Video models
MiniMax H3: Choose Your Video Workflow
A starting point for H3 video generation that separates hosted products, local workflows and model files.
MiniMax H3 is a video generation family. It serves a different task from M3.1 Flash: M3.1 Flash reasons over text and multimodal inputs, while H3 creates video.
Hosted service or local ComfyUI?
Choose a hosted MiniMax video product when you want a managed endpoint and do not want to provision local accelerators. Choose ComfyUI when you need a local node workflow and have hardware and software that match the current model requirements. Hosted model names, limits and output options can differ from local checkpoints; verify the exact service before building around a feature.
Choose by input
H3 workflows can be organized around text-to-video, image-to-video and reference-driven generation. Start from the official workflow or API documentation for the input type you have. Do not assume that a feature in a hosted Hailuo/H3 service exists in every local checkpoint.
Prompt structure
MiniMax provides separate base-mode and reference-mode prompt guides. In general, describe the subject or reference, then action, timed shots, camera and audio. Follow the selected workflow’s own input format; this is a summary of the official guidance, not an independent prompt test.
Before downloading or offering a hosted service, read the current H3 Community License. The official local-deployment guide notes territory restrictions that may require a separate license; do not assume the terms match M3.1 Flash. For local setup, follow the ComfyUI H3 guide; if a workflow asks for unfamiliar files, use the H3 model file index. For other MiniMax model families, see the model list.
Sources and version notes
Checked October 23, 2026. We label official claims, third-party measurements, our own testing and community reports separately — see the sources policy for what each label means.