[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-94428":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":14,"contributorsCount":15,"subscribersCount":15,"size":15,"stars1d":15,"stars7d":15,"stars30d":16,"stars90d":15,"forks30d":15,"starsTrendScore":15,"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":15,"starSnapshotCount":15,"syncStatus":26,"lastSyncTime":27,"discoverSource":28},94428,"ComfyUI-MiniMaxH3-Easy","nkxx188\u002FComfyUI-MiniMaxH3-Easy","nkxx188","The easiest way to use MiniMax H3. One compact workflow for T2V, I2V, first\u002Flast-frame, and reference video generation, with a unified multi-media input, powerful @ references, and inline dialogue blocks. Less wiring. More creative control.",null,"JavaScript",266,31,1,5,0,110,54.52,"MIT License",false,"main",true,[],"2026-08-24 04:01:22","# ComfyUI-MiniMaxH3-Easy\n\n[中文说明 \u002F Chinese documentation](README_CN.md)\n\n`ComfyUI-MiniMaxH3-Easy` integrates MiniMax H3 text-to-video, image-to-video,\nand reference-to-video generation into one streamlined ComfyUI workflow surface.\nThe interaction layer has been deliberately polished to make media input,\nreference selection, and prompt editing simple to understand and quick to use.\n\nThe idea is simple: keep the power of ComfyUI, while removing the repetitive\nmedia wiring and reference bookkeeping that normally make MiniMax H3 workflows\nhard to read and harder to learn.\n\n## Highlights\n\n### One `Media` input for mixed media\n\nThe main node uses one visible `Media` input for images, videos, and audio.\nMultiple links can enter the same port. Image, video, and audio order numbers\nare tracked independently, and each media type has its own wire color and\npreview style.\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"images\u002Fmixed-media-input-en.png\" alt=\"Mixed media input\" width=\"560\">\n\u003C\u002Fp>\n\nThis keeps the graph compact without losing ordering information. Drag from\n`Media` to an empty area of the canvas to quickly create a compatible media\nnode. Click the number in the middle of a virtual media wire to open the small\ndelete menu.\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"images\u002Fquick-create-node-en.png\" alt=\"Quick-create media node\" width=\"460\">\n\u003C\u002Fp>\n\n### A complete `@` reference editor\n\n`@` is available in **Reference Video** mode. Type `@` to select a connected\nimage, video, or standalone audio resource. The popup presents images first,\nvideos second, and audio last, with a preview for each item.\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"images\u002Fmention-popup-en.png\" alt=\"Reference popup\" width=\"320\">\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"images\u002Freference-editor-en.png\" alt=\"Reference editor\" width=\"720\">\n\u003C\u002Fp>\n\nReferences use **By index** by default because it is concise and easy to scan.\n**By filename** is available when the filename itself is more meaningful.\n\nThe chips are an editing interface only. When the workflow runs, the node\nautomatically converts them into the reference format recommended by MiniMax,\nincluding `\u003CPicture N>`, `\u003CVideo N>`, and `\u003CAudio N>`. Users do not need to\nmanually write or maintain those tags.\n\nA video's synchronized soundtrack is handled together with that video. A\nstandalone audio input remains an independent reference, so users only need to\nconnect the media they actually want to use.\n\n### Simple dialogue blocks\n\nType `#` in the prompt editor to insert an editable dialogue block.\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"images\u002Fdialogue-block-en.png\" alt=\"Dialogue block\" width=\"560\">\n\u003C\u002Fp>\n\n- Press `Enter` to finish the block.\n- Press `Shift+Enter` to add a line break inside it.\n- Click the block at any time to edit it again.\n\nThe block is automatically converted to MiniMax's recommended dialogue format\n`\u003Cd>...\u003C\u002Fd>` when the prompt is sent. The rest of the prompt stays ordinary\nprompt text, so users can describe the scene naturally without learning the\nunderlying markup.\n\n## Nodes and connections\n\n### MiniMax H3 Easy Loader\n\nThe all-in-one loader exposes separate choices for:\n\n- FL2VA model;\n- Ref2VA model;\n- Qwen3-VL text encoder;\n- video VAE;\n- audio VAE.\n\nOfficial and common community filename variants are recognized, including\nBF16, FP8, INT8, INT4, NVFP4, NF4, and GGUF releases.