Character identity
Keep faces, hair, body proportions, and distinctive features recognizable across new scenes.
Reference to Video AI
Reference to Video AI Generator
Upload reference images, videos, or audio and generate smooth, coherent videos where your character speaks and acts with perfect facial consistency and natural motion. No more reshoots!
Reference to Video
Choose a model, add compatible image, video, or audio references, then mention them in your prompt.
Reference assets
9 images · 3 videos · 3 audio files
0/4000
Model
Format
Sample Video
Choose a template to load its prompt and reference direction.
08 reference to video use cases
Stories need continuity. Not lucky generations.
See how reference to video keeps identity, wardrobe, products, and visual worlds recognizable across drama, UGC, anime, fashion, music, and game scenes.
More consistency workflows
Explore fashion, product, performer, and mascot continuity in the original eight reference to video cases.
Reference-first generation
What is reference to video?
Reference to video is an AI generation workflow that uses one or more source assets as creative evidence. Instead of asking a text prompt to remember a face, product, outfit, movement, camera style, and sound at once, reference to video lets each source define a specific part of the new shot.
A strong reference to video result does not reproduce a source frame for frame. It preserves the details that create continuity while allowing the setting, action, framing, and story beat to change. This makes reference to video useful for episodic creators, AI influencers, product campaigns, anime characters, fashion sequences, music videos, and branded worlds.
The difference is control: references stay active throughout the shot, defining what must remain recognizable and what may change.
What can you control with references?
Each source should own one production decision. Clear roles make the result easier to direct, review, and refine.
Product appearance
Preserve product shape, color, materials, packaging, and label placement in new shots.
Motion and performance
Use a short video to guide gestures, timing, body movement, and performance energy.
Camera movement
Reference a pan, orbit, push-in, handheld move, or other camera behavior that is hard to describe.
Scene and style
Carry a location, lighting language, palette, or visual world into a newly composed shot.
Audio and timing
Guide dialogue, voice, music, beats, and action timing on models that accept audio references.
Two multimodal reference workflows
Start with the smallest useful set of sources. These two workflows separate appearance, motion, and audio so every reference has one clear job.
Seedance 2.0 Fast720P example format
Image + video reference
Use the image to anchor a person, product, outfit, or visual style. Use the video to guide movement, performance timing, framing, and camera direction. The planned example will be delivered at 720P.
@Image1 · appearance@Video1 · motion
@Image1 defines the subject and appearance. Follow the action and camera movement from @Video1 while creating a new scene.
MiniMax H3 Reference768P example format
Image + video + audio reference
Add audio when voice, music, rhythm, or synchronization is part of the brief. The image owns appearance, the video owns performance, and the audio owns sound and timing. The planned example will use H3's 768P output.
@Image1 · appearance@Video1 · performance@Audio1 · timing
Keep the subject from @Image1, follow the performance in @Video1, and synchronize the reveal to @Audio1.
Fix common reference-to-video consistency problems
Reference to video consistency failures often come from conflicting sources, not a lack of prompt adjectives. Simplify the evidence, state what must stay fixed, and test one creative change at a time.
The face changes between shots
For reference to video identity control, use one clean front or three-quarter image, remove conflicting portraits, and describe the person once. Add a second angle only when it contributes new information.
Two characters swap clothes or roles
Assign each person a separate reference token and repeat the mapping in the action sentence. Keep wardrobe descriptions attached to the correct name.
The product is redesigned
Give reference to video a clean product hero image plus an angle showing geometry, branding, and material. Ask it to preserve shape, label placement, color, and scale.
Motion or audio is ignored
Trim source media to the exact movement or beat. In a reference to video prompt, state one job for every source and remove redundant assets.
The outfit changes between scenes
Use a full-body reference with the complete outfit visible. Name the garment, material, color, and accessories as fixed details, and avoid conflicting wardrobe images.
The camera does not follow the prompt
Use a short video showing the exact pan, orbit, push-in, or handheld movement. Assign that clip only to camera behavior instead of asking it to define the subject as well.
Text, image, or reference to video?
Choose by what must remain under control. Text creates from a description, image to video animates an existing frame, and reference to video uses several sources as constraints for a newly composed shot.
DecisionText to videoImage to videoReference to video
Starting point
Text to videoA written description
Image to videoOne image treated as the opening frame
Reference to videoSeveral assets with separate creative jobs
Best control
Text to videoConcept, subject, action, and mood
Image to videoComposition and appearance near the first frame
Reference to videoIdentity, product, motion, camera, style, and audio
Best use
Text to videoOpen-ended ideation and new scenes
Image to videoA single short animation from one still image
Reference to videoStories, campaigns, UGC series, character IP, and performance-led shots
Prompt method
Text to videoDescribe everything the model must create
Image to videoDescribe movement away from the uploaded frame
Reference to videoAssign roles to @Image, @Video, and @Audio sources
How reference to video works
Build a controlled reference to video brief by assigning every source a specific production job before generating a new scene.
Upload clean references
A reference to video project starts with clear images for identity, products, outfits, or style. Add short video and audio references when they contribute motion, camera, voice, or timing.
Assign every reference a role
Tell the model what @Image1, @Video1, and @Audio1 control, then describe the action, setting, camera, and details that must not change.
Generate and review consistency
Review the reference to video result for faces, wardrobe, product geometry, roles, motion, and audio timing. Change one instruction at a time.
