[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-94934":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":9,"language":10,"languages":9,"totalLinesOfCode":9,"stars":11,"forks":12,"watchers":13,"openIssues":13,"contributorsCount":14,"subscribersCount":14,"size":14,"stars1d":14,"stars7d":15,"stars30d":15,"stars90d":14,"forks30d":14,"starsTrendScore":16,"compositeScore":17,"rankGlobal":9,"rankLanguage":9,"license":18,"archived":19,"fork":19,"defaultBranch":20,"hasWiki":21,"hasPages":19,"topics":22,"createdAt":9,"pushedAt":9,"updatedAt":23,"readmeContent":24,"aiSummary":25,"trendingCount":14,"starSnapshotCount":14,"syncStatus":26,"lastSyncTime":27,"discoverSource":28},94934,"ComfyUI_MiniMax_H3_Extender","tritant\u002FComfyUI_MiniMax_H3_Extender","tritant","ComfyUI node for MiniMax H3 that chains multiple video clips with motion context, disk caching, dynamic image references, audio reference support, and seamless final video\u002Faudio decoding.",null,"Python",129,13,1,0,22,3,56.64,"Apache License 2.0",false,"main",true,[],"2026-08-24 04:01:23","# ComfyUI MiniMax H3 Extender\n\nA ComfyUI custom node for **MiniMax H3** designed to generate long, continuous video sequences from multiple clips while preserving motion, visual continuity, and audio continuity between generations.\n\nThe node combines **Ref2VA conditioning, Motion Context, disk caching, multi-clip generation, image references, audio references, and final video\u002Faudio decoding** into a much simpler workflow.\n\n---\n## 🎬 New — Video & Audio References for MiniMax H3\n\nThe Extender now supports **MiniMax H3 video references** directly inside the workflow.\n\nYou can now use external video frames as `\u003CVideo N>` references, combine them with image references, and optionally attach the matching video audio for motion, timing, camera behavior and lip-sync guidance.\n\n### What’s new\n\n- up to **3 video references**\n- up to **3 matching video-audio references**\n- up to **3 standalone audio references**\n- `\u003CVideo 1>`, `\u003CVideo 2>`, `\u003CVideo 3>` support in prompts\n- stable logical video slots with no automatic remapping\n- dynamic AV inputs in the node UI\n- optional paired `ref_video_audio_N` inputs\n- long standalone audio references are automatically sliced along the clip timeline\n- video soundtracks are cropped to the effective reference-video duration\n- reference-video inputs are automatically aligned to H3’s required `17k+5` frame structure\n- external Prompt Pack and Reference Pack sockets remain at the bottom of the node\n\n### Automatic FPS correction\n\nMiniMax H3 expects reference-video frames at **24 fps**, but real source videos can be 23.976, 25, 30, 45, 50, 60 fps, etc.\n\nThe Extender now accepts the original source FPS through:\n\n- `ref_video_fps_1`\n- `ref_video_fps_2`\n- `ref_video_fps_3`\n\nThe reference-video frame batch is then automatically **resampled to 24 fps while preserving the original duration** before it is sent to H3.\n\nThis fixes cases where the end of a reference video appeared to be missing because a non-24-fps image batch was previously interpreted as if it were already 24 fps.\n\nTypical workflow:\n\n~~~text\nLoad Video\n   ↓\nGet Video Components\n   ├── images → ref_video_1\n   ├── fps    → ref_video_fps_1\n   └── audio  → ref_video_audio_1\n~~~\n\n### Character replacement\n\nVideo references can be combined with image references for workflows such as:\n\n- character replacement\n- motion transfer\n- pose and body-performance transfer\n- camera-motion transfer\n- timing preservation\n- lip-sync guidance from the original video audio\n\nFor example:\n\n- `\u003CPicture 1>` defines the new character identity\n- `\u003CVideo 1>` provides the original performance, timing and camera behavior\n- `ref_video_audio_1` provides the matching audio reference\n\nThis keeps the Extender’s existing internal image-reference system fully intact while adding proper H3 Ref2VA video\u002Faudio support.\n\n\n**Added support for an external prompt pack through the new MiniMax H3 Prompt Pack Bridge node.**\n\n## Reference Pack Bridge\n\nA new node **MiniMax H3 Reference Pack Bridge** has been added.\n\nThis allows external ComfyUI `IMAGE` outputs to be injected directly into the Extender’s existing internal reference slots, while keeping the current internal reference system fully intact.\n\nYou can now mix both approaches in the same workflow:\n\n- internal references loaded with the Extender\n- external references coming from `Load Image`, crop\u002Fresize nodes, video frames, Kontext-style preprocessing, or any other `IMAGE` source\n- each external input maps directly to its matching `\u003CPicture N>` slot\n- empty external slots leave the existing internal reference untouched\n- reference indices are never compacted or remapped\n- once imported, the external image becomes a normal internal reference with thumbnail, project persistence and Save\u002FLoad compatibility\n- disconnecting the external input does not remove the imported internal reference\n- unchanged external images are not rewritten on every run\n\nThe Bridge is optional and does not replace the current reference workflow. It simply adds a flexible external entry point for advanced ComfyUI pipelines.