feat: OUTPUT节点图片来源标签 + 连线悬停高亮 + DashScope Qwen + GPT Image修复

CLAUDE.md

@@ -0,0 +1,109 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
+## Project Overview
+
+**Infinite-Canvas (AI Studio)** — a monolithic AI image/video/LLM generation platform supporting multiple API providers through a unified UI. Single-developer project, Chinese-language primary audience.
+
+## Architecture
+
+```
+main.py              — FastAPI monolith (~11,700 lines): API routes, data models, all business logic
+static/              — Frontend: vanilla HTML/JS/CSS (no bundler, no framework)
+  index.html         — Main studio shell (SPA with iframe-based page routing)
+  canvas.html        — Node-based workflow canvas for image/video generation pipelines
+  smart-canvas.html  — Chat-like AI composer with inline image generation
+  gpt-chat.html      — LLM chat interface
+  enhance.html       — Image enhancement/upscaling
+  klein.html         — Flux2-Klein model interface
+  zimage.html        — Z-Image model interface
+  online.html        — Online API image generation
+  angle.html         — ModelScope angle control
+  api-settings.html  — API provider configuration UI
+  asset-manager.html — Local asset library management
+  comfyui-settings.html — Local ComfyUI instance management
+  js/                — Page-specific JS modules (canvas.js, smart-canvas.js, etc.)
+  js/i18n/           — Internationalization (zh-CN + en), loaded by i18n.js
+  vendor/            — Locally mirrored third-party libs (Tailwind CSS, Lucide icons, Three.js)
+  system-prompts/    — Markdown prompt templates for smart-canvas
+  runninghub/        — Static RunningHub provider config + thumbnails
+data/                — JSON file storage (api_providers, conversations, canvases, prompt libraries, etc.)
+API/                 — `.env` file for API keys (gitignored)
+assets/              — input/output/library/uploads directories for generated/uploaded media
+workflows/           — ComfyUI workflow JSON files
+packages/            — Offline pip wheel cache for portable Python
+python/              — Bundled portable Python runtime (Windows)
+tools/               — PowerShell scripts for Jimeng CLI install/login
+```
+
+## Running the Project
+
+```bash
+# Windows — double-click run.bat or:
+python\python.exe main.py
+
+# macOS — double-click mac-启动服务.command or:
+python3 main.py
+
+# The server starts on http://127.0.0.1:3000/
+```
+
+No build step. The frontend is static HTML served by FastAPI's `StaticFiles` mount. CDN dependencies are mirrored locally in `static/vendor/` for offline use.
+
+## Key Technical Details
+
+### Backend (`main.py`)
+
+- **Framework**: FastAPI with CORS middleware (allow all origins), WebSocket support for live stats
+- **Data persistence**: JSON files in `data/` directory, protected by per-resource `threading.Lock` instances
+- **API providers** are configured in `data/api_providers.json`, managed via `/api/providers` endpoints. Provider protocols: `openai`, `apimart`, `gemini`, `volcengine`, `runninghub`, `jimeng`
+- **API keys** stored in `API/.env` (key=value format), read via `load_env_file()`
+- **Image generation flow**: `/api/online-image` (single) or `/api/canvas-image-tasks` (batched, sequential queue with `QUEUE_LOCK`)
+- **Video generation**: `/api/canvas-video` with provider-specific dispatch (Volcengine Ark, Jimeng/dreamina CLI, RunningHub, OpenAI-compatible)
+- **LLM chat**: `/api/canvas-llm` and `/api/conversations` endpoints, conversation state stored as JSON files
+- **Local ComfyUI**: Polls configurable ComfyUI instances (`COMFYUI_INSTANCES` list), proxy image viewing through `/api/view`
+- **ModelScope**: API calls to `modelscope.cn` for free model access, uses repo file API (not raw web)
