How to Keep a Character Consistent with Reference Images
Use one to three reference images on Kensa's Reference to Video tool to put the same character, mascot or product into a new scene with Veo 3.1 or Seedance 2.0.
- Time to read
- 4 min read
- Model
- Veo 3.1
- Typical credits
- Typically 13 credits (8 s)
Quick Steps
Pick one to three reference images
Choose clear images of the same subject: a character, a mascot or a product. JPG, PNG or WEBP, up to 10MB each. Different angles of the same subject help the model keep it consistent.
Describe the new scene
Write the prompt as one shot: what the subject from your images does, where it happens, how the camera moves and what the light looks like. Refer to it as "the character from the reference images" instead of describing its looks again.
Choose the model and the format
Veo 3.1 makes an 8-second clip with audio for a flat 13 credits. Seedance 2.0 is billed per second (7.7 credits/s at 480p). Pick 16:9 for landscape or 9:16 for vertical video.
Generate and review
Generate, check whether the subject still looks right, and change one part of the prompt if it doesn't. If a generation fails, the credits it reserved are returned.
How to Keep a Character Consistent with Reference Images
A text prompt alone gives you a different-looking character every time. The Reference to Video tool fixes that: you upload one to three images of your subject, describe a new scene, and the model generates a clip in which the subject keeps the look from your images. This guide covers how to choose the images, how to write the prompt and what each model costs.
What you need
- One to three reference images of the same subject: a character, a mascot, a product. JPG, PNG or WEBP, up to 10MB each. Images only; video and audio files can't be used as references.
- Credits. Veo 3.1 costs a flat 13 credits per 8-second clip. Seedance 2.0 is billed per second: 7.7 credits/s at 480p and 16.9 credits/s at 720p. New accounts created with Google sign-in get 15 free credits, enough for one Veo 3.1 clip.
Reference to Video or Image to Video?
| Image to Video | Reference to Video | |
|---|---|---|
| What your image does | Becomes the first frame of the clip | Shows the model what the subject looks like |
| First frame | Your image, exactly | A new scene the model composes |
| Images | 1 (or a start and end frame) | 1 to 3 of the same subject |
| Use it when | You want this exact photo to move | You want the same subject in a different scene |
If you want a product photo to come alive as it is, use Image to Video. If you want the same mascot walking through a market, on a stage and at a desk in three different clips, use Reference to Video.
Step 1: Choose good reference images
The images are the model's only source for what the subject looks like, so their quality matters more than anything else:
- Show the whole subject clearly. No cropped heads, no hands covering the product.
- Keep the background simple. A plain or blurred background tells the model what belongs to the subject.
- Use the same subject in every image. Two different characters in your references produce a mix of both.
- Add angles, not variations. A front view plus a side view helps; three different outfits confuse the model.
- Avoid text and watermarks on the images; they tend to show up in the video.
Step 2: Describe the new scene
Write one shot, in this order: the subject and what it does, where it happens, the camera move, then the light and style. Refer to the subject through the images instead of describing it again:
The character from the reference images walks through a rainy night market holding a paper umbrella, neon signs reflecting on the wet ground. Slow tracking shot from the side, shallow depth of field, cinematic lighting.
For a product:
The sneaker from the reference image rests on a wet city street at night. A runner steps into frame and the camera pushes in on the shoe, neon reflections, cinematic lighting.
If the result drifts from your subject, remove adjectives that describe its appearance from the prompt; they compete with the images.
Step 3: Choose the model and the format
| Veo 3.1 | Seedance 2.0 | |
|---|---|---|
| Price | 13 credits per clip | 7.7 credits/s (480p), 16.9 credits/s (720p) |
| Clip length | 8 seconds, with audio | Priced per second |
| Good for | Quick, cheap tests of a scene | Longer or higher-resolution clips |
Choose 16:9 for landscape video (YouTube, product pages) or 9:16 for vertical video (TikTok, Reels, Shorts). Both models support these two formats. The exact credit cost is shown before you generate.
Step 4: Generate and review
Most clips are ready in a few minutes. Check two things: does the subject still look like your images, and does the scene match your prompt? If not, change one thing at a time (a clearer image, a simpler action, a different camera move) and generate again. If a generation fails, the credits reserved for it are returned to your balance.
What gets blocked
Images and prompts are checked before generation, and the models apply their own safety filters. These are the most common reasons a reference-image request is refused:
- Photos of real people, especially celebrities and public figures
- Brand logos, trademarks and copyrighted characters
- Content that breaks the usage policy
Original characters, illustrated mascots, your own products and stylized artwork work best.
Try it now
Open Reference to Video, upload an image of your character or product and try one of the example prompts above.