@enslo/sd-metadata 2.1.0 → 2.1.1

package/CHANGELOG.md

@@ -0,0 +1,403 @@
+# Changelog
+
+All notable changes to this project will be documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [2.1.1] - 2026-02-24
+
+### Fixed
+
+- **Malformed input hardening** (#163): Add defensive checks for corrupted or malicious binary data across all format readers and writers
+  - PNG writer: validate chunk boundaries before slicing
+  - EXIF reader: early bailout for implausibly large tag data sizes
+  - JPEG reader: minimum segment length validation for APP1 and COM markers
+  - WebP reader: fix off-by-one in chunk iteration boundary check
+  - A1111 parser: input length guard for settings regex
+
+### Maintenance
+
+- Migrate to standard monorepo structure (#160)
+- Use standard APIs for binary operations (#162)
+
+## [2.1.0] - 2026-02-22
+
+### Added
+
+- **IIFE build for userscript support** (#157): Add `dist/index.global.js` (minified IIFE bundle) that exposes a `sdMetadata` global variable, enabling userscripts (Tampermonkey, Violentmonkey, etc.) to load the library via `@require`
+
+## [2.0.1] - 2026-02-19
+
+### Fixed
+
+- **ComfyUI DetailerForEach support** (#150): Extract prompt, negative prompt, and sampling settings from workflows using `DetailerForEach` (ComfyUI Impact Pack inpainting/detailer nodes)
+- **ComfyUI SamplerCustom support** (#150): Extract metadata from `SamplerCustom` nodes with dedicated handling for its input layout (cfg on node, noise_seed via reference, steps/scheduler from BasicScheduler)
+- **ComfyUI conditioning chain traversal** (#150): Follow conditioning through intermediate nodes (e.g. ControlNet apply) to reach CLIPTextEncode for prompt text extraction
+
+## [2.0.0] - 2026-02-19
+
+### ⚠️ Breaking Changes
+
+- **Forge family software identifiers** (#129): The `'forge'` identifier now refers specifically to the current [Forge](https://github.com/lllyasviel/stable-diffusion-webui-forge) main line. Images previously detected as `'forge'` (which was a catch-all for all `f`-prefixed versions) are now split into distinct identifiers:
+
+  | v1.x `software` | v2.0.0 `software` | Tool |
+  | --- | --- | --- |
+  | `'forge'` | `'forge'` | [Forge](https://github.com/lllyasviel/stable-diffusion-webui-forge) (current main line) |
+  | `'forge'` | `'forge-classic'` | [Forge Classic](https://github.com/Haoming02/sd-webui-forge-classic/tree/classic) |
+  | `'forge'` | `'reforge'` | [reForge](https://github.com/Panchovix/stable-diffusion-webui-reForge) |
+  | `'forge'` | `'easy-reforge'` | [EasyReforge](https://github.com/Zuntan03/EasyReforge) |
+
+  If your code checks `metadata.software === 'forge'`, update it to handle the new identifiers. TypeScript users with exhaustive `switch` statements on `GenerationSoftware` will see compile errors for the new values.
+
+### Added
+
+- **`embed()` function** (#126): Write metadata in SD WebUI (A1111) format to any supported image
+  - Unlike `write()` (which preserves the original format), `embed()` always writes in A1111 plain-text format
+  - Accepts user-created `EmbedMetadata` or parsed `GenerationMetadata` directly
+  - Optional `extras` field on `EmbedMetadata` for arbitrary key-value pairs in the settings line (e.g., `Version`, `Lora hashes`)
+  - Extras override structured fields at their original position; new keys append at the end
+  - Supports PNG (tEXt/iTXt), JPEG (Exif), and WebP (Exif)
+- **`stringify()` function** (#125): Unified metadata formatting for display