+- **Jimeng (即梦)**: CLI-based integration — spawns `dreamina` as subprocess for login, submit, and poll
+- **DashScope Qwen**: `generate_dashscope_qwen_image()` for Aliyun DashScope Qwen image generation. `is_qwen_image_model()` and `is_dashscope_image_provider()` added for provider dispatch
+- **GPT Image**: `is_gpt_image_model()` generalizes the old `is_gpt_image_2_model()` to support all GPT image models. Multipart form uses `image[]` field name when multi-image
+- **Auto-update**: `/api/check-update` checks GitHub/ModelScope for new VERSION; `/api/update-from-github` stages and applies updates with rollback support
+- **App version** stored in `VERSION` file (format: `YYYY.MM.DD`), automatically appended as query param to static assets for cache busting
+
+### Frontend
+
+- **No framework** — vanilla JS with DOM manipulation
+- **Tailwind CSS** loaded from local mirror (`static/vendor/js/tailwindcss-cdn.js`)
+- **Icons**: Lucide (local mirror)
+- **3D**: Three.js 0.160.0 (local mirror) — used in canvas for node graph visualization
+- **i18n**: Custom system in `static/js/i18n/` — `StudioI18n` global, translations loaded as JS modules. `tr(key)` for translation, `langIsEn()` for language check
+- **Theme**: Dark/light via CSS custom properties on `html.theme-dark`, toggled by `theme.js`
+- **Inter-page communication**: `window.postMessage` for cross-iframe events (lang change, canvas updates, provider changes)
+- **Version cache busting**: Static assets get `?v=<VERSION>` query strings generated by `versioned_static_html()` in Python
+
+### Data Model
+
+- **Canvas save/sync**: Uses optimistic concurrency — `saveCanvas()` sends `base_updated_at`, server returns 409 if stale (another tab saved first). `touchCanvasOpened()` bumps `updated_at` on open. Polling (`checkRemoteCanvasVersion()`) checks `/api/canvases/{id}/meta` every 2.5s. WebSocket broadcasts ignored when `client_id` matches own `CLIENT_ID`.
+- **Event delegation**: Node hover detection uses `board mousemove` + `e.target.closest('.node')` — no per-element listeners. Drag/resize/knife modes skip hover to avoid conflicts. in `data/canvases/<id>.json`. Node types:
+  - `image` — reference image (paste/drag/upload)
+  - `prompt` — text prompt
+  - `generator` — API image generation (OpenAI-compatible providers)
+  - `msgen` — ModelScope free image generation (Z-Image, Qwen-Edit, custom models)
+  - `rh` — RunningHub workflow/AI app execution
+  - `comfy` — Local ComfyUI (text-to-image, enhance, edit, custom workflow)
+  - `ltxDirector` — ComfyUI LTXDirector video with multi-segment timeline
+  - `video` — API video generation
+  - `llm` — LLM chat node
+  - `output` — Collects generated images/videos from upstream nodes
+  - `group` / `promptGroup` — Logical grouping of nodes
+  - `loop` — Serial/parallel batch execution with configurable count
+  - Output nodes now display `providerLabel` and `modelLabel` tags on each image (set in `appendOutputImages()`)
+  - Link hover highlighting via `hoveredNodeId` in board `mousemove` event delegation (`renderLinks()` adds `link-hover` class)
+- **Asset Library**: `data/asset_library.json` — metadata index for files in `assets/library/`
+- **Prompt Libraries**: `data/prompt_libraries.json` — categorized prompt templates
+- **History**: `history.json` — task execution log with results
+- **Conversations**: `data/conversations/<id>.json` — LLM chat history
+
+## File Size Warning
+
+`main.py` is ~11,700 lines. When editing, use targeted search (Grep) to locate the relevant section. Major section markers (comments like `# --- 路由接口 ---`) help navigation. The file is too large to read in one operation — always use `offset`/`limit` parameters.

