Text to video
Describe the subject, setting and motion. Explore a scene from scratch.
MiniMax H3 / MiniMax H3 Turbo
An idea. An image. A world of references. Turn your starting point into video with MiniMax H3 on Phosor.
Browse freely. Sign in to generate. Pay per use.

Text, image and reference inputs, with native audio generation. Choose MiniMax H3 or MiniMax H3 Turbo in each mode.
| Capability | Supported |
|---|---|
| Video output | 4–15 seconds · 24 fps · MP4 |
| Resolution | 480p or 768p; landscape 16:9 outputs are 832×480 or 1344×768 |
| Aspect ratios | 16:9 · 4:3 · 1:1 · 3:4 · 9:16 |
| Image to video | Starting image required; ending image optional |
| Reference images | Up to 9 with images only; up to 5 when combined with video or audio |
| Image dimensions | Longest edge ≤2048px; aspect ratio ≤4:1 in either orientation |
| Reference videos | Up to 3; combined duration ≤15.1s; frame rate ≤24 fps |
| Reference audio | Up to 3 clips; each ≤10s; requires an image or video reference |
| Image upload formats | JPEG · PNG · WebP; published default maximum 20 MiB per image |
| Video upload formats | MP4 · WebM · MOV |
| Audio upload formats | MP3 · WAV · OGG · M4A · AAC · FLAC |
Reference counts are per type and cannot necessarily all be combined at their maximums. Files must be readable with supported codecs. Current service limits apply.
768p · 15-second output · Reference to video. Observed completion times; shorter bars mean less waiting.
6 completed runs
3 completed runs · median 411 s
7 completed runs
3 completed runs · median 421 s
Historical Phosor production tests: H3, September 2026; Turbo, October 2026. Turbo results use prewarmed models with face enhancement off. Times include request processing and transfers, not just inference. Different test batches; ranges describe these runs and may vary with inputs, loading and queue time.
01 / Start creating
Choose the input that fits your shot. Each mode is available in MiniMax H3 and MiniMax H3 Turbo.
Describe the subject, setting and motion. Explore a scene from scratch.
Give a still image its next moment. Guide the motion from your opening frame.
A prompt and a starting image. An ending image is optional.
Bring subjects, places and movement together. Tell H3 how to use each reference.
A prompt and at least one reference image or video. Audio references are optional; audio alone is not enough.
02 / For developers
Use the exact model ID from the playground. Submit a request, check its status and retrieve the finished video.
curl -X POST https://phosor.ai/api/v1/inference/submit \
-H "X-API-Key: $PHOSOR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "minimax/h3/text-to-video",
"prompt": "A hiker walks along a snowy ridge at sunrise. A slow tracking shot follows from behind.",
"duration": 5,
"resolution_tier": "480p",
"aspect_ratio": "16:9"
}'Set PHOSOR_API_KEY to your own API key. This example submits a 5-second, 480p request; generation is billed.
03 / Good to know
Yes. Open a model page and sign in to create in the playground. The same models are available through the Phosor API.
Image to video. Provide a starting image with image_url and optionally an ending image with end_image_url.
H3 is billed per output video second. Resolution and MiniMax H3 or MiniMax H3 Turbo determine the rate; reference images and videos may add charges. The pricing page lists current output and input rates.
Review current counts, upload limits and supported formats in the playground and API documentation before submitting.
MiniMax H3 / Phosor
Choose an input. Shape the shot. Create with H3.
Try MiniMax H3