Validated reference images
Use private JPEG, PNG, or WebP character and style references.
Reference-driven anime direction
Build a coherent illustrated character world across multiple music-driven shots instead of regenerating an unrelated anime image every scene.
Character reference · style reference · shot continuity · original audio
MiniMaxMusic Studio is an independent third-party service and is not the official MiniMax website.
Choose a generated song or upload an authorized MP3/WAV.
Select a 15- or 30-second highlight, output ratio, and visual direction.
Generate and edit a versioned storyboard before any H3 job is submitted.
Attach validated reference images and approve an immutable server quote.
Render each H3 shot independently, review it privately, and rerender only what needs work.
Compose a final H.264/AAC MP4 with the original song as the only master audio.
Anime-style generation becomes more coherent when the project defines character silhouette, costume, line treatment, palette, environment, and camera language before creating shots.
The Studio stores validated character and style images as private reference assets, applies selected ids to the current Storyboard version, and invalidates approval when those references change. Every H3 attempt therefore has an auditable creative context.
Use private JPEG, PNG, or WebP character and style references.
Regenerate a new plan without overwriting historical shot versions.
Preserve character, world, color script, and directional motion across the hook.
Direct the final shot toward a memorable key visual.
Lock good shots and rerender only the inconsistent ones.
Create an expressive character-led vertical chorus clip.
Use one consistent illustrated identity and kinetic environments.
Build coherent scale and world rules with anime references.
Reserve negative space and add accurate text during final composition.
No generative workflow can guarantee pixel-identical continuity. Validated references, stable prompts, a consistent storyboard, per-shot review, and selective rerendering materially improve control.
You select a music range, generate a versioned storyboard, edit every shot, apply validated visual references, approve the storyboard, and request a server-side quote. No H3 job starts merely because a song was uploaded.
Successful private H3 shots are normalized and concatenated by a dedicated FFmpeg compositor. H3 audio is ignored; the selected portion of the original song becomes the only master soundtrack.
The first production formats are a 15-second teaser and a 30-second social clip in 16:9, 9:16, or 1:1. The project architecture is designed to support longer formats later without treating them as one opaque generation.