Turn a set of character, product, location, or style images into one directed video. Upload up to nine references, explain what each one should control, and generate with the MiniMax H3 Ref2VA workflow online.
Upload up to 9 subject or style images. Refer to them as Image 1, Image 2, and so on in your prompt.
Ready to generate
Your generated video will appear here
Reference-led creative directions
Explore the kinds of subjects, action, products, stylized worlds, and locations you can anchor with visual references, then bring your own images to the generator above.
MiniMax H3 Ref2VA is the reference-to-audio-video variant of MiniMax H3. The model can condition a new video on multimodal context; this online workflow focuses on its image-reference path, accepting up to nine images for subjects, objects, settings, and visual style. The MiniMax H3 task is available in the generator on this page.
The strongest workflow gives every image a clear job. Use complementary views and name their roles so the generated scene has better visual anchors.
Upload portraits, profiles, full-body views, and wardrobe details of the same character, then describe the movement and camera path. The references give the model more evidence for the face, silhouette, clothing, and accessories you want to retain in a dynamic shot.
Provide a clean product image, a location reference, and a material or lighting reference, then state how they belong together. Ref2VA can use those inputs as a shared visual brief for product films, campaign concepts, and designed environments.
Refer to uploaded images by their order and explain what to preserve from each one: identity, wardrobe, object design, location, composition, or motion intent. Clear roles reduce conflicts and leave the remaining prompt free to direct the new action.
Start with references that agree on the important details, then direct the new shot in plain language.
Keep MiniMax H3 selected in the model menu; this page opens its Reference Images workflow by default.
Add the clearest character, object, setting, and style references you need. Remove redundant or contradictory images.
Name what each image controls, then specify the subject action, camera movement, environment, pacing, and final beat.
Set the duration, aspect ratio, and resolution, generate the clip, then refine references or wording after reviewing continuity.
Use multiple references when a shot must preserve recognizable visual decisions while introducing new movement, framing, or context.
Anchor face, hair, costume, and accessories while directing a new performance and camera move.
Keep product form and materials recognizable while placing the object inside a designed environment.
Reference architecture, landscape, and lighting so a new action still belongs to the intended place.
Use artwork or a visual board to guide color, texture, rendering language, and atmosphere.
Combine character and scene references to explore a shot before committing to a larger production.
Bring recurring props, environments, and art direction into one coherent fantasy or branded world.
Practical answers about reference limits, prompting, output settings, and the workflow available on this page.
Yes. MiniMax identifies H3-Base-Ref2VA as its reference-to-audio-video variant for generating video and audio from text plus reference media. This page uses the available MiniMax H3 task's image-reference mode.
You can upload up to nine images in the MiniMax H3 Reference Images mode shown here. Use fewer when every reference does not add a distinct, useful detail.
Use sharp, unobstructed images with compatible identity, clothing, object design, or art direction. Multiple angles help when the shot reveals parts of a subject that one portrait cannot show.
First assign a role to each uploaded image, then describe the new shot in order: subject action, camera movement, setting, sound or dialogue, and the ending state.
No. Reference images provide stronger identity and design guidance, but they do not guarantee exact identity in every frame or across separate generations. Clear, compatible references improve the odds.
Yes. Switch to Multimodal Reference to combine image, video, and audio inputs. Reference Images remains the default workflow on this page.
The current MiniMax H3 task offers 5–15 second clips, common landscape, portrait, square, and ultrawide ratios, plus 480P, 768P, 2K, and 4K choices. The live generator shows the current credit cost before submission.
Gather the views that define your subject and scene, assign each one a role, and direct a new shot with MiniMax H3.