+  - Accepts `ParseResult`, `GenerationMetadata`, or `EmbedMetadata` directly
+  - `ParseResult`: `success` → SD WebUI format, `unrecognized` → raw text, `empty`/`invalid` → empty string
+  - `GenerationMetadata` / `EmbedMetadata`: directly formats as SD WebUI text
+- **`softwareLabels` constant**: Read-only mapping from `GenerationSoftware` identifiers to human-readable display names
+- **`EmbedMetadata` type**: User-created custom metadata — `BaseMetadata & Pick<NovelAIMetadata, 'characterPrompts'> & { extras? }`
+- **`BaseMetadata` type export**: Previously internal, now available for direct use
+- **`GenerationSoftware` type export**: String literal union of all supported software identifiers
+
+### Fixed
+
+- **TensorArt parser** (#138): Extract sampler and scheduler from the ComfyUI node graph. Previously these fields were missing because `generation_data` does not contain them.
+- **ComfyUI CLIP skip** (#137): Extract `clipSkip` from the `CLIPSetLastLayer` node. Previously, CLIP skip was silently dropped from all ComfyUI-based tool output (ComfyUI, TensorArt, Stability Matrix).
+- **Fooocus parser** (#130): Rewrite based on actual Fooocus source code analysis. The previous implementation was non-functional due to reading from the wrong metadata location. Now supports both JSON and A1111 text metadata schemes.
+- **Easy Diffusion parser** (#131): Improved based on source code analysis. Removed phantom key format that never appeared in actual images, improved detection reliability, and added upscale/denoise field extraction.
+
+### Changed
+
+- **SD WebUI family detection** (#129): Explicit Version-field pattern matching for SD WebUI (`v` + digit) and all Forge variants. Added real SD WebUI sample files, promoting Stable Diffusion WebUI from experimental to fully supported.
+
+### Deprecated
+
+- **`writeAsWebUI()`**: Use `embed()` instead (now a thin wrapper around `embed()`)
+- **`formatAsWebUI()`**: Use `stringify()` instead (now a thin wrapper around internal `buildEmbedText()`)
+- **`formatRaw()`**: Use `stringify()` instead
+
+### Migration Guide
+
+| v1.x                            | v2.0.0                   | Notes                                     |
+| ------------------------------- | ------------------------ | ----------------------------------------- |
+| `writeAsWebUI(image, metadata)` | `embed(image, metadata)` | Also accepts user-created `EmbedMetadata` |
+| `formatAsWebUI(metadata)`       | `stringify(metadata)`    | Also accepts `ParseResult` directly       |
+| `formatRaw(raw)`                | `stringify(readResult)`  | Handles all statuses automatically        |
+| `software === 'forge'`          | Check all Forge variants | See Breaking Changes above                |
+
+## [1.8.1] - 2026-02-15
+
+### Fixed
+
+- **A1111 detection criteria** (#122): Fix `writeAsWebUI` round-trip for minimal metadata (no sampling settings)
+  - Detection now accepts `Steps:` OR `Sampler:` OR `Negative prompt:` (was requiring both `Steps:` AND `Sampler:`)
+
+### Changed
+
+- **Unify detection and parsing responsibilities** (#122): `detectSoftware()` is now the single source of truth for format identification
+  - Remove redundant detection guards from parsers (a1111, novelai, comfyui, fooocus, ruined-fooocus)
+  - Remove fallback parser chain from router
+
+## [1.8.0] - 2026-02-14
+
+### Added
+
+- **ComfyUI SamplerCustomAdvanced support** (#118): Parse distributed node workflows
+  - Trace sampling parameters across RandomNoise, CFGGuider, KSamplerSelect, and BasicScheduler nodes
