
+**新模型支持**
+* [**Flux2 Klein**](https://github.com/Comfy-Org/ComfyUI/pull/11890)：重大新模型家族支持 — Flux2 Klein 具备扩展的模型能力
+* [**Z-Image ControlNet Lite**](https://github.com/Comfy-Org/ComfyUI/pull/11849)：阿里云-Pai Z-Image ControlNet 的轻量版本
+
+**新节点**
+* **Meshy 3D API 节点** ([#11843](https://github.com/Comfy-Org/ComfyUI/pull/11843))
+* **ResizeImageMaskNode**：添加裁切至倍数模式
+* **蓝图目录**用于内置模板
+
+**性能与技术**
+* NVFP4 LoRA 优化（多轮）
+* 节流的 ProgressBar WebSocket 更新
+* Lanczos 灰度图像放大修复
+* VAELoader 元数据加载
+* 仓库 URL 已更新为 Comfy-Org/ComfyUI
+
+</Update>
+
+<Update label="v0.9.1" description="2026年1月13日">
+
+**问题修复**
+- 更新 LTXAV 内存估算
+
+</Update>
+
+<Update label="v0.9.0" description="2026年1月13日">
+
+**新特性**
+* [**Basic Asset Support**](https://github.com/Comfy-Org/ComfyUI/pull/11315)：模型资产管理的基础——模型增删改查操作端点
+* **NVFP4 Model LoRA** ([#11837](https://github.com/Comfy-Org/ComfyUI/pull/11837))：LoRA 支持现已兼容 NVFP4 量化模型
+* [**SigLIP 2 NAFlex**](https://github.com/Comfy-Org/ComfyUI/pull/11831)：支持 SigLIP 2 NAFlex CLIP 视觉模型
+* [**图像对比节点**](https://github.com/Comfy-Org/ComfyUI/pull/11343)：直观对比两个图像
+* [**JoinAudioChannels**](https://github.com/Comfy-Org/ComfyUI/pull/11728)：音频通道合并工具
+
+**模型支持**
+* Z-Image 模型的 ModelScope LoRA ([#11805](https://github.com/Comfy-Org/ComfyUI/pull/11805))
+* 修复 Chroma 和文本编码器的 FP4/FP8 问题
+* VAEEncodeForInpaint 现支持 Wan万相 VAE 的 tuple downscale_ratio
+* 通过高效时间步嵌入处理降低 LTX2 显存占用
+
+**API 节点**
+* **Vidu2** 节点 ([#11760](https://github.com/Comfy-Org/ComfyUI/pull/11760))
+* **Topaz Enhance** 修复
+* **Gemini** 安全拦截处理
+
+**技术相关**
+* AMD gfx1200：默认启用 PyTorch 注意力
+* CI 容器版本自动更新
+* 重构模型加载以降低内存使用
+* 前端补丁更新至 1.36.14
+
+</Update>
+
 <Update label="v0.8.1" description="2026年1月8日">
 
 **模型支持**
@@ -654,6 +1147,7 @@ translationFrom: changelog/index.mdx
 - 将工作流模板更新到v0.7.69
 
 </Update>
+
 <Update label="v0.8.0" description="2026年1月7日">
 
 **新增模型与节点**
@@ -711,6 +1205,7 @@ translationFrom: changelog/index.mdx
 - 图像处理节点转换为 V3 架构
 - 优化 Lumina/Z 图像模型（移除未使用组件）
 </Update>
+
 <Update label="v0.6.0" description="2025年12月24日">
 
 **新模型支持**
@@ -731,6 +1226,7 @@ translationFrom: changelog/index.mdx
 - 修复在 --gpu-only 模式下 ZImageFunControlNet 的掩码连接问题
 
 </Update>
+
 <Update label="v0.5.1" description="2025年12月18日">
 
 - 新增 GPT-Image-1.5 API 节点
@@ -806,6 +1302,7 @@ translationFrom: changelog/index.mdx
 - 请从[这里](https://www.comfy.org/download)下载最新的桌面版本以获取最新功能。
 
 </Update>
+
 <Update label="v0.3.76" description="2025年12月2日">
 
 **前端界面与用户体验**
@@ -835,6 +1332,7 @@ translationFrom: changelog/index.mdx
 - Kling O1 模型支持
 
 </Update>
+
 <Update label="v0.3.75" description="2025年11月26日">
 
 **新模型支持**
@@ -881,6 +1379,7 @@ translationFrom: changelog/index.mdx
 - 更新前端至 v1.30.6 并升级 Transformers 库
 
 </Update>
+
 <Update label="v0.3.71" description="2025年11月21日">
 
 **模型兼容性与增强功能**
@@ -903,6 +1402,7 @@ translationFrom: changelog/index.mdx
 - 修复了工作流命名问题，提升了复杂处理管道的整体稳定性
 
 </Update>
+
 <Update label="v0.3.70" description="2025年11月19日">
 
 **CUDA 12.6支持与分发**
@@ -923,6 +1423,7 @@ translationFrom: changelog/index.mdx
 - 改进了发布自动化和分发流程，确保更可靠的更新
 
 </Update>
+
 <Update label="v0.3.69" description="2025年11月18日">
 
 **内存与性能优化**
@@ -944,6 +1445,7 @@ translationFrom: changelog/index.mdx
 - 统一所有模型的RoPE函数实现
 
 </Update>
+
 <Update label="v0.3.68" description="2025年11月5日">
 **模型支持**
 - 支持 [ChronoEdit 14B](https://research.nvidia.com/labs/toronto-ai/chronoedit/) 模型
@@ -971,6 +1473,7 @@ translationFrom: changelog/index.mdx
 - 工作流模板更新至v0.2.11
 
