Omni-modal generation
One series covers text-to-video, first/last-frame image-to-video, and reference-driven generation, so you can move between creative workflows without changing providers.
MiniMax H3 Reference to Video supports reference images, reference videos and reference audios, with a maximum of 9 images, 3 videos and 3 audios. Audio cannot be used independently; at least one reference image or one reference video is required. First frame/last frame fields are not allowed!
The MiniMax H3 Image to Video model of this version supports video generation from the first frame, the last frame, as well as both the first and last frames.
minimax-h3-t2v is the text-to-video variant of MiniMax H3.
MiniMax's omni-modal model (Hailuo 3) turns text, images, and reference assets into native 2K video with synchronized dialogue, sound effects, and music. Try it free — new Juzi AI accounts include trial credit, no credit card required.
One series, three public model IDs
minimax-h3-t2v
Text to video. Prompt plus a required aspect ratio.
minimax-h3-i2v
Image to video. First frame required, optional last frame.
minimax-h3-r2v
Reference to video. Up to 9 images, 3 videos, 3 audio clips.
Prompt Direction
A street chef flips a wok over open flame at a night market, sparks rising in slow motion. Camera orbits from a low angle to a close-up. Sizzling oil, crowd murmur, and a distant scooter horn blend into the night ambience.
According to MiniMax's announcement, H3 is a general-purpose omni-modal generation model that understands text, images, video, and audio — and produces video with natively synchronized sound.
One series covers text-to-video, first/last-frame image-to-video, and reference-driven generation, so you can move between creative workflows without changing providers.
H3 generates native 2K footage with dialogue, sound effects, and music modeled together — including lip-sync — instead of bolting sound on afterwards.
The reference-to-video mode accepts up to 9 images, 3 videos, and 3 audio clips to lock characters, style, and voice across shots and scenes.
MiniMax H3 runs as an async task on Juzi AI. Pick the public model ID that matches your inputs, submit, and poll until the video is ready.
Use minimax-h3-t2v for text prompts, minimax-h3-i2v when you have a first frame, and minimax-h3-r2v for multi-asset references.
Write up to 7,000 characters of direction, attach images, videos, or audio within the limits of each mode, and pick an aspect ratio.
Juzi AI queues the request and returns a task ID while keeping MiniMax provider-side naming internal.
Check task status until completion, then retrieve the 2K video with its synchronized audio track.
Use the public Juzi AI model IDs directly. The adapter maps each mode to MiniMax's native video endpoint while your application depends on a single stable request contract.
# MiniMax H3 text-to-video request payload = { "model": "minimax-h3-t2v", "prompt": "Low-angle orbit around a street chef flipping a wok at a night market, sparks in slow motion, sizzling oil and crowd ambience", "duration": 10, "resolution": "2K", "aspect_ratio": "16:9" } response = requests.post( "https://japi.apipod.ai/v1/videos/generations", json=payload, headers={"Authorization": "Bearer $API_KEY"} )
The model story above follows MiniMax's official announcement. The values below mirror the current validation and adapter behavior for MiniMax H3 in Juzi AI.
Public models
3 model IDs
minimax-h3-t2v, minimax-h3-i2v, and minimax-h3-r2v.
Video length
4-15 seconds
duration is an integer between 4 and 15 seconds.
Resolution
2K
Native 2K is the supported resolution for all three modes.
Aspect ratios
21:9 to 9:16
21:9, 16:9, 4:3, 1:1, 3:4, 9:16. Required for t2v; i2v and r2v adapt to inputs.
Reference assets
Up to 15
r2v accepts up to 9 images, 3 videos, and 3 audio clips; i2v takes 1-2 frames.
Execution
Async task
Submit generation, store the task ID, and fetch status until completion.
MiniMax H3 is MiniMax's omni-modal video generation model — the Hailuo 3 generation — announced on July 31, 2026. It generates native 2K video with synchronized dialogue, sound effects, and music from text, images, or reference assets. MiniMax has also announced plans to release the model weights.
Yes. New Juzi AI accounts receive a free trial credit with no credit card required, which you can spend on MiniMax H3 or any other model on the platform. After the credit is used, billing is pay-as-you-go per generation.
Use minimax-h3-t2v for pure text-to-video, minimax-h3-i2v when you have a first frame (with an optional last frame), and minimax-h3-r2v when you need reference images, videos, or audio to lock characters and style. All three map to MiniMax's native video endpoint internally.
The current Juzi AI contract accepts durations from 4 to 15 seconds per generation, matching MiniMax's single-generation limit.
All three MiniMax H3 modes generate native 2K output. MiniMax has announced a 768P tier in closed beta; Juzi AI currently validates 2K for this series.
21:9, 16:9, 4:3, 1:1, 3:4, and 9:16. Text-to-video requires an explicit aspect ratio, while image-to-video and reference-to-video adapt to the provided inputs.
Yes. According to MiniMax's announcement, H3 models dialogue, sound effects, and music together with the video, including lip-sync. Describe the sound design in your prompt to direct it.
MiniMax's official API lists 2K generation at $0.13 per second, which MiniMax positions as less than a third of mainstream model pricing. On Juzi AI you pay per generation with unified billing — check the console for current rates and spend your free trial credit first.
Start with free trial credit, then scale MiniMax H3 text, image, and reference workflows through Juzi AI's unified generation pipeline.