+  - Support UNETLoader as alternative model loader (fallback from CheckpointLoaderSimple)
+  - Resolve hires scale from rgthree node references (`clip_scale`)
+- **HF Space upscale settings** (#119): Extract `use_upscaler` as hires settings
+  - `upscale_method` → `hires.upscaler`
+  - `upscaler_strength` → `hires.denoise`
+  - `upscale_by` → `hires.scale`
+
+### Changed
+
+- **Bundle size optimization** (#117): Internal refactoring to reduce bundle size and consolidate codebase (no public API changes)
+
+## [1.7.1] - 2026-02-01
+
+### Fixed
+
+- **PNG chunk encoding strategies** (#114): Corrected encoding for format conversion
+  - saveimage-plus and CivitAI now use Unicode escape encoding (was incorrectly using raw UTF-8)
+  - Stability Matrix and TensorArt now use per-chunk encoding to match original behavior
+    - `parameters` (Stability Matrix) and `generation_data` (TensorArt) use raw UTF-8
+    - Other chunks use Unicode escape encoding
+  - This ensures Japanese and other non-ASCII characters are correctly preserved during round-trip conversion
+
+## [1.7.0] - 2026-02-01
+
+### Added
+
+- **ComfyUI extended node support** (#108): Support for additional custom nodes
+  - **Prompt nodes**: Power Prompt (rgthree), DF_Text_Box (Derfuu)
+  - **Latent nodes**: SDXL Empty Latent Image (rgthree)
+  - **Seed nodes**: CR Seed
+  - **Hires detection**: LatentUpscale, LatentUpscaleBy
+- **InvokeAI upscale settings** (#109): Extract upscale information from InvokeAI images
+  - `upscale.upscaler`: Upscaler model name (e.g., RealESRGAN_x4plus_anime_6B)
+  - `upscale.scale`: Scale factor
+- **ComfyUI denoise in SamplingSettings** (#110): Extract denoising strength from KSampler nodes
+  - Only captured when `denoise < 1.0` (img2img workflows)
+  - `denoise = 1.0` (txt2img default) is omitted
+
+### Changed
+
+- **Internal refactoring** (#107): Type cleanup and developer experience improvements (no public API changes)
+
+## [1.6.1] - 2026-01-29
+
+### Fixed
+
+- **CivitAI Orchestration JPEG→PNG conversion** (#102): Store all metadata in a single `prompt` chunk instead of separating into multiple chunks, improving round-trip compatibility
+  - This fixes an issue introduced in v1.6.0 where the chunk structure was not optimal
+
+## [1.6.0] - 2026-01-29
+
+### Added
+
+- **`strict` option for `read()`** (#95): Control dimension extraction behavior
+  - When `true`, dimensions come strictly from metadata only (returns 0 if not present)
+  - When `false` (default), missing dimensions fall back to image headers
+
+### Fixed
+
+- **CivitAI detection and format conversion** (#96): CivitAI images are now correctly detected and can be converted between formats
+  - Previously, CivitAI Orchestration images were incorrectly detected as ComfyUI or SD WebUI
+  - Images using Hires fix or Upscaler on CivitAI now convert to PNG correctly (previously became `unrecognized`)
+
+### Changed
+
+- **Parser refactoring** (#96, #98): Improved code quality across all parsers
+  - Immutable patterns for better maintainability
+  - Function decomposition for ComfyUI parser
+
+## [1.5.0] - 2026-01-26
+
+### Added
+
+- **ArrayBuffer support** (#89): All public API functions (`read`, `write`, `writeAsWebUI`) now accept both `Uint8Array` and `ArrayBuffer` as input
+  - Browser users can pass `ArrayBuffer` directly from `File.arrayBuffer()` or `fetch().arrayBuffer()` without manual conversion
+  - Node.js `Buffer` (which extends `Uint8Array`) continues to work as before
