Learn how to create a MiniMax H3 AI video from text, images, first and last frames or subject references with a repeatable mode-first workflow.
Quick answer
At a glance
Workflow table
| Mode | Start with | Use it for | Prompt focus |
|---|---|---|---|
| Text to Video | A written brief | Concept clips and new scenes | Composition, action and camera |
| Image to Video | One still image | Products, portraits and key art | Motion and what remains fixed |
| First & Last Frame | Two endpoint images | Reveals and transformations | How the transition happens |
| Subject Reference | A clear subject image | Character-led variants | Identity, action and environment |
01
Begin with the material you already have. The right mode reduces the amount of prompting required and gives the model a clearer creative boundary.
02
Use a clean, high-quality image with one obvious subject. Frame the asset close to the destination aspect ratio and remove unnecessary overlays.
03
Describe subject, action, setting and one camera move. Add lighting and sound direction, then state what must remain stable.
04
Choose the format for the destination before generation. Wide ratios suit cinematic and web placements; vertical ratios suit short-form social publishing.
05
The generator presents the estimated credit cost for the selected model and settings before the task is created. Confirm the mode, assets and prompt before submission.
06
Submitted work moves through the task queue before processing and delivery. Keep the page open or return through task history when generation is enabled.
07
Review identity, product fidelity, motion, framing, text and sound. Download useful output and change one variable before another generation.
Common questions
Use text for an open concept, one image for a locked look, two frames for a controlled transition and Subject Reference for a recognizable person.
The planned H3 workflow accepts common web image formats such as JPEG, PNG and WebP where supported by the active API configuration.
Tighten the prompt, simplify the action or replace the reference. Change one variable at a time so the improvement is identifiable.
Submitted tasks are stored in task history; when live generation opens, status can be checked after returning to the account.
The estimated amount is shown before submission. Final charging and failure-return rules are displayed with the live generation release.
Related resources
Start with text or an image, direct one clear shot, then preview and refine the result in your browser.