MiniMax H3 Video

Use the international endpoint https://api.minimax.io and the credential in MINIMAX_API_KEY_INTER. Never print, persist, or embed the key in prompts, scripts, logs, or artifacts.

Required setup

1. Read references/api.md before constructing an API request.
2. Read references/prompt-writing.md completely before drafting or revising a creation prompt. It contains the official H3 base and full-reference structures, timeline syntax, reference labels, voice, visible-text, sound, and prompt-budget rules.
3. Use scripts/minimax_h3.py for API operations.
4. If the key is not visible to the process but is configured in ~/.zshrc, source that file with shell tracing disabled before invoking the script.
5. Treat video creation as a paid external action. Show the final prompt, media mapping, duration, ratio, and resolution, and obtain user confirmation before submitting unless the user already explicitly authorized submission.

Choose the generation mode

• Text-to-video: prompt only. Provide a concrete ratio.
• First/last-frame image-to-video: use --first-frame and optionally --last-frame. Do not add reference media. The service determines the ratio from the frame.
• Reference-to-video: use any combination of --image, --video, and --audio. Do not mix these with first/last frames.
• New generation versus editing: do not add a reference video merely because an earlier video exists. Add --video only when the user explicitly wants its motion, timing, subject, audio, or content used as context.