+
+## [1.4.2] - 2026-01-25
+
+### Fixed
+
+- **JPEG/WebP Format Conversion** (#85): Fixed metadata parsing for HF-Space, Ruined Fooocus, TensorArt, and Stability Matrix after cross-format conversion (PNG ↔ JPEG/WebP)
+  - HF-Space and Ruined Fooocus: Now correctly parsed from Comment JSON (was `unrecognized`)
+  - TensorArt and Stability Matrix: Now correctly detected from Comment JSON (was misidentified as `comfyui`)
+- **Unicode Round-trip** (#85): Non-ASCII characters (e.g., Japanese) are now preserved during PNG → JPEG/WebP → PNG conversions
+
+### Changed
+
+- **Internal Refactoring** (#86): Removed unused `exifSoftware` and `exifDocumentName` support (no public API changes). These Exif tags were not populated in real NovelAI files.
+
+## [1.4.1] - 2026-01-24
+
+### Fixed
+
+- **NovelAI WebP Data Loss** (#79): Fixed an issue where `Software` and `Title` (DocumentName) tags were lost when converting NovelAI images to WebP. They are now correctly preserved in Exif tags.
+- **NovelAI Version Detection** (#79): Improved detection of NovelAI version strings to better support newer versions.
+
+### Changed
+
+- **Internal Refactoring** (#80): Decoupled encoding strategies from specific tool names for better maintainability (no public API changes).
+
+## [1.4.0] - 2026-01-24
+
+### ⚠️ BREAKING CHANGES
+
+- **Remove `force` option from `write()`** (#76): Blind conversion for unrecognized formats is no longer supported.
+- **Metadata drop behavior**: When converting unrecognized metadata to a different format (e.g., PNG → JPEG), metadata is now dropped instead of attempting a blind conversion.
+
+### Added
+
+- **`WriteWarning` type** (#76): Notifies when metadata was intentionally dropped during conversion (e.g., unrecognized cross-format).
+- **Graceful unrecognized handling**: Unrecognized metadata is still preserved when writing back to the same format or using `writeAsWebUI` with structured metadata.
+
+### Documentation
+
+- Updated API examples to use the new `result.warning` pattern.
+
+## [1.3.0] - 2026-01-23
+
+### Added
+
+- **`formatRaw()` function** (#72): Format raw metadata as plain text for display
+  - Works with unrecognized metadata (`status: 'unrecognized'`)
+  - Multiple entries separated by blank lines
+  - Ideal for fallback display when parsing fails
+
+### Changed
+
+- **Migrate to pnpm** (#74): Replace npm with pnpm for faster, more disk-efficient package management
+- **Upgrade Vitest to v4** (#74): Fix esbuild security vulnerability (GHSA-67mh-4wv8-2f99)
+
+### Documentation
+
+- Improve README quality and fix package import names (#73)
+- Restructure Usage section with collapsible advanced examples
+- Emphasize `formatAsWebUI` cross-tool normalization capability
+
+## [1.2.0] - 2026-01-22
+
+### Added
+
+- **`writeAsWebUI()` function** (#69): Write any GenerationMetadata to PNG/JPEG/WebP in SD WebUI (A1111) format
+  - Create custom metadata from scratch
+  - Convert metadata from any tool to WebUI-compatible format
+  - Automatic encoding strategy (tEXt for ASCII, iTXt for non-ASCII in PNG)
+- **`formatAsWebUI()` function** (#69): Format metadata as human-readable text in SD WebUI format
+  - Tool-agnostic standard format for displaying generation metadata
+  - Supports NovelAI character prompts with position comments
+  - Normalizes line endings for cross-platform compatibility
+
+## [1.1.1] - 2026-01-21
+
+### Documentation
+
+- Update README CDN usage example to reference v1.1.0
