1038lab /ComfyUI-MiniMax-H3-Promptor

About ComfyUI-MiniMax-H3-Promptor

These ComfyUI nodes write MiniMax H3 video prompts for you. H3_Vision reads your images, video and audio once, then H3_Promptor turns that report into a structured prompt. Cloud providers need paid API keys, and local Qwen models need a second node pack.

Best for: Prompt writers who want cinema-style MiniMax H3 prompts without re-reading their media on every run.

• Reads up to 9 images, 3 videos and 3 audios in one H3_Vision node
• Picks the task mode for you from the media you connect (T2V, I2V, L2VA and more)
• Writes a shot list paced to a length you set, from 4.0 to 15.0 seconds
• Composes prompts by hand with 8 official task-mode starters and @ tag menus
• Refines one shot, subject or section with AI, then restores the old version
• Unloads local models from VRAM after each run so video generation doesn't crash

1. 1
   Install the node pack into your ComfyUI custom_nodes folder and restart ComfyUI.
2. 2
   Open the settings panel (gear icon) and set up your LLM providers and API keys.
3. 3
   Add the H3_Vision node, connect or drag in your images, videos and audio, then pick a provider.
4. 4
   Add the H3_Promptor node, connect the VISION_CONTEXT output, write your scene description and set the duration.
5. 5
   Leave task_type on Auto, or force a mode such as L2VA, and choose English or Chinese output.
6. 6
   Optionally open H3_PromptComposer or H3_PromptEditor to refine the prompt before sampling.