README.md

@@ -25,8 +25,9 @@ https://www.fhl.mom/register?aff=86L574B4T2N9  （包含codex和GPT image 2模
 3. 火山引擎调用（人脸认证还在修复bug）
 4. Modelscope免费LLM模型和图像模型调用
 5. 即梦CLI调用，可直接调用即梦高级会员的积分，支持文生图/图生图/文生视频/图生视频
-6. 支持调用本地局域网的ComfyUI
-7. 扩展图片/360全景图预览截图/视频帧抽取/循环节点等诸多功能
+6. 支持阿里云百炼 DashScope 兼容模式下的 LLM 调用，并对 `qwen-image-*` 图片模型自动走千问原生图片生成/编辑接口
+7. 支持调用本地局域网的ComfyUI
+8. 扩展图片/360全景图预览截图/视频帧抽取/循环节点等诸多功能
 
 --------
 

graphify-out/GRAPH_REPORT.md

@@ -0,0 +1,157 @@
+# Graph Report - d:/ziyong/Infinite-Canvas  (2026-06-07)
+
+## Corpus Check
+- 56 files · ~953,777 words
+- Verdict: corpus is large enough that graph structure adds value.
+
+## Summary
+- 5379 nodes · 14474 edges · 25 communities detected
+- Extraction: 37% EXTRACTED · 63% INFERRED · 0% AMBIGUOUS · INFERRED: 9060 edges (avg confidence: 0.5)
+- Token cost: 0 input · 0 output
+
+## God Nodes (most connected - your core abstractions)
+1. `tr()` - 106 edges
+2. `tr()` - 92 edges
+3. `Vector3` - 76 edges
+4. `push()` - 62 edges
+5. `render()` - 59 edges
+6. `map()` - 59 edges
+7. `render()` - 56 edges
+8. `handleClick()` - 56 edges
+9. `scheduleSave()` - 55 edges
+10. `Vector2` - 55 edges
+
+## Surprising Connections (you probably didn't know these)
+- None detected - all connections are within the same source files.
+
+## Communities
+
+### Community 0 - "Three.js Object Helpers"
+Cohesion: 0.0
+Nodes (92): ArrowHelper, AxesHelper, BatchedMesh, Box2, Box3, Box3Helper, BoxGeometry, BoxHelper (+84 more)
+
+### Community 1 - "Smart Canvas Composer"
+Cohesion: 0.01
+Nodes (734): activeAssetCategory(), activeAssetLibrary(), activeComposerNode(), activeInputImagesFor(), activePromptLibrary(), activePromptTemplateGroups(), activePromptTemplateNodeId(), activeSettingsSubject() (+726 more)
+
+### Community 2 - "Three.js Core Primitives"
+Cohesion: 0.0
+Nodes (267): addContour(), addUniform(), allocTexUnits(), AmbientLight, AnimationClip, AnimationLoader, AnimationObjectGroup, ArcCurve (+259 more)
+
+### Community 3 - "Canvas Workflow Graph"
+Cohesion: 0.01
+Nodes (698): actionFailed(), activeCanvasAssetCategory(), activeCanvasAssetLibrary(), activeCanvasMediaCategory(), activeCanvasPromptLibrary(), activeCanvasPromptLibraryItems(), activeCanvasPromptTemplateGroups(), activeCanvasWorkflowCategory() (+690 more)
+
+### Community 4 - "FastAPI Backend Core"
+Cohesion: 0.01
+Nodes (689): BaseModel, Exception, add_asset_library_item(), add_prompt_library_category(), add_prompt_library_item(), ai_config(), AIReference, api_headers() (+681 more)
+
+### Community 5 - "Tailwind CSS Library"
+Cohesion: 0.01
+Nodes (510): _a(), aa(), ac(), add(), addToError(), Ae(), after(), Ah() (+502 more)
+
+### Community 6 - "API Provider Settings"
+Cohesion: 0.03
+Nodes (170): addModel(), addMsLora(), addProvider(), applyDetectedProtocol(), applyModelPicker(), applyProviderOnboardingDefaults(), applyRhEditorGraphTransform(), applyRhImageSlotDefaults() (+162 more)
+
+### Community 7 - "Asset Manager UI"
+Cohesion: 0.04
+Nodes (167): activeAssetCategory(), activeAssetLibrary(), activeAvatarProvider(), activeLocalFolder(), activePromptCategories(), activePromptLibrary(), activeWorkflowCategory(), activeWorkflowLibrary() (+159 more)
+
+### Community 8 - "Three.js Animation"
+Cohesion: 0.02
+Nodes (11): AnimationAction, AnimationMixer, Audio, AudioAnalyser, AudioListener, CubicInterpolant, DiscreteInterpolant, Interpolant (+3 more)
+
+### Community 9 - "ComfyUI Instance Settings"
+Cohesion: 0.06
+Nodes (78): addComfyInstance(), addDropdownOption(), addMiniNode(), applyActiveRandomValues(), applyGraphTransform(), applyLanguage(), attachPanZoom(), bindMiniCanvas() (+70 more)
+
+### Community 10 - "LTX Director Timeline"
+Cohesion: 0.1
+Nodes (6): beforeRegisterNodeDef(), clamp(), hideWidget(), isCanvasLTXNode(), parseInitial(), TimelineEditor
+
+### Community 11 - "Three.js Math Utils"
+Cohesion: 0.06
+Nodes (3): Euler, makeClipAdditive(), Quaternion
+
+### Community 12 - "Agent HTML Builder"
+Cohesion: 0.18
+Nodes (18): build(), export_config(), extract_summary(), find_callout(), infer_meta(), _inline_md(), main(), md_to_html() (+10 more)
+
+### Community 13 - "Theme System"
+Cohesion: 0.23
+Nodes (13): applyScale(), applyTheme(), autoScale(), broadcastScale(), currentScaleMode(), ensureScaleStyle(), isFramed(), normalizeScaleMode() (+5 more)