\n\n### 🎬 Save Preview\n\nThe **Final Decode \u002F Preview** node now includes a dedicated **Save Preview** button.\n\nIt saves the currently assembled Extender preview exactly as shown, including:\n\n- seam handling\n- assembled audio\n- per-clip color corrections\n- ComfyUI workflow metadata\n- ComfyUI prompt metadata\n\nThe saved MP4 can be dragged back into ComfyUI to restore the workflow.\n\n### 🎨 Real-Time Per-Clip Color Editor\n\nEach generated clip now has its own **color editor**, accessible directly from the clip card with the `🎨` button.\n\nYou can adjust:\n\n- Saturation\n- Contrast\n- Brightness\n\nThe editor provides a **live looping preview around the selected clip**, including a short part of the previous and next clips, making it much easier to visually match colors between continuations.\n\nCorrections are:\n\n- stored independently for every clip\n- editable again at any time\n- non-destructive\n- preserved in `.ext` Save\u002FLoad projects\n- automatically included in Preview, Full Batch and Save Preview\n\nColor correction is applied only to the decoded video and **does not modify Motion Context, latent data or clip validation**.\n\nA small `✓` next to the palette indicates that a clip has an active color correction.\n\n### ⚡ Full Batch Integration\n\nFull Batch correctly preserves and applies all previously configured clip corrections during final assembly.\n\nYou can therefore:\n\n1. Generate several clips\n2. Color-correct individual clips\n3. Continue generating additional clips\n4. Run Full Batch\n5. Obtain the complete sequence with all previous adjustments preserved\n\n### 💾 Save \u002F Load Project\n\nPer-clip color adjustments are stored inside `.ext` projects.\n\nLoading a project restores the clip settings and color correction state, allowing you to continue exactly where you stopped.\n\n- **Internal reference manager**\n\n  - Add thumb editor.\n  - Up to 9 image references can now be loaded directly inside the Extender.\n  - No external `Load Image` nodes are required anymore.\n  - References are shown as thumbnails directly in the node.\n  - Reference slots stay fixed, so prompt numbering remains stable.\n  - Double-click a thumbnail to view it larger.\n\n- **Portable `.ext` projects**\n  - `Save Project` now embeds the actual reference images inside the project archive.\n  - `Load Project` restores prompts, refs, validation state, cache, preview and resolution settings.\n  - A saved project can therefore be reopened on another machine without needing the original image files.\n\n- **Named clip cards**\n  - Each clip card can now have its own optional name, making long sequences much easier to organize.\n\n- **Improved audio joins** (work in progress)\n  - Audio is now rebuilt from PCM and encoded only once at the end instead of concatenating separate AAC streams.\n  - Additional gain matching, declicking and smooth entry ramps reduce audible bumps between clips.\n\n---\n\nhttps:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002Fa18cc6a5-2340-474e-9d3b-b784cd41584a\n\n\n\u003Cimg width=\"2307\" height=\"1028\" alt=\"image\" src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002Fc1126ae8-2d4b-416c-a8c2-b839cd4c6b15\" \u002F>\n\n\n\u003Cimg width=\"2376\" height=\"1163\" alt=\"Capture d'écran 2026-08-20 062155\" src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F072885ca-8faa-4d58-b525-4b12d57fe75c\" \u002F>\n\n\n\u003Cimg width=\"1860\" height=\"691\" alt=\"image\" src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F9f2354a4-d8b7-485d-904f-76481e8fba15\" \u002F>\n\n\n\u003Cimg width=\"2048\" height=\"1103\" alt=\"image\" src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002Ff67ba34a-5b8a-4d3b-9c18-d6314db2c873\" \u002F>\n\n\n\u003Cimg width=\"2091\" height=\"948\" alt=\"image\" src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F905dcd01-09f8-4dfe-9d20-2f949637f938\" \u002F>\n\n\n\u003Cimg width=\"2557\" height=\"1212\" alt=\"Capture d'écran 2026-08-15 083401\" src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F99ca1fc4-d8b9-4662-a869-1fa06e5e58e1\" \u002F>\n\n## Keeping References Consistent Across Clips\n\nWhen using image references, it is strongly recommended to place a `subject_definitions` block at the **beginning of every clip prompt**.\n\nThis helps MiniMax H3 keep the same subjects, identities, clothing, visual roles and even the environment associated with a reference image from one clip to the next.\n\n### Example\n\n> **subject_definitions:**  \n> `\u003CPicture 1>` is the reference image defining the exact visual appearance, identity, face, hairstyle, body proportions, clothing, accessories, and overall look of `\u003CSubject 1>`, as well as the established environment and visual context of the scene.  \n> `\u003CPicture 2>` is the reference image defining the exact visual appearance, identity, face, hairstyle, body proportions, clothing, accessories, and overall look of `\u003CSubject 2>`.  \n> `\u003CSubject 1>` is the exact same woman shown in `\u003CPicture 1>`.  \n> `\u003CSubject 2>` is her friend, the exact same woman shown in `\u003CPicture 2>`.  \n> `\u003CSubject 3>` is the same street environment and scene context established in `\u003CPicture 1>`.\n\nPlace this `subject_definitions` block at the **very beginning of every clip prompt**.\n\nThe important point is that a reference image does not have to represent only a character.\n\nA reference can also define the **environment, location or visual context** of the sequence.\n\nIn the example above:\n\n- `\u003CPicture 1>` defines the appearance of `\u003CSubject 1>`\n- `\u003CPicture 2>` defines the appearance of `\u003CSubject 2>`\n- `\u003CPicture 1>` also defines the street environment used as `\u003CSubject 3>`\n\nSo the same reference image can be used both to preserve a character **and** to preserve the environment established by that image.\n\nRepeating these definitions at the beginning of each clip prompt helps MiniMax H3 maintain the intended reference roles throughout the entire sequence.\n\n---\n\n## Features\n\n- Multi-clip MiniMax H3 generation\n- Continuous Motion Context between clips\n- Video and audio latent continuity\n- Disk-based latent cache\n- Clip-by-clip generation workflow\n- Full batch generation mode\n- Per-clip prompt\n- Per-clip seed\n- Seed modes:\n  - Randomize\n  - Fixed\n  - Increment\n  - Decrement\n- Per-clip duration\n- Clip validation system\n- Dynamic image reference inputs\n- Up to **9 image references**\n- Optional **audio reference**\n- Shared references automatically applied to all clips\n- Native MiniMax H3 Ref2VA conditioning\n- Native ComfyUI sampling progress\n- Final video preview\n- Final video + audio export\n- Seam correction between generated clips\n- Audio seam correction \u002F declick\n- H.264, H.265 \u002F HEVC and FFV1 export\n- Persistent disk cache allowing generation to be resumed\n\n---\n# Clip Validation\n\nEach clip has a **Validated** checkbox.\n\nThe validation system allows you to progressively build a long sequence without regenerating clips that have already been accepted.\n\nWhen a clip is validated:\n\n- its generated latent remains stored in the disk cache;\n- the clip is locked and reused directly;\n- it becomes the Motion Context source for the next clip;\n- it will not be sampled again while it remains valid.\n\nA typical workflow is:\n\n    Clip 1 → Generate\n    Clip 1 → Validate\n\n    Clip 2 → Generate\n    Clip 2 → Retry if needed\n    Clip 2 → Validate\n\n    Clip 3 → Generate\n    Clip 3 → Retry if needed\n    Clip 3 → Validate\n\nThis makes it possible to build a sequence one clip at a time while preserving all previously accepted generations.\n\nValidation always forms a continuous chain from the beginning of the sequence.\n\nFor example:\n\n    Clip 1 ✅\n    Clip 2 ✅\n    Clip 3 ⬜\n    Clip 4 ⬜\n\nIf an earlier clip is unvalidated, every clip after it is automatically unvalidated as well, because each continuation depends on the previous generated clip.\n\nChanging a generation parameter also invalidates the sequence when necessary. This includes changes to:\n\n- prompt\n- seed\n- duration\n- model\n- sampling settings\n- image references\n- audio reference\n- Motion Context settings\n\nThe affected clip and all dependent clips after it must then be regenerated.\n\nIn **clip_by_clip** mode, the Extender works on the first unvalidated clip in the sequence.\n\nThe intended workflow is therefore:\n\n    Generate → Preview → Retry if needed → Validate → Continue\n\nThis allows long MiniMax H3 sequences to be created progressively without repeatedly regenerating clips that are already approved.\n\n---\n\n# Installation\n\n## ComfyUI Manager\n\nSearch for:\n\n    MiniMax H3 Extender\n\nand install it directly from ComfyUI Manager.\n\n## Manual Installation\n\nOpen a terminal in:\n\n    ComfyUI\u002Fcustom_nodes\u002F\n\nThen run:\n\n    git clone https:\u002F\u002Fgithub.com\u002Ftritant\u002FComfyUI_MiniMax_H3_Extender.git\n\nRestart ComfyUI after installation.\n\n### ℹ️ About the old Disk Join nodes\n\nThe old low-level **Motion Context Disk Join** workflow is now considered **deprecated** and is no longer actively maintained.\n\nNew workflows should use the main **MiniMax H3 Extender** node, which now handles cache management, references, trimming, validation, preview, seam correction, project Save\u002FLoad, color correction and final assembly internally.\n\n---\n\nThanks again to everyone testing the node and reporting edge cases.\n\nThe Extender is becoming much more comfortable to use for long H3 sequences.\n","这是一个为 ComfyUI 设计的 MiniMax H3 视频生成扩展节点，支持多片段视频链式生成，保持运动、视觉与音频连续性。核心功能包括 Ref2VA 条件控制、运动上下文建模、磁盘缓存、动态图像\u002F视频\u002F音频参考输入、自动 FPS 校正（适配 24fps 输入要求）、以及端到端音视频解码。适用于需长时序连贯视频生成的创作场景，如角色动作迁移、镜头运动复用、唇形同步驱动、视频风格化续写等 AI 视频工作流。",2,"2026-08-18 02:30:10","CREATED_QUERY"]