Identity that holds
Faces, outfits, and palette carry across shots.
GPT Image 2 is live — sharper edits, cleaner text, up to 4K
Reference to Video
Multi-reference generation — anchor characters, wardrobe, and art direction, then generate shots that stay in the same world.
AI Model 3 credits
Prompt0/2000
Avoid copyrighted characters, logos, music and sensitive or explicit content — it may be blocked.
Resolution
Aspect Ratio
Batch generation1
Estimated cost: 3 credits
Your creation appears here
Pick a model, describe the shot, and hit Generate.
Why Reference to Video
Why creators choose Reference to Video
Reference to video is for work that needs more than a single clip: recurring characters and a steady look.
Many references
Up to 30 images on Seedance 2.5, nine on MiniMax H3 Max.
Reference stills
Seedance 2.5, Wan 3.0, and MiniMax H3 Max hold style across shots from reference images.
Reusable sets
The same references work across every shot.
Reference to video guide
How reference to video keeps characters consistent
Reference to video generates clips from several reference images, and on some models from video and audio cues as well. It exists to solve one problem: keeping the same character, outfit, and look from one shot to the next. This page explains how to use it.
How reference to video works
Instead of a single first frame, you provide a set of reference stills that define the subject and style, plus a prompt for the new shot. Seedance 2.5 accepts up to 30 reference images, Wan 3.0 up to 10, and MiniMax H3 Max up to nine.
Building a reference set
Choose references that agree with each other. Use one clear face, the outfit, and the setting, in similar lighting. If a reference contradicts another, the model has to compromise and consistency suffers. In the prompt, say what each image contributes, such as "the jacket from reference two".
Character
A clear view of the face and body.
Wardrobe and props
The items that must stay the same.
Setting and style
The place, palette, and art direction.
Making a sequence
Reuse the same reference set for every shot in a scene and change only the prompt. Generate an establishing shot, a medium shot, and a close-up with the same references, and the sequence will read as one world. Draft at low resolution first, then re-run the shots you keep.
How to use Reference to Video
How to use Reference to Video in three steps
Pick agreeing images for character, wardrobe, and setting.
00:0000:1500:30
Who uses Reference to Video
Who uses Reference to Video for image work
Short-film makers
Scenes with the same cast throughout.
Brand storytellers
A consistent spokesperson across a campaign.
Fashion brands
The same model and outfit across angles.
Series creators
Episodes that keep a recognizable look.
More models
Reference to Video and every other model in one studio
Switch engines without switching tools — GPT Image 2 plus the strongest public image and video models, side by side in the GPTImagine 2 studio.
Multi-reference editingCharacter consistencyText rendering1K–4K outputConversational editsStyle transferVideo generationBatch runs
Muse VideoNewFlagship multimodal video with native audio
Seedance 2.5HotMulti-reference Seedance with native audioSeedance 2.0Previous-gen Seedance up to 4KSeedance 2.0 FastFastFastest Seedance tier for drafts
MiniMax H3Fast text/image to video up to 2KMiniMax H3 MaxHotReference-driven video consistency
Veo 3.1Google’s cinematic drafting engine
Wan 3.0Flexible duration and adaptive framing
Kling 3.0Dramatic motion from a single prompt
GPT Image 2.5NewSharp text and precise edits to 4KGPT Image 2Versatile image generation to 4KNano Banana 2.1NewFast drafts and precise edits to 4KNano Banana ProHotHigh-fidelity image output to 4KNano Banana 2Fast drafts and conversational editsNano Banana 2 LiteFastLightweight tier for quick, low-cost draftsFAQ
Reference to Video FAQ
What is reference-to-video?
Instead of a single first frame, you supply multiple reference stills that define the subject, style, and mood of the generated clip.
How many references can I add?
Seedance 2.5 accepts up to 30 reference images per shot; MiniMax H3 Max takes up to 9.
Does it keep characters consistent?
That is its core purpose — identity, outfit, and palette carry across every generated clip.
Can I mix reference types?
Reference mode accepts still images; Muse Video is stills-first.