Friends dance
Two young adult friends in green and blue share light knee bounces, smiles and low hand gestures. Generated from one complete original scene image.
MiniMax H3 · 768 × 1344

Original ideas. Real video workflows.
MiniMax H3 Max Ref usually refers to fal’s H3 Max reference-to-video variant. Use images, motion and sound references to assign distinct roles instead of asking every reference to control everything. VidLoop’s embedded creator currently runs MiniMax H3 through Metaso; it is not fal H3 Max.
Compare the actual provider and model before generating. The samples on this page are MiniMax H3 outputs, not H3 Max benchmarks.
Upload your references and keep one explicit subject, one scene and one short action. For the H3 Max endpoint, follow the official fal documentation linked below.
These original MiniMax H3 examples use different casts and one complete scene image each. They demonstrate generated booth movement; they are not a named competitor’s output or an exact music-video remake.
Two young adult friends in green and blue share light knee bounces, smiles and low hand gestures. Generated from one complete original scene image.
MiniMax H3 · 768 × 1344
An original golden retriever and black cat sway together, then the cat raises a paw. Generated from one complete original illustrated scene image.
MiniMax H3 · 768 × 1344
An original robot and pink plush alien share a small dance and friendly waves. Generated from one complete original character scene image.
MiniMax H3 · 768 × 1344

MiniMax H3 Max Ref uses assigned reference roles to guide subject, appearance and movement. Read the linked endpoint specification before uploading: a reference image is different from an opening frame. VidLoop’s Metaso MiniMax H3 is a separate model route; its displayed identity remains explicit.
Use this workflow
Prepare MiniMax H3 Max Ref inputs by assigning one role per file: subject identity, wardrobe, location or motion. Choose compatible framing and avoid conflicting faces. In VidLoop, separate portraits use Reference to Video; a complete scene uses Image to Video. Audio and video reference support depends on the actual provider endpoint.
Use this workflow
Build a MiniMax H3 Max Ref prompt from subject, reference assignment, action, camera and duration. Ask for one achievable movement in a short draft. Specify what must remain stable. A detailed reference description helps the model, but it cannot guarantee identity, physical accuracy or exact timing.
Use this workflow
Review MiniMax H3 Max Ref results against every assigned source. Inspect faces during turns, hands during gestures, wardrobe continuity and scene stability. Listen to generated audio separately. Simplify conflicting instructions before a revision; a successful creative regeneration uses another quoted amount. Keep the original inputs and model record.
Use this workflow| Starting material | Task | Important distinction |
|---|---|---|
| One complete scene | Image to video | Animate the supplied composition |
| Separate subject photos | Reference to video | Guide identity within a new scene |
| Video movement | Motion Control | Transfer movement with a compatible model |
| MiniMax H3 Max endpoint | Official fal reference API | Different provider variant from VidLoop H3 |
Upload clear permitted portraits or a complete scene; keep faces visible, remove overlays. For fictional scenes, select a text preset.
Apply your preset. Review model, mode, duration, quote and balance. MiniMax H3 starter credits fund H3 only; general credits cover other supported models.
Inspect identity, anatomy, movement, camera drift. Revise an instruction responsible for visible errors before download.
Download your reviewed video. Add licensed sound or captions, label synthetic performances, keep source files with generation settings for revisions.
Choose a subscription for regular creation, or a one-time credit pack below. Compare credit quotes by model, duration, resolution and audio. Check out securely with PayPal; monthly plans issue credits each billing period and annual plans release credits monthly.
MiniMax H3 · Member benefit
Every paid plan includes MiniMax H3 credits. A 5-second 768P video uses just 5 credits; a 5-second 2K video uses just 10.
768P
1
credit / sec
2K
2
credits / sec
These rates apply to MiniMax H3 credits, issued monthly on both monthly and annual plans.
Annual credits release monthly.
Monthly creation credits
400 general + 150 MiniMax H3
Monthly creation credits
1,200 general + 300 MiniMax H3
Monthly creation credits
2,800 general + 600 MiniMax H3
One-time general credits for all available models. Credits never expire, with no auto-renewal. Subscription benefits and gifted MiniMax H3 credits are not included.
US$7.99
One-time payment
US$21.99
One-time payment
US$44.99
One-time payment
Billed in USD; credits arrive after payment confirmation. Expiring general credits are used first. Failed tasks return eligible credits to their original balance.
Keep creating with tools that fit your source material.
Made something with Vidloop? Tell other creators how it went and what you would change.
Be the first to share your experience. There are no published creator reviews for this page yet.
We review feedback before publishing and use only the public name you choose. For help with a task or billing, contact support.
Inputs, actual capabilities and your next draft.
VidLoop currently offers MiniMax H3 via Metaso. The linked fal service provides the H3 Max reference endpoint. They should not be labeled as the same model.
Use separate portraits when each image defines one performer. Use one complete scene when you want the entire opening composition preserved. These are different inputs and use different modes.
Local planning tools are free. Generation uses the displayed model quote. MiniMax H3 starter credits fund H3 only. Compare official subscriptions and credit packs below.
Confirmed failed generations return reserved credits. A delay alone does not confirm failure. Keep the task reference and contact support if its status remains unresolved.
Use permitted photos and licensed sound. Public availability does not grant reuse rights. Label AI performances that could be mistaken for real events.
These are original VidLoop examples with actual model, input and prompt records. Code renders are labeled separately. No competitor affiliation or integration is implied.
Choose the matching input, review a short draft and keep the original source and settings with your finished video.
Make your video