Choose the right reference to video model
The best reference to video model depends on your sources and the detail that matters most. The composer validates supported asset types and limits before it sends a job.
Seedance
The complete Seedance lineup
Choose Seedance 2.0 Mini, 2.0 Fast, or 2.5. Mini supports 480P/720P, Fast supports 720P, and 2.5 supports 480P/720P with generation up to 30 seconds.
MiniMax H3
High-resolution multimodal reference
MiniMax H3 supports 768P and 2K with image, direct-video, and audio references. Images and audio add no extra charge; direct videos are billed by total input plus output duration.
Gemini Omni
Fast Gemini video generation
Omni Flash 1.1 supports 720P, 1080P, and 4K output at 8 or 10 seconds. It accepts up to seven weighted reference slots, with one video using two slots.
Learn the reference to video workflow
How to keep the same character consistent
Prepare identity references, name the details that cannot change, and review face, outfit, and proportions across every shot.
Read the character guideHow image, video, and audio references work together
Give appearance to images, movement to video, timing to audio, and connect every source with explicit reference tokens.
Read the multimodal guideReference to video terms and concepts
Review identity consistency, scene continuity, image to video, multimodal prompting, and related production terms.
Open the glossaryPRICING
Choose Your Video Generation Plan
Lite
$12/month
$144 billed yearly
1,500 credits / month
Up to 37 videos
- 720P, 1080P & 4K output
- Up to 30s per video, depending on settings
- Image, video & audio references
Credits issued monthly. No rollover.
Most Popular
Pro
$29/month
$348 billed yearly
4,000 credits / month
Up to 100 videos
- 720P, 1080P & 4K output
- Up to 30s per video, depending on settings
- Image, video & audio references
Credits issued monthly. No rollover.
Ultra
$89/month
$1,068 billed yearly
13,000 credits / month
Up to 325 videos
- 720P, 1080P & 4K output
- Up to 30s per video, depending on settings
- Image, video & audio references
Credits issued monthly. No rollover.
Compare plans
No video reference
| Model / Resolution | Lite1,500 credits | Pro4,000 credits | Ultra13,000 credits |
|---|---|---|---|
| Seedance 2.0 Mini · 480P40 credits / 5s | Up to 37 videos | Up to 100 videos | Up to 325 videos |
| Seedance 2.0 Mini · 720P80 credits / 5s | Up to 18 videos | Up to 50 videos | Up to 162 videos |
| Seedance 2.0 Fast · 720P180 credits / 5s | Up to 8 videos | Up to 22 videos | Up to 72 videos |
| Seedance 2.5 · 480P190 credits / 5s | Up to 7 videos | Up to 21 videos | Up to 68 videos |
| Seedance 2.5 · 720P380 credits / 5s | Up to 3 videos | Up to 10 videos | Up to 34 videos |
| MiniMax H3 · 768P110 credits / 5s | Up to 13 videos | Up to 36 videos | Up to 118 videos |
| MiniMax H3 · 2K160 credits / 5s | Up to 9 videos | Up to 25 videos | Up to 81 videos |
| Omni Flash 1.1 · 720P120 credits / 8s | Up to 12 videos | Up to 33 videos | Up to 108 videos |
| Omni Flash 1.1 · 1080P180 credits / 8s | Up to 8 videos | Up to 22 videos | Up to 72 videos |
| Omni Flash 1.1 · 4K350 credits / 8s | Up to 4 videos | Up to 11 videos | Up to 37 videos |
How video generation credits are calculated
Total credits = base generation credits + video-reference credits
Without a native video reference
Base credits = output rate × output seconds
Text prompts, reference images, and reference audio are charged only for the generated duration.
With native video references
Total credits = base generation credits + video-reference credits
Seedance and H3 charge video references per second. Reference durations are added together before rounding up; images and audio are free. Omni uses a fixed video-reference add-on, not a per-second fee.
Seedance and MiniMax H3 rates
| Model | Resolution | 5s output · no video reference | 5s reference + 5s output |
|---|---|---|---|
| Seedance 2.0 Mini | 480P | 40 | 40 + 5 × 2 = 50 |
| Seedance 2.0 Mini | 720P | 80 | 80 + 5 × 4 = 100 |
| Seedance 2.0 Fast | 720P | 180 | 180 + 5 × 8 = 220 |
| Seedance 2.5 | 480P | 190 | 190 + 5 × 10 = 240 |
| Seedance 2.5 | 720P | 380 | 380 + 5 × 20 = 480 |
| MiniMax H3 | 768P | 110 | 110 + 5 × 22 = 220 |
| MiniMax H3 | 2K | 160 | 160 + 5 × 32 = 320 |
Omni Flash 1.1 fixed credits
Reference duration does not increase the charge| Resolution | 8s output · no video reference | 3s reference + 8s output |
|---|---|---|
| 720P | 120 | 120 + 50 = 170 |
| 1080P | 180 | 180 + 60 = 240 |
| 4K | 350 | 350 + 70 = 420 |
Seedance and H3 charge video references per second. Reference durations are added together before rounding up; images and audio are free. Omni uses a fixed video-reference add-on, not a per-second fee.
Reference to video FAQs
Reference to Video
Create a new shot from the references you already have
Use images for appearance, video for motion, and audio for sound and timing. Give every source a clear role, then direct what the next shot should change.