\n\nTo use only one transformer model, set the other model selector to `None`. The\nremaining model will automatically be used for text-to-video,\nI2V\u002Ffirst-last-frame, and reference-video generation. When both models are\navailable, the node prefers FL2VA for text-to-video and I2V\u002Ffirst-last-frame\ngeneration, and Ref2VA for reference-video generation. At least one of the two\ntransformer models must be selected.\n\n### MiniMax H3 Easy\n\nThis is the main generation node. It outputs:\n\n- `Model` — connect this to a model-only LoRA, Sage Attention patch, or directly\n  to the sampler;\n- `H3 Context` — connect this to **MiniMax H3 Easy Output**.\n\n### MiniMax H3 Easy Output\n\nThis node expands `H3 Context` into the standard workflow outputs:\n\n- Conditioning;\n- Latent;\n- Video VAE;\n- Audio VAE;\n- FPS.\n\nThe sampler, acceleration nodes, video\u002Faudio processing, and save nodes remain\noutside the main node so the workflow stays compatible with the rest of\nComfyUI.\n\n## Modes\n\n### I2V or First\u002FLast Frame\n\n- No media connected: text-to-video.\n- One image connected: image-to-video.\n- Two images connected: first\u002Flast-frame generation.\n- Video and audio links are not accepted in this mode.\n\n### Reference Video\n\n- Up to nine reference images, three reference videos, and three standalone\n  audio clips.\n- The `@` editor is enabled.\n- Reference order and prompt references are kept synchronized automatically.\n\n## Parameter design\n\n### Resolution and aspect ratio\n\nResolution presets follow the MiniMax H3\u002FComfyUI megapixel-style budgets:\n\n`360P`, `416P`, `480P`, `540P`, `640P`, `720P`, `768P`, `832P`, `928P`,\n`1024P`, `1080P`, and `Custom`.\n\nPresets calculate the canvas from the selected aspect ratio and align the final\ndimensions to multiples of 32. Available ratios are `1:1`, `2:3`, `3:2`, `3:4`,\n`4:3`, `9:16`, `16:9`, and `21:9`.\n\nSelecting `Custom` reveals width and height and hides the aspect-ratio control.\nCustom width and height must be multiples of 32.\n\n### Duration\n\nDuration is set in seconds from **4 to 20**. The requested duration is aligned\nto MiniMax H3's frame rules internally.\n\n### Advanced options\n\nAdvanced options are off by default. When enabled, they reveal:\n\n- first\u002Flast-frame setup in I2V or First\u002FLast Frame mode;\n- reference image size in Reference Video mode;\n- `@` display mode in Reference Video mode.\n\nReference images use a short-edge limit of **1K** or **2K**. Images below the\nlimit keep their original resolution; larger images are resized proportionally\ninstead of being forced down to the output video's resolution.\n\n## Installation and models\n\nInstall this directory as:\n\n```text\nComfyUI\u002Fcustom_nodes\u002FComfyUI-MiniMaxH3-Easy\n```\n\nPlace models in the standard folders:\n\n```text\nComfyUI\u002Fmodels\u002Fdiffusion_models\u002F\nComfyUI\u002Fmodels\u002Ftext_encoders\u002F\nComfyUI\u002Fmodels\u002Fvae\u002F\n```\n\nFor `.gguf` transformer or text-encoder files, install\n[ComfyUI-GGUF](https:\u002F\u002Fgithub.com\u002Fcity96\u002FComfyUI-GGUF) and restart ComfyUI.\nGGUF files are routed automatically to their GGUF loader; regular safetensors\nfiles continue to use native ComfyUI loading.\n\n## License and attribution\n\nThis project is released under the [MIT License](LICENSE). \n\nIf you reference, reuse, or adapt a substantial part of this project, please\ncredit the original author and mention `ComfyUI-MiniMaxH3-Easy` in your project\ndocumentation.\n\nPlease do not present the project's multi-media input design, `@` reference\neditor, dialogue-block conversion, or related implementation as entirely your\nown work.\n\n## Important notes\n\n- I2V or First\u002FLast Frame mode accepts at most two images.\n- Reference Video mode accepts at most nine images, three videos, and three\n  standalone audio clips.\n- A video's synchronized audio is paired with that video automatically and does\n  not consume a separate audio slot.\n- Image, video, and audio numbering is independent.\n- The node supports both the legacy ComfyUI canvas and Nodes 2.0.\n- Chinese browsers show Chinese parameter labels; other browsers show English\n  labels.\n- Model-only LoRA and attention\u002Facceleration patches connect after the main\n  node's `Model` output.\n","这是一个为 ComfyUI 设计的 MiniMax H3 多模态视频生成工作流插件，旨在简化文本到视频（T2V）、图像到视频（I2V）、首尾帧控制及参考视频生成等任务。核心功能包括统一的混合媒体输入端口（支持图片、视频、音频自动识别与独立序号管理）、可视化 @ 引用编辑器（自动生成 MiniMax 标准格式如 \u003CPicture N>）、以及 # 触发的对话块编辑（转为 \u003Cd>...\u003C\u002Fd> 格式）。技术上通过节点抽象减少手动连线与引用维护，提升工作流可读性与易用性。适用于 AI 视频创作、短视频原型设计、多模态提示工程验证等需要快速迭代视频生成效果的场景。",2,"2026-08-08 02:30:09","CREATED_QUERY"]