Seedance 2.5 Flash Reference to Video

ByteDance's Seedance 2.5 Flash reference-to-video model, a faster and lower-cost tier of Seedance 2.5 that generates up to 30 seconds of 480p or 720p video from reference images, videos, and audio, and can edit or extend a reference clip.
API docs
bytedance/seedance/2.5/flash/reference-to-video

ByteDance's Seedance 2.5 Flash reference-to-video model, a faster and lower-cost tier of Seedance 2.5 that generates up to 30 seconds of 480p or 720p video from reference images, videos, and audio, and can edit or extend a reference clip.

Try the model

Playground

Open playground
Input
The text prompt to generate a video from.
Output duration in seconds (4-30), or `auto` to let the model pick a length from the prompt and inputs. `auto` is pre-authorized at 30 seconds and billed on the actual output length. When a reference video is edited or extended, use `auto` with `adaptive` aspect ratio so the output follows the source clip.
Output aspect ratio. Omit to let the model choose from the prompt and reference inputs.
The resolution of the video to generate.
Advanced settings

PNG, JPEG, WebP, or GIF · 20 MiB maximum each · up to 10

The URLs of reference images to guide video generation. Up to 10 images. Refer to them in the prompt as @Image1, @Image2, etc.
The URLs of reference videos to guide video generation. Up to 5 MP4 or MOV files; combined duration must not exceed 30 seconds.
The URLs of reference audio files to guide video generation. Up to 5 files; combined duration must not exceed 30 seconds.
`reference` generates a new video from the references, `edit` edits a reference video (replace subjects, add or remove objects, repaint), `extend` extends a reference video forward or backward, `auto` (or omitted) lets the model infer it from the prompt. `edit` and `extend` require at least one video; `edit` forces `adaptive` aspect ratio and `auto` duration, `extend` forces `adaptive` aspect ratio. Write the prompt to match, otherwise the task can fail with a task-type mismatch.
Whether to generate synchronized audio (voice, SFX, BGM) for the video.
Random seed for reproducible results. Omit or use -1 for a random seed. Range: -1 to 4294967295.
Result
Example
  1. What would you like to do next?

    Open original

Saved example result. No new generation was requested.

Displayed run estimate: $1.2. Final charges follow the model’s billing rules below.

Seedance 2.5 Flash Reference to Video

Seedance 2.5 Flash reference-to-video is the faster, lower-cost tier of Seedance 2.5: combine up to 10 reference images, 5 reference videos, and 5 reference audio clips to lock a character, set, motion, or voice in a new 480p or 720p clip.

Fields

FieldTypeDefaultNotes
promptstring, required—Up to 50,000 characters.
imagesimage URL[] (≤10)—Reference images, Image 1, Image 2, …
videosvideo URL[] (≤5)—Reference clips, Video 1, Video 2, … Combined length up to 30 seconds.
audiosaudio URL[] (≤5)—Reference audio, Audio 1, Audio 2, … Combined length up to 30 seconds.
task_typeauto / reference / edit / extendautoReference sub-task; see below.
durationauto / 4–305Output length in seconds (4–30), or auto to let the model choose.
aspect_ratio16:9 / 9:16 / 1:1 / 4:3 / 3:4 / 21:9 / adaptive—Output frame shape. Omit to let the model choose.
resolution480p / 720p720pSets the per-second rate.
generate_audiobooleantrueNative dialogue, music, and effects. Same price on or off.
seedinteger -1–4294967295randomFix it to compare prompt changes; -1 is random.

Billing

Without reference videos you pay for output seconds: $0.12 at 480p, $0.24 at 720p. With reference videos, the input video seconds (at least 2) are added to the output length and billed at a lower rate: $0.075 at 480p, $0.15 at 720p. With duration: "auto" the quote assumes 30 seconds and the final charge uses the generated length.

Output480p720p
5s, images only$0.60$1.20
10s, images only$1.20$2.40
5s + 5s reference video$0.75$1.50
10s + 10s reference video$1.50$3.00

Failed runs are not charged.

Prompting

  • Give each reference a role: “Image 1 is the dancer, Video 1 sets the choreography, Audio 1 is the music”.
  • Numbering follows array order within each type.

Limits

  • Up to 10 images, 5 videos, and 5 audio clips per request.
  • Reference video seconds count toward billing (minimum 2, up to 30).
  • Output is 480p or 720p. For 1080p, draft mode, or more references, use bytedance/seedance/2.5/reference-to-video.

Task types

task_type selects the reference sub-task: reference (new video guided by the references), edit (edit a reference video), extend (extend a reference video forward or backward), or auto (default, inferred from the prompt). edit and extend need at least one video and use the source clip's aspect ratio; edit also follows the source clip's length and is billed on the actual output length. If the prompt contradicts an explicit task_type, the task fails.

{"prompt": "Replace the man in @Video1 with the character in @Image1.", "images": ["https://example.com/character.png"], "videos": ["https://example.com/source.mp4"], "task_type": "edit"}

Also on SandBase

Other Seedance 2.5 endpoints and tiers: bytedance/seedance/2.5/flash/text-to-video, bytedance/seedance/2.5/flash/image-to-video, bytedance/seedance/2.5/reference-to-video, bytedance/seedance/2.0/fast/reference-to-video.