Skip to main content

Seedance 2.5 character consistency with a character sheet

Seedance 2.5 keeps a character consistent by reading the reference images you upload and the role you give each one in the prompt. It takes up to 30 images, and ByteDance's guide says multi-view images are supported, so the front, side and back of a turnaround sheet fit naturally. This page shows which pack images to upload and in what order, prompt templates for one or two characters, how to add a first frame or swap a person in a video, what a clip costs, and how to fix a drifting face.

We have not made an example with Seedance 2.5 yet. This page maps the pack to ByteDance's published limits and prompt guide; the examples on this site were made with Kling 3.0.

Seedance 2.5 (official limits)

Checked 2026-10-23
  • Released by ByteDance Seed on 31 July 2026. Available in Jimeng (Dreamina) and Doubao, and through the API on BytePlus ModelArk (model ID dreamina-seedance-2-5-260628).
  • 4 to 30 seconds per generation (or -1 to let the model choose), with multi-round extension for longer stories. Output 480p, 720p or 1080p; aspect ratios 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 or adaptive.
  • Omni reference: up to 30 images, 10 videos and 10 audio clips (50 assets) in one request; videos and audio each total 30 seconds or less. Each image under 30 MB, 300 to 6000 px per side, aspect ratio 0.4 to 2.5.
  • Reference images are named in the prompt as @Image 1, @Image 2… in upload order. The guide recommends 1 to 8 subjects for image references; multi-view images are fine for up to 5 subjects.
  • Reference images or videos that contain real human faces cannot be uploaded directly; ByteDance offers trusted model outputs, preset digital characters or authorized real-person assets instead.
  • BytePlus API price examples (16:9, no input video): $0.103 per second at 480p, $0.231 at 720p and $0.569 at 1080p. Billing is per token, so input video adds cost, and failed or moderated generations are not charged.

Sources: ByteDance Seed: Introducing Seedance 2.5 · BytePlus ModelArk: Seedance 2.5 tutorial · BytePlus ModelArk: Seedance 2.5 prompt guide · BytePlus ModelArk: Pricing

From turnaround sheet to Seedance 2.5 in five steps

  1. Get a character sheet: download a pack from the library or a free character. Every pack already has the front, side and back as separate files, plus full body, outfit, prop, scene and first-frame images.
  2. Upload the images below in this order. Seedance numbers references by upload order, and its guide says to upload subjects in the order they first appear in the shot.
  3. Bind every @Image to one job in the prompt ("@Image 2 is her profile"). Do not rely on names written inside the image.
  4. Write the action as timed beats (0–5 s, 5–10 s…). Seedance 2.5 responds to whole-second timestamps; Seedance 2.0 did not.
  5. On the API, try Draft mode first: a 480p preview billed as a normal 480p video. If the face holds, turn the same draft into the final 1080p video without re-uploading the references.

Browse character turnaround sheets

Which views to upload, in which order

Eight images cover one character, the product, the room and a first frame, leaving 22 slots for more characters. Keep the order the same for every shot so @Image numbers never change.

  1. @Image 102_turnaround_views/view_front.pngFace and front of the outfit
  2. @Image 202_turnaround_views/view_side.pngProfile and hair from the side
  3. @Image 302_turnaround_views/view_back.pngHair and outfit from behind
  4. @Image 401_character_reference/full_body.pngHeight and proportions
  5. @Image 503_outfits/outfit_work.pngThe job outfit
  6. @Image 604_props/prop_role.pngThe product or tool
  7. @Image 705_scenes/scene_workspace.pngThe location
  8. @Image 808_video_starters/video_starter_landscape.pngFirst frame (optional)

The single model_sheet.png also works for one to five characters, because Seedance 2.5 supports multi-view images. With more than five characters, ByteDance says separate single-view images are more stable, so the split files are the safer default.

Why three separate views beat one good angle

A single image, even a well-chosen three-quarter view, only shows part of the character. When the character turns, the model has to invent the profile and the back, and that is where the face, hair and outfit start to drift. Upload the front, side and back as separate references and give each one a single role in the prompt, as the official guide recommends. If you also have a three-quarter view, add it as an extra reference rather than using it on its own.

