Describe the shot
Prompt tips: Keep each clip to one clear action. For image-to-video, use a first frame and, if needed, a last frame to define the shot's start and end.
Create 5–15-second 480p or 768p videos from text, first and last frames, or multimodal reference media. MiniMax H3 Max supports a dedicated reference-to-video workflow using images, video and audio.

MiniMax H3 Max is a video generation model in the H3 family. It supports text-to-video, image-to-video using a first frame, a last frame or both, and multimodal reference-to-video. The current EvoLink routes support 480p or 768p MP4 output, with durations from 5 to 15 seconds. Reference mode accepts images, videos and audio, which can help guide a shot’s visual identity, motion or timing beyond what a text prompt alone can specify.
Prompt tips: Keep each clip to one clear action. For image-to-video, use a first frame and, if needed, a last frame to define the shot's start and end.
Generation mode: Text-to-Video / Image-to-Video / Reference-to-Video / Video-to-Video. Reference image: 9. Video references are supported. Audio references are supported. Only use source material you can process, and say what each reference should guide. A reference cannot guarantee identical details.
Duration: 5s / 6s / 7s / 8s / 9s / 10s / 11s / 12s / 13s / 14s / 15s. Resolution: 768p / 480p. Aspect ratio: adaptive / 21:9 / 16:9 / 4:3 / 1:1 / 3:4 / 9:16. Check the controls for the active generation mode before submitting. Watch the entire clip for continuity, unintended objects, timing and audio where available before using it.
Generation mode
Text-to-Video / Image-to-Video / Reference-to-Video / Video-to-Video
Duration
5s / 6s / 7s / 8s / 9s / 10s / 11s / 12s / 13s / 14s / 15s
Resolution
768p / 480p
Aspect ratio
adaptive / 21:9 / 16:9 / 4:3 / 1:1 / 3:4 / 9:16
Reference image
9 · JPEG / PNG / WebP / image/heic / image/heif · 30 MB
Reference video URL
MP4 / MOV · 50 MB
Reference audio
WAV / MP3 · 15 MB
Prompt (max)
7000
MiniMax H3 Max is a video generation model in the H3 family. It supports text-to-video, image-to-video using a first frame, a last frame or both, and multimodal reference-to-video. The current EvoLink routes support 480p or 768p MP4 output, with durations from 5 to 15 seconds. Reference mode accepts images, videos and audio, which can help guide a shot’s visual identity, motion or timing beyond what a text prompt alone can specify.

Choose text, keyframes or multimodal references depending on how much control your shot needs.
Generate a 5–15-second clip from a prompt, with 21:9, 16:9, 4:3, 1:1, 3:4 or 9:16 framing.
Generate a 5–15-second clip from a prompt, with 21:9, 16:9, 4:3, 1:1, 3:4 or 9:16 framing.
Animate from a first frame, a last frame or both to define the visual start and finish.
Animate from a first frame, a last frame or both to define the visual start and finish.
Use up to 9 images, 3 videos and 3 audio clips, with a maximum of 12 reference assets per request.
Use up to 9 images, 3 videos and 3 audio clips, with a maximum of 12 reference assets per request.
Choose 480p for lower-cost exploration or 768p for a cleaner result to review.
Choose 480p for lower-cost exploration or 768p for a cleaner result to review.
Suitable for quick ad iterations, e-commerce product movement, first- and last-frame shots, and video guided by multimodal references.
MiniMax H3 Max is a video generation model in the H3 family. It supports text-to-video, image-to-video using a first frame, a last frame or both, and multimodal reference-to-video. The current EvoLink routes support 480p or 768p MP4 output, with durations from 5 to 15 seconds. Reference mode accepts images, videos and audio, which can help guide a shot’s visual identity, motion or timing beyond what a text prompt alone can specify. Ad concepts,E-commerce videos,Reference-led clips,Storyboard iterations Start with one subject, one action and a clear camera direction. Compare the result with your brief; generated motion can vary.
Prompt tips: Keep each clip to one clear action. For image-to-video, use a first frame and, if needed, a last frame to define the shot's start and end.
Prompt tips: Keep each clip to one clear action. For image-to-video, use a first frame and, if needed, a last frame to define the shot's start and end. Start with one subject, one action and a clear camera direction. Compare the result with your brief; generated motion can vary.
Generation mode: Text-to-Video / Image-to-Video / Reference-to-Video / Video-to-Video.
