Choose a clear reference
Use an image where the defining face, character, object, or product features are unobstructed and easy to distinguish.
MiniMax H3 Reference to Video Generator
Create a new 2K video around a recognizable person, character, object, or visual target. Add reference media for identity, then use the prompt to direct the new scene and action.
ReferenceFramesFirst & Last FrameTextText to Video
ImagesAdd reference images0/9up to 9JPG/PNG/WebP · up to 9
VideosAdd reference videos0/3≤3 · 15sMP4/MOV · ≤3 · 15s total
AudioAdd reference audio0/3≤3MP3/WAV · ≤3 · needs image/video
Prompt
Upload up to 9 images, 3 videos, and 3 audio clips (12 references total). Use @Image, @Video, or @Audio in your prompt to identify each reference.
Inspiration
MiniMax H316:95s2K
MiniMax H316:95s2KGenerate MiniMax H3 Reference-to-Video · Free
Separate identity from composition
Use MiniMax H3 subject reference to keep the character recognizable
Reference-to-video separates identity guidance from scene composition. Unlike image-to-video, the uploaded subject does not have to become the opening frame. This gives the model room to place a recognizable subject in a different environment, action, or camera setup. Strong references show the defining features clearly; a concise prompt then handles what should happen in the new shot.
A four-step workflow
From a clear subject reference to a new MiniMax H3 video
Add only useful media
Each reference should have a job. Remove conflicting angles, lighting, or styling that does not serve the target shot.
Write the new scene
Describe the action, location, camera, lighting, and mood. State which identity details should remain recognizable.
Evaluate identity and motion separately
First check whether the subject reads correctly, then judge action and camera. Revise the part that failed rather than rewriting everything.
Reference prompt anatomy
Write the scene around the referenced subject
Prompt formula
Referenced subject + new action + new setting + identity constraints + camera + lighting + sound
Example prompt
Keep the referenced presenter recognizable, including facial structure and hairstyle. She walks through a quiet modern gallery while explaining an exhibit, steady waist-up tracking shot, soft skylight, natural gestures, low room ambience.
Credits for identity-led video
MiniMax H3 reference-to-video pricing for consistent subjects
Compare one-time credit packs for MiniMax H3 reference-to-video generation when recognizable characters, products, or presenters need to carry into new scenes.
Clear answers
MiniMax H3 reference-to-video questions, answered
What is reference-to-video?
Reference-to-video uses uploaded media to guide a recognizable subject or visual target while a prompt creates a new scene, action, and camera setup.
Can MiniMax H3 keep the same character consistent across videos?
MiniMax H3 reference-to-video is the best workflow for consistent character or subject identity. Use clear reference media, keep defining features consistent, and change only the scene, action, or camera instructions you want to test.
What is the difference between image-to-video and reference-to-video?
Image-to-video anchors the opening composition to an uploaded frame. Reference-to-video uses the upload as identity guidance, allowing the generated scene to differ from the source composition.
How do I improve subject consistency?
Use clear, compatible references with unobstructed defining features. Keep the first test shot simple, and state which identity details must remain recognizable.
How many references should I upload?
Use only references that add distinct, compatible information. The generator shows the current file limits; more files are not automatically better when they conflict.