Keep one face per image: crop expression grids out of the sheet, and paste the same locked appearance description into every shot's prompt.

Prompt template: one character

Prompt template

References: @Image 1, @Image 2 and @Image 3 are the front, side and back of [Name]: [skin, eyes, hair]; @Image 4 sets her height and proportions; she wears the outfit in @Image 5 and holds the [product] in @Image 6. @Image 7 is the location. @Image 8 is the first frame. Use the references only for appearance, not for pose or lighting. Style: [art style], 16:9, single continuous take. 0–5 s: [action + emotion] 5–10 s: [camera move + detail] 10–15 s: [closing shot] Keep her face, hairstyle and outfit identical throughout; only one [Name] in frame, no twins. No subtitles, no background music.

Prompt template: two characters

Prompt template

References: @Image 1 and @Image 2 are [Name A] (front and side). @Image 3 and @Image 4 are [Name B] (front and side). @Image 5 is the location. [Name A] appears first and stands on the left; [Name B] enters at 4 s and stands on the right. 0–4 s: [..] 4–10 s: [..] 10–15 s: [..] Keep each character exactly as in their own references; do not swap faces, clothes or heights between them. No subtitles.

Upload the character who appears first as @Image 1. ByteDance lists a mismatched character, where the first person on screen gets someone else's face, as a known issue and fixes it by reordering uploads to match first appearance.

Character sheet plus a first frame

Setting an image's role to first_frame turns the request into a first-frame task with a locked aspect ratio. To keep the full character references in the same request, the prompt guide's alternative is to upload the frame as a normal reference and write "@Image 8 is the first frame", as in the template above. The result follows the frame closely but not pixel for pixel.

Put your character into an existing video

Seedance 2.5 can edit a video: replace the person in a 4 to 30 second clip with the character from your references. The output keeps the input's length and aspect ratio. ByteDance recommends clips under 20 seconds and one to five reference images for edits, so use the front, side and back.

Prompt template

Replace the character in @Video 1 with the character in @Image 1, @Image 2 and @Image 3 (front, side and back of [Name]). Keep the motion, camera, timing and background of @Video 1 unchanged.

Character swap guide

Longer than 30 seconds

Generate up to 30 seconds, then extend: "Extend @Video 1… keep the character, scene and style consistent", and include the same character images again. Seedance 2.5 keeps the main characters consistent across extensions; ByteDance recommends mov output for extension and editing.

What one attempt costs on the API

Estimated from BytePlus's official per-second examples (16:9, no input video). Other aspect ratios and input videos change the token count.

  • 15-second test at 480p: about $1.55
  • 15 seconds at 720p: about $3.47
  • 30 seconds at 720p: about $6.93
  • 30 seconds at 1080p: about $17.07

Reference images are not listed as a separate charge in BytePlus's video pricing. App prices on Jimeng and Doubao use their own credits.

Face drift and other problems

The face changes when the character turns.
Upload the side and back views as their own images and name them in the prompt ("@Image 2 is her profile"). A front view alone leaves the model to invent the other angles.
The first character on screen gets the wrong face.
Reorder the uploads so each character's images come in the order the characters first appear, then renumber the prompt. This is the fix in ByteDance's prompt guide.
The face drifts by the third or fourth clip of a sequence.
Creators report this when the sheet is weak, for example a single front view. Use the same images in the same order and the same identity sentence for every clip, and extend the previous clip instead of starting fresh.
The upload is rejected.
Seedance 2.5 does not accept reference images with real human faces. Illustrated and 3D characters are the easy case; for a photorealistic person, use one of ByteDance's portrait options.
With many characters, faces mix or a twin appears.
Stay within 1 to 8 characters, give each its own images, and add "only one [Name] in frame, no twins". Above five characters, use single-view images instead of the whole sheet.

Start with a free character

Every character's full 19-image pack is free: eight starter characters need no account, the rest a free sign-up. All nine example ads use pack images only.

Browse free characters