Video model · Live
MiniMax H3 AI video generator
Write the shot and generate it, or attach one still image and describe how it should move. The media mode and the model stay locked to MiniMax H3 for the whole conversation.
MiniMax H3 workspace
The media mode and model are locked to MiniMax H3; the Agent chooses text-to-video or image-to-video from your instruction and the media you attach.
MiniMax H3 specifications
Every value below is read from the runtime model catalog that routes and prices the request, not maintained a second time in marketing copy.
- Output
- Video
- Supported modes
- Text-to-Video, Image-to-Video
- Duration
- 5–15s
- Resolutions
- 768P, 2K
- Aspect ratios
- adaptive, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16
- Generated audio
- Not supported
- Images per reference operation
- 1
What MiniMax H3 does here
The page exposes only the operations and settings Ideart AI can route and validate for this model today.
Text-to-video
Generate a clip from a written brief alone, with nothing attached.
Image-to-video from one frame
Attach exactly one still image and describe the motion. MiniMax H3 takes a single frame here, not a first-and-last pair.
768P or 2K
Two resolutions are offered and 2K is the default; the resolution is chosen before the clip is generated.
Five to fifteen seconds
Clip length is chosen before generating, anywhere from five up to fifteen seconds.
Seven aspect ratios
Adaptive, 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16 cover wide, square and vertical delivery from the same prompt.
Picture only
MiniMax H3 returns no audio track. Sound is a separate pass, or a job for a model that generates it.
How a MiniMax H3 clip gets made
Treat the first render as a direction check, then change one variable at a time.
Describe one shot
State the subject, the action, the setting, the camera behaviour and the pacing — one shot, not a sequence.
Decide where it starts
Say nothing about attachments for a text-to-video clip, or attach the single image the shot should open on.
Set the delivery format
Choose duration, 768P or 2K, and the aspect ratio the finished clip has to fit before you generate.
Refine in the same conversation
Keep the useful take in the thread and ask for one focused change rather than rewriting the whole brief.
Prompts to send to MiniMax H3
Open a clip to read the prompt beside it, then send that prompt straight into the locked workspace above. These clips are creative references from Ideart AI's media library, published so the prompt structure can be studied and reused; they are not presented as verified MiniMax H3 outputs.
Writing MiniMax H3 prompts
Because this model returns picture only, the prompt is best spent on what can be seen.
Say what changes
Describe what is different at the last second compared with the first, instead of naming a mood and stopping.
Name one camera move
Pick a single primary move — push-in, orbit, tracking, crane, or locked-off — and let the subject carry the rest.
Spend the detail on the frame
Light, materials, lens and framing all land in the output here; sound direction does not, so leave it out.
Where MiniMax H3 fits
Short, silent, high-resolution shots that need to be produced quickly and cut into something else.
Move an approved product still
Turn one signed-off product photograph into a short moving shot instead of booking the set again.
- One image in, one clip out
- Vertical, square or wide framing
- 2K delivery for large placements
Animate a portrait frame
Give a still portrait a small, believable movement for a vertical feed without redoing the shoot.
- Single-frame image-to-video route
- 9:16 and 3:4 for vertical feeds
- Five to fifteen second cuts
Block an establishing shot at 2K
Test a camera path and a lighting idea at the highest resolution this model offers before anyone builds the real set.
- 2K default resolution
- 21:9 and 16:9 framing
- Prompt-only, no source media
MiniMax H3 next to the Seedance models
The table is generated from the same runtime settings the generation service uses, so it stays aligned as operations and routes change.
| Specification | MiniMax H3 | Seedance 2.5 | Seedance 2.0 |
|---|---|---|---|
| Output | Video | Video | Video |
| Supported modes | Text-to-Video, Image-to-Video | Text-to-Video, Image-to-Video, Reference-to-Video, Video Edit, Video Extend | Text-to-Video, Image-to-Video, Reference-to-Video |
| Duration | 5–15s | 4–30s | 4–15s |
| Resolutions | 768P, 2K | 480p, 720p | 480p, 720p, 1080p |
| Aspect ratios | adaptive, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 | auto, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 | adaptive, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 |
| Generated audio | Not supported | Supported | Supported |
| Images per reference operation | 1 | 30 | 9 |
Read before you use an output
Generative video is probabilistic. Watch the whole clip rather than judging one frame.
- Faces, hands, small text, repeated objects and reflections can drift between frames.
- MiniMax H3 returns no audio, so anything that has to be heard is a separate step.
- Image-to-video accepts one image; a shot that needs an explicit opening and ending frame belongs on a model that takes two.
- There is no reference, edit or extend route on this model — those requests need a model that declares them.
- A clip is one shot, not a montage. Describe a sequence and the actions compete; generate the shots one at a time.
MiniMax H3 questions
Which modes can the Agent use with MiniMax H3?
Two: text-to-video, and image-to-video from a single still. A request that needs mixed references, an edit of an existing clip, or an extension routes to a model that declares those operations.
How long and how large can one clip be?
Five to fifteen seconds, at 768P or 2K, in any of the seven listed aspect ratios. 2K is the default. What each combination costs is on the pricing page, not restated here.
Does MiniMax H3 generate sound?
No. The runtime does not request audio for this model, so every result is picture only. Seedance 2.5 and Seedance 2.0 both return generated audio.
Are the clips on this page MiniMax H3 outputs?
No. They are labelled creative references from the project's media library, published so the prompts beside them can be read and reused.
Why is this page not indexed yet?
The page stays noindex until its first-party examples and publication review are complete. The route works; it simply does not advertise unfinished evidence to search engines.
Start a MiniMax H3 clip
Write one shot, choose the duration and resolution, and keep refining the useful take in your own private conversation.