AI video generation · WaveSpeed latest-model listing
Use MiniMax H3 with PicWish AI 2.0 Beta
WaveSpeed describes MiniMax H3 as an open-weights video model supporting text, image and reference inputs with native stereo audio and clips up to 15 seconds.
Try this workflow on Polox AI ↗Text/image/reference-to-video
Native stereo audio
Up to 15-second clips
Open-weights workflow exploration
Where MiniMax H3 fits
PicWish AI 2.0 Beta is written for commerce teams improving product photos, backgrounds, marketing variants and lightweight motion. Our editorial lens is to measure whether a model preserves product truth while accelerating catalog and campaign production. That makes MiniMax H3 useful when its particular controls solve a real production constraint, not merely because the model is new.
For ai video generation, begin with a narrow brief: define the subject, intended audience, composition or shot, visual references and the detail that must remain unchanged. Use a small comparison batch before increasing duration, resolution or variation count. This keeps the evaluation focused on repeatability rather than a single lucky output.
A practical production sequence
Protect the original product reference, clean the background, generate a controlled scene, compare details against the source, and animate only an approved still.
- Pass 1: test text/image/reference-to-video against the same approved brief and record what changed.
- Pass 2: test native stereo audio against the same approved brief and record what changed.
- Pass 3: test up to 15-second clips against the same approved brief and record what changed.
- Pass 4: test open-weights workflow exploration against the same approved brief and record what changed.
For image-to-video work, choose the source image for silhouette, depth, lighting and readable subject identity before asking for movement. Then describe subject action, environmental motion and camera motion separately. For image generation, hold the main visual constraint steady while varying one dimension—composition, lens language, lighting or style—at a time.
Evaluation and responsible use
Check identity and object consistency, prompt adherence, temporal stability, typography, hands and fine geometry, then confirm whether audio and motion support the intended story. Compare usable results and revision time rather than relying on headline specifications alone.
Generated product imagery should not invent features, labels or performance claims; retain the original photo for verification.
This page is an independent workflow guide. Availability and specifications should be verified at the source: WaveSpeed MiniMax H3 overview ↗.