Generation mode: Text-to-Video / Image-to-Video / Reference-to-Video / Video-to-Video. Reference image: 9. Video references are supported. Audio references are supported. JPEG / PNG / WebP / image/heic / image/heif · 30 MB. MP4 / MOV · 50 MB. WAV / MP3 · 15 MB. Only use source material you can process, and say what each reference should guide. A reference cannot guarantee identical details.
Duration: 5s / 6s / 7s / 8s / 9s / 10s / 11s / 12s / 13s / 14s / 15s. Resolution: 768p / 480p. Aspect ratio: adaptive / 21:9 / 16:9 / 4:3 / 1:1 / 3:4 / 9:16.
Duration: 5s / 6s / 7s / 8s / 9s / 10s / 11s / 12s / 13s / 14s / 15s. Resolution: 768p / 480p. Aspect ratio: adaptive / 21:9 / 16:9 / 4:3 / 1:1 / 3:4 / 9:16. Check the controls for the active generation mode before submitting. Output quality can affect the credit estimate. Confirm the displayed total before generating.
Ad concepts,E-commerce videos,Reference-led clips,Storyboard iterations
Suitable for quick ad iterations, e-commerce product movement, first- and last-frame shots, and video guided by multimodal references. Ad concepts,E-commerce videos,Reference-led clips,Storyboard iterations Choose the framing for your destination and check how uploaded images are cropped.
Watch the entire clip for continuity, unintended objects, timing and audio where available before using it.
Watch the entire clip for continuity, unintended objects, timing and audio where available before using it. Start with one subject, one action and a clear camera direction. Compare the result with your brief; generated motion can vary. Check the controls for the active generation mode before submitting.
Suitable for quick ad iterations, e-commerce product movement, first- and last-frame shots, and video guided by multimodal references.
View example prompt 1
Suitable for quick ad iterations, e-commerce product movement, first- and last-frame shots, and video guided by multimodal references.
A high-end sneaker rotates slowly on a clean studio platform as soft daylight reveals the fabric texture.
View example prompt 2
Prompt tips: Keep each clip to one clear action. For image-to-video, use a first frame and, if needed, a last frame to define the shot's start and end.
Animate the opening frame with a smooth camera push, ending naturally on the supplied final frame.
View example prompt 3
Ad concepts,E-commerce videos,Reference-led clips,Storyboard iterations
A precise shot brief with ordered actions, camera direction, timing, lighting, and output constraints.
MiniMax H3 Max FAQ
MiniMax H3 Max is an H3-family AI video model for text-to-video, first- or last-frame image-to-video, and multimodal reference-to-video generation.
Reference mode supports up to 9 images, 3 videos and 3 audio clips, with a maximum of 12 assets in total. You must include at least one reference image or video; audio alone is not accepted.
The current routes support 480p and 768p MP4 output, with integer durations from 5 to 15 seconds.
H3 Max adds multimodal reference-to-video generation with image, video and audio inputs. H3 Max Turbo is the faster, lower-cost tier for text-to-video and first-frame image-to-video workflows.
The generator displays the current estimated cost before you submit. The cost depends on resolution and duration. Reference mode may also add charges based on reference-video length and the number of reference images.
Generation mode: Text-to-Video / Image-to-Video / Reference-to-Video / Video-to-Video. Suitable for quick ad iterations, e-commerce product movement, first- and last-frame shots, and video guided by multimodal references.
Prompt tips: Keep each clip to one clear action. For image-to-video, use a first frame and, if needed, a last frame to define the shot's start and end. Start with one subject, one action and a clear camera direction. Compare the result with your brief; generated motion can vary.
Duration: 5s / 6s / 7s / 8s / 9s / 10s / 11s / 12s / 13s / 14s / 15s. Check the controls for the active generation mode before submitting.
Resolution: 768p / 480p. Check the controls for the active generation mode before submitting. Output quality can affect the credit estimate. Confirm the displayed total before generating.
Aspect ratio: adaptive / 21:9 / 16:9 / 4:3 / 1:1 / 3:4 / 9:16. Check the controls for the active generation mode before submitting. Choose the framing for your destination and check how uploaded images are cropped.
Generation mode: Text-to-Video / Image-to-Video / Reference-to-Video / Video-to-Video. Reference image: 9. Video references are supported. Audio references are supported. JPEG, PNG, WebP, image/heic, image/heif · 30 MB. MP4, MOV · 50 MB. WAV, MP3 · 15 MB. Only use source material you can process, and say what each reference should guide. A reference cannot guarantee identical details. Watch the entire clip for continuity, unintended objects, timing and audio where available before using it.
Image
JPG, JPEG, PNG, WEBP, HEIC, HEIF · 30MB
Drop an image or click to upload.