+- Add README version update step to release workflow
+- Remove trailing slash from package.json homepage URL
+
+## [1.1.0] - 2026-01-21
+
+### ⚠️ BREAKING CHANGES
+
+- **Replace `workflow` with `nodes` in ComfyUIMetadata** (#58): The `workflow?: unknown` field has been replaced with typed `nodes: ComfyNodeGraph`.
+  - **What `workflow` was**: ComfyUI UI layout data containing node positions, sizes, link connections, and visual groupings—information used by ComfyUI's graph editor but not required for image reproduction.
+  - **What `nodes` is**: The actual node graph (from PNG `prompt` chunk) mapping node IDs to their inputs and class types—the essential data needed to reproduce the generation.
+  - **Why this change**: Previously, `workflow` only existed in ComfyUI images and was `undefined` for other ComfyUI-compatible tools (TensorArt, Stability Matrix, SwarmUI). Now `nodes` provides actual node graph data from all tools.
+- **Remove `type` field from GenerationMetadata** (#57): Use `metadata.software` instead of `metadata.type` for type narrowing.
+- **Internal types no longer exported** (#60): `BaseMetadata` and `GenerationSoftware` are now internal types.
+
+### Added
+
+- **SwarmUI workflow preservation** (#59): Store ComfyUI node graph in `exifMake` segment when converting PNG to JPEG/WebP, enabling complete round-trip preservation
+- **Typed ComfyUI node graph** (#58): New types `ComfyNode`, `ComfyNodeGraph`, `ComfyNodeReference`, `ComfyNodeInputValue` for type-safe workflow access
+- **Expanded type exports** (#60): Export `ModelSettings`, `SamplingSettings`, `HiresSettings`, `UpscaleSettings`, `CharacterPrompt`, and chunk/segment types
+- **Type documentation** (#60): Comprehensive type reference at `docs/types.md`
+
+### Fixed
+
+- **TensorArt seed extraction** (#61): Extract actual seed from KSampler node when `generation_data.seed` is -1 (random placeholder)
+
+### Changed
+
+- **Metadata type consolidation** (#60): Renamed `A1111Metadata` to `StandardMetadata`, merged `InvokeAIMetadata` into `StandardMetadata`
+- **ComfyUIMetadata restructure** (#60): Split into `BasicComfyUIMetadata` (nodes required) and `SwarmUIMetadata` (nodes optional)
+
+### Documentation
+
+- Improved GenerationMetadata documentation in README (#62)
+- Added comprehensive type reference docs (#60)
+- Improved README usage examples (#56)
+
+## [1.0.2] - 2026-01-19
+
+### Fixed
+
+- **Incorrect SD-WebUI Detection**: Fixed false positives where non-AI generated images (e.g., from photo editing software) were incorrectly detected as SD-WebUI/A1111 format (#47)
+- **ComfyUI Detection**: Improved detection for ComfyUI files, fixing cases where valid ComfyUI metadata was not recognized (#47, #50)
+
+### Changed
+
+- Refactored detection logic with tier-based structure for improved maintainability (#47)
+
+## [1.0.1] - 2026-01-13
+
+### Changed
+
+- Updated package homepage URL to point to the demo site (<https://sd-metadata.pages.dev/>) instead of the GitHub repository
+
+## [1.0.0] - 2026-01-13
+
+### Added
+
+#### Core Features
+
+- **Multi-format Support**: PNG (tEXt / iTXt), JPEG (COM / Exif), WebP (Exif)
+- **Unified API**: Simple `read()` and `write()` functions work across all formats
+- **TypeScript Native**: Written in TypeScript with full type definitions included
+- **Zero Dependencies**: Works in Node.js and browsers without any external dependencies
+- **Format Conversion**: Seamlessly convert metadata between PNG, JPEG, and WebP
