Create AI video from a prompt or guide it with images, video, and audio in the same request. MiniMax H3 generates up to 15 seconds at 2K with native stereo sound and precise multimodal control.
Direct text, image, video, and audio inside one creative context, render up to 2K with sound generated alongside the picture, and keep type and detail legible all the way through.
H3 reads text, stills, clips, and audio as one brief rather than four separate inputs. Give each asset a job — take the camera move from this video, the face from this image, the vocal from this track — and it works out how they fit together.
Try it NowPicture and stereo audio come out of the same pass, so dialogue, music, room tone, and effects land with the action instead of being timed to it afterwards. Up to 15 seconds, up to 2K, six aspect ratios including 21:9.
Try it NowLabels, credits, signage, packaging, and UI copy come back readable instead of smeared into lettering-shaped noise. That is the difference between a clip you can put in front of a client and one you have to reshoot.
Try it NowShot order, timing, transitions, camera movement, and the details that must not drift all belong in the brief. H3 responds to what you rule out as reliably as to what you ask for, which is what makes a second take worth running.
Try it NowThree steps to your first multimodal AI video.
Step 01
Open Spira’s video generator and select MiniMax H3. Pick your duration, aspect ratio, and either 768p or 2K output.
Step 02
Attach reference images, a reference video, or a reference audio track, then describe the subject, action, camera movement, lighting, and pacing.
Step 03
Generate the clip, check how closely it followed your references, refine the direction if needed, then download the strongest result.
| Pricing | |||||
| Starting price | From $9/month | Pay as you go | From $110/month | From $19/month | From $18/seat/month |
| GPT Image 2 | $0.03 | Not supported | $0.55 | $0.35 | $0.2635 |
| Nano banana image | $0.01 | $0.04 | $0.55 | $0.1 | $0.047 |
| Seedance2.0 Video, 720P | $0.15/s | $0.30/s | $0.66/s | $0.59/s | $0.21/s |
| Veo3.1, 720p | $0.06/s | $0.20/s | $1.37/s | $0.07/s | $0.48/s |
| Gemini Omni Video, 720 P | $0.06/s | Not supported | Not supported | Not supported | Not supported |
| Kling 3.0 | $0.06/s | $0.084/s | $0.8/s | $0.086/s | $0.105/s |
H3 is a general-purpose multimodal model that connects text, images, video, and audio in one context. It can generate video and native stereo sound together, follow detailed creative instructions, transfer motion from video, and make reference-driven changes without splitting those jobs across separate models.
Up to 9 reference images, 3 reference videos, and 3 reference audio clips, with a combined limit of 12 items. Reference videos and audio clips must each be 2 to 15 seconds long, and audio cannot be the only reference.
Yes. One image can define the first frame, two images can define the first and last frames, and a reference video can guide motion, camera language, timing, or a targeted transformation. In multimodal reference mode, images can also define identity, products, or visual style instead of acting as frames.
Yes, through a reference generation rather than a separate edit mode. Attach the clip as a reference video and describe only what should change — a character, a costume, a background, the lighting — and leave the rest of the brief alone. The reference clip must be 2 to 15 seconds, and its length is added to what the job costs.
Clips run from 4 to 15 seconds at 768p or 2K, in 21:9, 16:9, 4:3, 1:1, 3:4, or 9:16. When you generate from references, the output can also follow the references’ own aspect ratio automatically.
Cost is based on the output resolution and the length of the clip, and reference videos add their own duration to the total. Spira shows the credit cost for your exact settings before you generate.