 </Update>
+
 <Update label="v0.3.67" description="2025年10月28日">
 
 **API 节点**
@@ -992,7 +1495,8 @@ translationFrom: changelog/index.mdx
 - 模板更新到版本 0.2.4
 
 </Update>
-<Update label="v0.3.66" description="October 21, 2025">
+
+<Update label="v0.3.66" description="2025年10月21日">
 
 **前端更新**
 - **子图组件编辑**：可通过新的参数面板直接编辑子图组件，无需进入子图内部
@@ -1079,7 +1583,8 @@ translationFrom: changelog/index.mdx
 - **选择工具箱重设计**：重新设计了节点选择工具箱
 
 </Update>
-<Update label="v0.3.62" description="2025年10月1日">
+
+<Update label="v0.3.62" description="2025年9月30日">
 - **Rodin3D-Gen2 参数修复**
 - **Seedance Pro 模型支持**
 </Update>
@@ -1102,6 +1607,7 @@ translationFrom: changelog/index.mdx
 - **采样器 CFG 增强**：在采样器 CFG 函数参数中添加了 'input_cond' 和 'input_uncond' 参数，提供更灵活的条件控制
 
 </Update>
+
 <Update label="v0.3.60" description="2025年9月23日">
 
 **新模型支持**
@@ -1126,6 +1632,7 @@ translationFrom: changelog/index.mdx
 - **前端版本更新**：更新到版本 1.26.13
 
 </Update>
+
 <Update label="v0.3.59" description="2025年9月10日">
 
 **字节跳动 Seedream 4.0 集成**
@@ -1171,6 +1678,7 @@ translationFrom: changelog/index.mdx
 - **减少 Windows 上的 RAM 使用量**
 
 </Update>
+
 <Update label="v0.3.55" description="2025年8月29日">
 
 **模型支持**
@@ -1231,6 +1739,7 @@ translationFrom: changelog/index.mdx
 - **文档清理**：从自述文件中移除未完全实现的模型，以避免用户混淆
 
 </Update>
+
 <Update label="v0.3.52" description="2025年8月23日">
 
 **增强模型支持和 Qwen Image ControlNet 集成等**
@@ -1260,7 +1769,7 @@ translationFrom: changelog/index.mdx
 - **导航模式回滚**：回滚导航默认到传统旧版模式，避免默认启用新版标准导航模式导致用户体验问题，用户仍旧可以在设置中启用标准导航模式
 </Update>
 
-<Update label="v0.3.51" description="2025年08月20日">
+<Update label="v0.3.51" description="2025年8月20日">
 
 **模型支持**
 - **Qwen-Image-Edit模型**：原生支持 Qwen-Image-Edit
@@ -1328,6 +1837,7 @@ translationFrom: changelog/index.mdx
 - **Kling API改进**：修复了Kling Image API节点的图像类型参数
 
 </Update>
+
 <Update label="v0.3.49" description="2025年8月5日">
 
 **界面优化与模型支持**
@@ -1356,6 +1866,7 @@ translationFrom: changelog/index.mdx
 - **模板更新**：更新了多个模板版本
 
 </Update>
+
 <Update label="v0.3.48" description="2025年8月2日">
 
 **API增强与性能优化**
@@ -1379,6 +1890,7 @@ translationFrom: changelog/index.mdx
 - **模板更新**：更新了多个模板版本
 
 </Update>
+
 <Update label="v0.3.47" description="2025年7月30日">
 
 **内存优化与大模型性能**
@@ -1394,6 +1906,7 @@ translationFrom: changelog/index.mdx
 - **内存分配改进**：改进了多大模型用户的内存管理
 
 </Update>
+
 <Update label="v0.3.46" description="2025年7月28日">
 
 **硬件加速和音频处理**
@@ -1434,6 +1947,7 @@ translationFrom: changelog/index.mdx
 - **代码清理**：移除了已弃用的代码
 
 </Update>
+
 <Update label="v0.3.45" description="2025年7月21日">
 
 **采样与训练功能改进**
@@ -1475,6 +1989,7 @@ translationFrom: changelog/index.mdx
 - **文档**：增强了fast_fp16_accumulation的文档
 
 </Update>
+
 <Update label="v0.3.44" description="2025年7月8日">
 
 **采样和模型控制增强**
@@ -1516,6 +2031,7 @@ translationFrom: changelog/index.mdx
 - **文件验证**：增强检查点加载安全措施
 
 </Update>
+
 <Update label="v0.3.43" description="2025年6月30日">
 
 **模型支持与工作流可靠性**
@@ -1559,7 +2075,6 @@ translationFrom: changelog/index.mdx
 
 </Update>
 
-
 <Update label="v0.3.40" description="2025年6月5日">
 
 **工作流工具和性能优化**