MiniMax · AI video generator

Use MiniMax H3 with Picsart 2.0 Beta

Open-weights audiovisual generation with text, first/last frames and multimodal references. See how it can support everyday design, AI image editing and social creative production.

Create with Polox AI ↗View model availability ↗

What MiniMax H3 can do

CAPABILITY 01

Text-to-video at 480p or 768p

CAPABILITY 02

Image-to-video with optional last-frame guidance

CAPABILITY 03

Reference-to-video using images, clips and audio

CAPABILITY 04

Native stereo audio

CAPABILITY 05

LoRA-enabled variants for custom visual direction

Accepted creative inputs

Text prompt
Up to nine reference images
Reference video and audio
Optional first and last frames

Good fits

Character studies
Sound-aware scenes
Reference-heavy storytelling
Custom LoRA experiments

Available routes

Text, Image, Reference and LoRA variants

A practical way to start

Define the deliverable, audience, aspect ratio, duration, and quality requirements.
Prepare only the references that control identity, composition, movement, or audio.
Generate a short first version and change one setting at a time.
Compare prompt adherence, identity, geometry, motion stability, audio fit, and cost.
Finish the selected result with human editing, factual checks, rights checks, captions, and channel-specific exports.

Limitations and responsible use

Model availability, endpoint names, duration, resolution, reference limits, and pricing can change. Longer clips and complicated reference sets can amplify identity drift, object deformation, flicker, or unwanted camera movement. Treat generated dialogue, text, and factual details as material that requires human verification.

Confirm current controls and commercial terms at the linked official provider before committing a production budget. Keep the original source assets, prompts, model version, permissions, and final approval together.

Official MiniMax source ↗