+- **Lossless Round-trip**: Preserves original metadata structure when converting back to native format
+- Image dimension extraction with automatic fallback from image headers
+
+#### Supported AI Tools
+
+**Fully Supported** (verified with sample files, native format support):
+
+- NovelAI - PNG and WebP native formats
+- ComfyUI - PNG native format
+- Stable Diffusion WebUI (AUTOMATIC1111) - Experimental (no samples yet)
+- Forge / Forge Neo - PNG, JPEG, and WebP native formats
+- InvokeAI - PNG native format
+- SwarmUI - PNG and WebP native formats
+
+**Extended Support** (sd-metadata specific parsers, cross-format conversion):
+
+- Civitai - Uses A1111-compatible format, JPEG primary format
+- TensorArt - Custom JSON format in PNG
+- Stability Matrix - Custom JSON format in PNG
+- HuggingFace Space - JSON format in Exif
+- Ruined Fooocus - PNG format
+
+**Experimental** (implemented from reference code, no sample verification):
+
+- Easy Diffusion
+- Fooocus
+
+#### Advanced Features
+
+- Segment/chunk tracking for lossless round-trip preservation
+- ComfyUI detection priority when both prompt and workflow exist
+- NovelAI WebP UTF-8 corruption auto-correction
+- Force conversion for unrecognized metadata formats with `{ force: true }` option
+
+### Known Limitations
+
+- **ComfyUI JPEG/WebP**: While reading supports major custom node formats (e.g., `save-image-extended`), writing always uses the `comfyui-saveimage-plus` format for best information preservation and ComfyUI drag-and-drop workflow compatibility
+- **NovelAI WebP**: Auto-corrects corrupted UTF-8 in Description field, breaking content-equivalent round-trip (intentional for data validity)
+- **SwarmUI PNG→JPEG/WebP**: Loses ComfyUI workflow data from `prompt` chunk to match native WebP format (metadata preserved)
+
+### Development
+
+- Comprehensive test suite with unit, sample, and integration tests
+- Test coverage reporting with Vitest
+- GitHub Actions CI with automated coverage reports
+- Lefthook pre-commit hooks for linting and type checking
+- Biome for code formatting and linting
+- CONTRIBUTING.md for community contributions
+
+[2.1.1]: https://github.com/enslo/sd-metadata/releases/tag/core@2.1.1
+[2.1.0]: https://github.com/enslo/sd-metadata/releases/tag/v2.1.0
+[2.0.1]: https://github.com/enslo/sd-metadata/releases/tag/v2.0.1
+[2.0.0]: https://github.com/enslo/sd-metadata/releases/tag/v2.0.0
+[1.8.1]: https://github.com/enslo/sd-metadata/releases/tag/v1.8.1
+[1.8.0]: https://github.com/enslo/sd-metadata/releases/tag/v1.8.0
+[1.7.1]: https://github.com/enslo/sd-metadata/releases/tag/v1.7.1
+[1.7.0]: https://github.com/enslo/sd-metadata/releases/tag/v1.7.0
+[1.6.1]: https://github.com/enslo/sd-metadata/releases/tag/v1.6.1
+[1.6.0]: https://github.com/enslo/sd-metadata/releases/tag/v1.6.0
+[1.5.0]: https://github.com/enslo/sd-metadata/releases/tag/v1.5.0
+[1.4.2]: https://github.com/enslo/sd-metadata/releases/tag/v1.4.2
+[1.4.1]: https://github.com/enslo/sd-metadata/releases/tag/v1.4.1
+[1.4.0]: https://github.com/enslo/sd-metadata/releases/tag/v1.4.0
+[1.3.0]: https://github.com/enslo/sd-metadata/releases/tag/v1.3.0
+[1.2.0]: https://github.com/enslo/sd-metadata/releases/tag/v1.2.0
+[1.1.1]: https://github.com/enslo/sd-metadata/releases/tag/v1.1.1
+[1.1.0]: https://github.com/enslo/sd-metadata/releases/tag/v1.1.0
+[1.0.2]: https://github.com/enslo/sd-metadata/releases/tag/v1.0.2
+[1.0.1]: https://github.com/enslo/sd-metadata/releases/tag/v1.0.1
+[1.0.0]: https://github.com/enslo/sd-metadata/releases/tag/v1.0.0

