Releases: nkxx188/ComfyUI-MiniMaxH3-Easy
Releases · nkxx188/ComfyUI-MiniMaxH3-Easy
Release list
v1.0.3
中文
主要更新
- 新增可选的提示词优化,支持 OpenAI 兼容 API 和 Gemini 原生 API。
- 新增按 MiniMax H3 模式选择的 Prompt Guide,并支持场景方案。
- 新增读取已连接媒体,并明确禁止在没有实际素材时编造媒体内容。
- 新增结构化提示词视图与原始提示词视图;原始视图直接显示
<Picture N>、<Video N>、<Audio N>和<d>...</d>。 - 改进多线
Media输入:图片、视频、音频独立编号,支持@引用,并允许断开的引用或数量不匹配时继续运行。 - 新增模型中转节点,可连接普通 ComfyUI 加载器,并支持单模型回退和 GGUF。
- 时长范围调整为 0.2–30 秒。
- 改进工作流参数保存、外部文本连接、重新优化和媒体输入交互。
- 参考生视频新增参考图尺寸控制:匹配生成分辨率、1K/1.5K/2K 像素面积和原图不缩放,并保持原始宽高比。
English
Highlights
- Added optional prompt optimization with OpenAI-compatible and Gemini Native APIs.
- Added mode-aware MiniMax H3 Prompt Guides with selectable scene guides.
- Added connected-media reading with an explicit no-media-invention rule.
- Added structured and literal raw-prompt views; raw mode shows
<Picture N>,<Video N>,<Audio N>, and<d>...</d>directly. - Improved the multi-link
Mediainput with independent media numbering,@references, and tolerant handling of disconnected or mismatched references. - Added a Model Bridge for ordinary ComfyUI loader outputs, single-model fallback, and GGUF workflows.
- Expanded duration to 0.2–30 seconds.
- Improved workflow persistence, external text connections, re-optimization, and media-input interaction.
- Added reference-image sizing controls for reference-to-video: match generation size, 1K/1.5K/2K pixel-area presets, and original-size mode with aspect-ratio preservation.
v1.0.2
中文更新说明
主要更新
- 优化从 Media 端口向左拖线快速创建加载节点的交互。
- 修复偶发弹出 ComfyUI 原生节点搜索的问题,现在可以直接创建对应的媒体加载节点。
- 优化画布交互性能,减少拖线、快速创建节点时的卡顿和掉帧。
- 修复已有媒体加载节点直接连接主节点时,连线序号和 @ 引用无法正常识别的问题。
- 模型加载器现在允许只安装 FL2VA 或 REF2VA 其中一种模型,另一种可以设置为“无”。未安装的模型对应模式将无法使用。
- 改进模型和媒体资源的识别与引用更新。
- 新增加速工作流,并补充工作流依赖插件和模型的中英文说明文档。
English Release Notes
Highlights
- Improved quick loader creation by dragging from the Media port.
- Fixed an issue where the native ComfyUI node search could appear unexpectedly.
- Improved canvas interaction performance to reduce lag and frame drops during dragging and node creation.
- Fixed media numbering and @ references when existing loader nodes are connected directly to the main node.
- The model loader now allows users to install only one of FL2VA or REF2VA. The other model can be set to “None”, but its corresponding mode will be unavailable.
- Improved model and media resource detection and reference updates.
- Added an acceleration workflow and bilingual workflow dependency documentation.