MiniMax H3 Ref2V (Ref2VA)
MiniMax H3 Ref2V:
References In.
Audio‑Video Out.
Ref2V is how most people search for MiniMax H3's reference-to-audio-video mode: you upload reference images, optional clips, and optional voice audio, and H3 performs a new scene with them as the cast — and generates the stereo soundtrack in the same pass. No subscription, pay per clip with credits.
Model Options
Choose the workflow that fits your scene
Start with a prompt, add references when needed, and pick the generation path that matches the output.
MiniMax H3 Ref2VA
The reference-to-audio-video mode of MiniMax H3, exposed on Moosky as an omni-reference composer. Feed it a reference sheet — face images, wardrobe or product shots, motion clips, voice takes — label each reference in the prompt, and H3 keeps the identity, the styling, and the voice while performing your new scene with synchronized stereo sound.
Other reference-to-video models
Ref2V is MiniMax H3's take on R2V. When a shot needs heavier motion or different framing, your same credits run Wan 3.0 R2V or Seedance 2.0 R2V — the full lineup and how to pick between them lives on the reference-to-video hub.
Why use it on Moosky AI
Powerful tools. Seamless experience.
Video and audio in one pass
Ref2V's defining trick: the A in Ref2VA. Dialogue, ambience, and score cues render synchronized with the picture — no separate TTS or dubbing step.
Omni reference sheet
Up to 9 images, 3 video clips, and 3 standalone audio files per job (12 files max) — identity, wardrobe, style, motion, and voice all in one bundle.
Character-consistent scenes
The same face, outfit, and product across takes. H3's default strength is talking heads and dialogue scenes that have to stay recognizably the same person.
Voice carried by reference
Attach a voice clip as a reference and the performed line keeps that timbre — across languages, with lip sync baked in.
11-language dialogue
Stable speech in Arabic, Chinese, English, French, German, Italian, Japanese, Korean, Portuguese, Russian, and Spanish.
Pay per clip
H3 Ref2V runs on Moosky credits (100 per $1). The exact cost of your references, length, and resolution is quoted in the composer before you generate.
How It Works
Three simple steps to finished output
Build the reference sheet
Upload face and wardrobe images, optional motion clips, and an optional voice take — ordered, because the labels follow the order.
Direct the scene
Prompt the new shot: setting, action, camera, and the line to be spoken. Reference each file by label (Image 1, Audio 1) so its job is explicit.
Render video + audio
Pick 4-15 seconds at 768P or 2K and generate — one pass returns the clip with its synchronized stereo track.
Why character-consistency creators are switching to Ref2V
Plain text-to-video cannot hold an identity across takes: the hero from shot one is a slightly different person in shot two. Ref2V fixes this by conditioning every render on a reference sheet instead of a description — the face from several angles, the wardrobe flat-lay, a motion clip, a ten-second voice take. The identity lives in your files, not your phrasing, so take nine looks like take one.
The workflow maps onto storyboarding: one reference sheet per role, H3 assigns each file a role from your labels (Image 1 the face, Audio 1 the voice), and you walk the panels changing only setting, blocking, and dialogue. Stereo audio renders in the same pass — lip-synced speech in eleven languages, ambience, score cues — so there is no dubbing round-trip.
When a take drifts, you add one more reference of the detail and regenerate the panel; short 4-to-15-second 768P renders keep retries cheap, escalating a hero shot to 2K when it earns it. One beautiful shot? Text-to-video is fine. A character who must exist in ten shots and sound like themselves in all of them? That is a Ref2V project — on credits, no subscription.
Credit-based
Simple, honest pricing
No subscriptions. No hidden fees. Pay only for the credits you use.
More from Moosky AI
Explore our complete suite of AI tools.
Frequently Asked Questions
What does Ref2V mean on MiniMax H3?
Ref2V is shorthand for reference-to-video: you give the model references instead of just a prompt, and it generates a clip conditioned on them. On MiniMax H3 the mode is formally Ref2VA — reference to audio-video — because the same pass also produces synchronized stereo sound (dialogue, ambience, music cues).
What can I feed the Ref2V composer?
Up to 9 reference images, 3 reference videos, and 3 standalone audio clips, 12 files total. Each video or audio clip should run 2-15 seconds. Standalone audio must accompany at least one image or video — audio can't be the only input.
How do I control what each reference does?
Order them intentionally and label them in the prompt: Image 1 is the face, Image 2 the wardrobe, Video 1 the camera move, Audio 1 the voice. H3 reads those labels and assigns each reference a role — identity, style, motion, or sound.
How is H3 Ref2V different from image-to-video?
Image-to-video animates one still — the frame itself becomes the shot. Ref2V conditions on identity instead: place the same person or product in new scenes, angles, and actions and they stay consistent take after take, and with H3 they speak with a consistent voice too.
How much does MiniMax H3 Ref2V cost on Moosky?
Generations are billed in credits (100 credits per $1). H3 pricing scales with output length, resolution (768P vs 2K), and input materials — the first 5 images are free, extra images and reference-video duration are billed. The composer shows the exact quote before you run, and failed runs are refunded.
Should I use H3 Ref2V or another R2V model?
Choose H3 when the scene depends on dialogue, a talking head, or a specific voice — that is its home turf. When a shot is mostly kinetic motion or complex camera work, Wan 3.0 R2V or Seedance 2.0 R2V may fit better. The reference-to-video hub compares them side by side.
Can I download the clips Ref2V generates?
Yes. Every finished clip lands in your Moosky queue as a downloadable MP4 at the 768P or 2K resolution you chose, with no separate export step and no watermark on the output. Failed runs are refunded, so you only pay credits for clips you keep.
Disclosure
Moosky AI is an independent platform provider and reseller of AI model access, including access to MiniMax H3. Moosky AI is not officially affiliated with, sponsored by, or endorsed by MiniMax. This page and its examples may include AI-assisted and synthetic content.