package/README.ja.md

@@ -63,7 +63,7 @@ npm install @enslo/sd-metadata
 > \* フォーマット固有の動作があるツール。詳細は[フォーマット固有の動作](#フォーマット固有の動作)を参照してください。
 
 > [!TIP]
-> **ツールサポートの拡大にご協力ください！** 実験的なツール（Easy Diffusion、Fooocus）やサポートされていないツールのサンプル画像を募集しています。これらのAIツールで生成したサンプル画像をお持ちの方は、ぜひご提供ください！詳細は[CONTRIBUTING.md](CONTRIBUTING.md)を参照してください。
+> **ツールサポートの拡大にご協力ください！** 実験的なツール（Easy Diffusion、Fooocus）やサポートされていないツールのサンプル画像を募集しています。これらのAIツールで生成したサンプル画像をお持ちの方は、ぜひご提供ください！詳細は[CONTRIBUTING.md](https://github.com/enslo/sd-metadata/blob/main/CONTRIBUTING.md)を参照してください。
 
 ## フォーマット固有の動作
 
@@ -145,7 +145,7 @@ fileInput.addEventListener('change', async (e) => {
 // ==UserScript==
 // @name        My Script
 // @namespace   https://example.com
-// @require     https://cdn.jsdelivr.net/npm/@enslo/sd-metadata@2.1.0/dist/index.global.js
+// @require     https://cdn.jsdelivr.net/npm/@enslo/sd-metadata@2.1.1/dist/index.global.js
 // ==/UserScript==
 
 const response = await fetch(imageUrl);
@@ -586,25 +586,7 @@ type RawMetadata =
 
 ## 開発
 
-```bash
-# 依存関係をインストール
-pnpm install
-
-# テストを実行
-pnpm test
-
-# ビルド
-pnpm build
-
-# リント
-pnpm lint
-
-# 型チェック
-pnpm typecheck
-
-# デモサイト起動
-pnpm demo
-```
+開発のセットアップとガイドラインについては、[コントリビューションガイド](https://github.com/enslo/sd-metadata/blob/main/CONTRIBUTING.md)を参照してください。
 
 ## ライセンス
 

package/README.md

@@ -63,7 +63,7 @@ When you convert from a native format to an extended format and back (e.g., PNG
 > \* Tools with format-specific behaviors. See [Format-Specific Behaviors](#format-specific-behaviors) for details.
 
 > [!TIP]
-> **Help us expand tool support!** We're actively collecting sample images from experimental tools (Easy Diffusion, Fooocus) and unsupported tools. If you have sample images generated by these or other AI tools, please consider contributing them! See [CONTRIBUTING.md](CONTRIBUTING.md) for details.
+> **Help us expand tool support!** We're actively collecting sample images from experimental tools (Easy Diffusion, Fooocus) and unsupported tools. If you have sample images generated by these or other AI tools, please consider contributing them! See [CONTRIBUTING.md](https://github.com/enslo/sd-metadata/blob/main/CONTRIBUTING.md) for details.
 
 ## Format-Specific Behaviors
 
@@ -145,7 +145,7 @@ For userscripts (Tampermonkey, Violentmonkey, etc.), load the IIFE build via `@r
 // ==UserScript==
 // @name        My Script
 // @namespace   https://example.com
-// @require     https://cdn.jsdelivr.net/npm/@enslo/sd-metadata@2.1.0/dist/index.global.js
+// @require     https://cdn.jsdelivr.net/npm/@enslo/sd-metadata@2.1.1/dist/index.global.js
 // ==/UserScript==
 
 const response = await fetch(imageUrl);
@@ -586,25 +586,7 @@ For detailed documentation of all exported types including `ModelSettings`, `Sam
 
 ## Development
 
-```bash
-# Install dependencies
-pnpm install
-
-# Run tests
-pnpm test
-
-# Build
-pnpm build
-
-# Lint
-pnpm lint
-
-# Type check
-pnpm typecheck
-
-# Start demo site
-pnpm demo
-```
+See the [Contributing Guide](https://github.com/enslo/sd-metadata/blob/main/CONTRIBUTING.md) for development setup and guidelines.
 
 ## License