1
Prepare a Reference Set
Collect 2–9 clear images of your character, product or style. Several angles of the same subject work better than many different subjects.
Grok ImagineReference to Video · Wan 3.0
Reference to Video Generator
Keep the same character, product or style across every clip
Upload up to nine reference images — a character from several angles, a product, an outfit, a location or a style — and describe a new scene. Wan 3.0 uses the references to keep the subject recognizable while your prompt decides the action, setting, camera and sound. 2 to 30 seconds at 480p, 720p or 1080p, in the browser.
How It Works
Reference to Video in 4 Steps
2
Upload the References
The generator above opens on the image tab with Wan 3.0 selected in reference mode. Add all your reference images.
3
Describe the New Scene
Write the setting, action, camera movement and sound. Refer to the subject the same way in every prompt, such as 'the woman in the red jacket'.
4
Generate and Reuse
Preview the result, then keep the same reference set for the next shot so the subject stays consistent across the whole sequence.
Reference to Video Features
Consistency Across Shots, Not Just One Clip
Up to 9 Reference Images
Combine several images of the same subject — front, side and full-body views, close-ups of details — so the model has more than one angle to work from.
Consistent Characters
Use the same reference set for every shot of a story, and the character keeps the same face, hair and outfit from clip to clip far more reliably than with text alone.
Products and Brand Assets
Upload a product from several angles to place it in new scenes for ads and listings while keeping its shape, colors and label recognizable.
Mix Subject, Outfit and Setting
Combine a character with a separate outfit, prop or location reference, then describe how they come together in the scene.
New Scene, Same Subject
Unlike image to video, the output does not have to start from one of your photos. The prompt can put the subject in a completely new place, pose and camera angle.
2–30 Seconds with Sound
Render at 480p, 720p or 1080p with audio generated in the same pass. Cost is about 2.74 credits per second at 480p, 5.48 at 720p and 10.96 at 1080p, shown before you submit.
HOW TO KEEP CHARACTERS CONSISTENT IN AI VIDEO
How to Use Reference Images with Wan 3.0
Consistency comes from the references, not from the prompt. Build one good reference set for each character or product, reuse it for every shot, and let the prompt change only what happens in the scene.
- 1
Build a character sheet
Use 3–5 images of the same subject: a clear front view, a side or three-quarter view and a full-body shot, all with the same hair, clothing and accessories.
- 2
Keep backgrounds simple
Plain or neutral backgrounds help the model separate the subject from its surroundings, so the new scene is not pulled toward the old one.
- 3
Describe the subject consistently
Name the same key traits every time — 'short silver hair, red leather jacket' — so text and references point in the same direction.
- 4
Change the scene, not the subject
Use the prompt for the setting, action, lighting and camera. Avoid prompting a different outfit or hairstyle unless you upload references for it.
- 5
Generate shot by shot
Create each shot of your story as its own generation with the same references, then edit the clips together.
Tips for Better Reference to Video Results
Quality beats quantity
Three sharp, consistent images help more than nine mixed ones. Remove references where the subject looks different.
One subject per set
Putting two different characters in one reference set can blend their features. Keep a set focused on one subject plus optional props or locations.
Make references with AI first
No photos yet? Create a character in the AI image generator, then generate a few angles of it to use as your reference set.
Check details every time
References guide the result but do not guarantee exact identity, logos or small text. Check faces and product details in each clip.
What People Make with Reference to Video
When the Same Subject Has to Appear Again
AI Short Films and Series
Keep a recurring cast recognizable across scenes and episodes.
Product Ads in New Settings
Place the same product in a kitchen, on a beach or in a studio without a new shoot.
Virtual Influencers and Mascots
Post new videos of the same character every week with a stable look.
Fashion and Outfit Showcases
Show one outfit or accessory on the same model in different scenes and poses.
Reference to Video FAQ
Common Questions About Reference to Video
What is reference to video?
Reference to video is an AI video workflow that uses one or more reference images as a guide for the subject instead of as the first frame. You upload images of a character, product or style and describe a new scene, and the model keeps the subject recognizable while creating new action and camera angles.
How many reference images can I use?
Up to nine images per generation with Wan 3.0, the default model on this page. If you switch to Grok Imagine, the limit is seven. Two to five consistent images of the same subject usually give the best balance between guidance and flexibility.
How do I keep the same character across multiple AI videos?
Use the same set of reference images for every clip, describe the character with the same key traits in each prompt, and change only the scene, action and camera. Generating one shot at a time and editing them together keeps the character more consistent than one long, complex prompt.
What is the difference between reference to video and image to video?
Image to video animates a single photo, so the video begins from that picture's framing and look. Reference to video uses images as guides for who or what appears, so the subject can be placed in a completely new scene, pose and camera angle.
Is reference to video free to try?
Wan 3.0 needs a paid plan or purchased credits — the 12 free signup credits cover the Grok models, not Wan 3.0. To try for free, switch the model to Grok Imagine: 12 credits are enough for one 6-second 480p video, and daily check-ins add 2–3 credits. Downloading video files requires a paid plan or a past credit purchase.
Does it guarantee an exact likeness?
No. References strongly guide faces, outfits and products, but AI video can still change small details such as logos, text, jewelry or hands, especially in fast motion. Check every clip and regenerate when an important detail drifts.
Can I combine a character with a product or a location?
Yes. Upload the character references together with an image of the product, outfit or location, then describe how they appear together. Keep the total at nine images or fewer (seven with Grok Imagine) and make the main subject the majority of the set.
How long are the videos and what do they cost?
2–30 seconds per generation at 480p, 720p or 1080p. Wan 3.0 costs about 2.74 credits per second at 480p, 5.48 at 720p and 10.96 at 1080p, rounded up per video — a 5-second 720p clip is 28 credits. The cost does not depend on the number of reference images and is shown before you generate. Generation usually takes a few minutes.
Is this the official Wan or Grok website?
No. This is an independent web studio, not affiliated with Alibaba or xAI. It runs Wan 3.0 and Grok Imagine through API providers, so there is nothing to install and no Alibaba Cloud account, API key or SuperGrok subscription is needed — you pay per generation with credits.
Can I use reference images of real people or brands?
Only if you have the rights and consent. You are responsible for the people, brands and assets in your references, and moderation may reject some real-person content. Paid plans include a commercial license for the videos in your account.