+
+### Community 14 - "get-pip Bootstrap"
+Cohesion: 0.31
+Nodes (9): bootstrap(), determine_pip_install_arguments(), include_setuptools(), include_wheel(), main(), monkeypatch_for_cert(), Install setuptools only if absent, not excluded and when using Python <3.12., Install wheel only if absent, not excluded and when using Python <3.12. (+1 more)
+
+### Community 15 - "Lucide Icon Library"
+Cohesion: 0.36
+Nodes (8): Ba(), cA(), dA(), iA(), ka(), MA(), Pa(), za()
+
+### Community 16 - "i18n Core Engine"
+Cohesion: 0.39
+Nodes (7): apply(), entries(), lang(), register(), set(), t(), toggle()
+
+### Community 17 - "Jimeng CLI Installer"
+Cohesion: 0.48
+Nodes (4): Convert-ToWslPath(), Invoke-WslScript(), Invoke-WslScriptCapture(), New-WslScriptFile()
+
+### Community 18 - "History Bulk Manager"
+Cohesion: 0.6
+Nodes (4): attach(), fmt(), injectStyles(), tr()
+
+### Community 19 - "Jimeng CLI Login"
+Cohesion: 0.6
+Nodes (3): Convert-ToWslPath(), Invoke-WslScript(), New-WslScriptFile()
+
+### Community 20 - "Image Preview Utility"
+Cohesion: 0.67
+Nodes (0): 
+
+### Community 21 - "i18n Module Loader"
+Cohesion: 1.0
+Nodes (0): 
+
+### Community 22 - "i18n Common Strings"
+Cohesion: 1.0
+Nodes (0): 
+
+### Community 23 - "i18n Studio Strings"
+Cohesion: 1.0
+Nodes (0): 
+
+### Community 24 - "i18n Validator"
+Cohesion: 1.0
+Nodes (0): 
+
+## Knowledge Gaps
+- **105 isolated node(s):** `Install setuptools only if absent, not excluded and when using Python <3.12.`, `Install wheel only if absent, not excluded and when using Python <3.12.`, `Patches `pip install` to provide default certificate with the lowest priority.`, `首次运行时提前创建配置目录，避免第一次保存 API Key 时才创建目录/文件。`, `保存 API 设置后，将 os.environ 里最新的值同步回模块级全局变量，     避免保存后需要重启才能生效。` (+100 more)
+  These have ≤1 connection - possible missing edges or undocumented components.
+- **Thin community `i18n Module Loader`** (1 nodes): `i18n.js`
+  Too small to be a meaningful cluster - may be noise or needs more connections extracted.
+- **Thin community `i18n Common Strings`** (1 nodes): `common.js`
+  Too small to be a meaningful cluster - may be noise or needs more connections extracted.
+- **Thin community `i18n Studio Strings`** (1 nodes): `studio.js`
+  Too small to be a meaningful cluster - may be noise or needs more connections extracted.
+- **Thin community `i18n Validator`** (1 nodes): `validate-i18n.js`
+  Too small to be a meaningful cluster - may be noise or needs more connections extracted.
+
+## Suggested Questions
+_Questions this graph is uniquely positioned to answer:_
+
+- **Why does `Vector3` connect `Three.js Object Helpers` to `Three.js Core Primitives`?**
+  _High betweenness centrality (0.008) - this node is a cross-community bridge._
+- **Why does `Vector2` connect `Three.js Object Helpers` to `Three.js Core Primitives`?**
+  _High betweenness centrality (0.007) - this node is a cross-community bridge._
+- **Why does `Vector4` connect `Three.js Object Helpers` to `Three.js Core Primitives`?**
+  _High betweenness centrality (0.006) - this node is a cross-community bridge._
+- **Are the 105 inferred relationships involving `tr()` (e.g. with `performUndo()` and `trf()`) actually correct?**
+  _`tr()` has 105 INFERRED edges - model-reasoned connections that need verification._
+- **Are the 91 inferred relationships involving `tr()` (e.g. with `trf()` and `actionFailed()`) actually correct?**
+  _`tr()` has 91 INFERRED edges - model-reasoned connections that need verification._
+- **Are the 61 inferred relationships involving `push()` (e.g. with `warn()` and `ve()`) actually correct?**
+  _`push()` has 61 INFERRED edges - model-reasoned connections that need verification._
+- **Are the 58 inferred relationships involving `render()` (e.g. with `performUndo()` and `renderMinimap()`) actually correct?**
+  _`render()` has 58 INFERRED edges - model-reasoned connections that need verification._

graphify-out/cache/065b91bfd18c261defa77c93c7344bd92a47fd72dec5fa4451bdb2a96bf56d2c.json

@@ -0,0 +1 @@
+{"nodes": [{"id": "smart_canvas", "label": "smart-canvas.js", "file_type": "code", "source_file": "d:\\ziyong\\Infinite-Canvas\\static\\js\\i18n\\smart-canvas.js", "source_location": "L1"}], "edges": []}