Skip to main content

Uncensored — This model declares no content filter. SpicyAPI forwards the request without adding a platform content scan; legal and consent duties remain with the customer.Browse every model at this filtering level
bytedance/seedance-2.0-mini/reference-to-video

Seedance 2.0 Mini reference-to-video from ByteDance builds one shot from your own material: up to 9 images, 3 video clips and 3 audio tracks, with no first frame. Output is 4 to 15 seconds at 480p or 720p, with sound in the same pass.

From $0.011/s
Run model

Workbench

Input

Reference to video

Reordering updates prompt references. Removing media also removes its references; review your prompt.

Required, and here it also does the casting. References are cited in the order you send them, so name which image is which character, what each clip contributes and what the audio is for. Briefs are not reviewed on our side; where Seedance's own layer refuses the wording, a first frame on image-to-video is the route that still gets there. at least 1 characters
Provide at least one of theseRequirement met
2/9

Drop files or media links, paste, or browse · Up to 9 files · Max 10 MB · JPG · PNG · WEBP · GIF

  • @Image1
  • @Image2
2 / 9 files
Up to 9 images you are authorised to use, holding a character, a product, a style or a set. Two or three angles of the same subject carry more than a stack of near-identical frames. Every request needs at least one reference image or reference clip; audio on its own is refused. This endpoint has no first- or last-frame field: when the opening picture has to be exact, use image-to-video instead. Each side must be between 300 and 6000 pixels; a smaller plate is rejected by the input check before you are charged. at most 9 items
affects price0/3

Drop a video here, or browse · Up to 3 files · Max 90 MB · MP4 · WEBM

0 / 3 files
Up to 3 clips supplying motion, pacing or camera language. Each clip should run 2 to 15 seconds and all of them no more than 15 seconds together; keep every file under 50 MB and at 1080p or below, or the request is rejected. Their seconds are billed on top of the output length, so trim them to the part you actually want referenced. at most 3 items
0/3

Drop an audio file here, or browse · Up to 3 files · Max 90 MB · MP3 · WAV

0 / 3 files
Up to 3 audio files, 2 to 15 seconds each and 15 seconds together, each under 15 MB, giving the shot a sonic anchor — delivery, tempo, a musical phrase — that motion and cut points align to. Audio cannot be the only reference: send at least one image or clip with it. It is an input, not the output track: whether the finished clip carries sound is still decided by `generate_audio`. at most 3 items
Output framing, adaptive by default: the clip takes its shape from the primary reference. When the references disagree or the placement is fixed, pick one of the six ratios from 21:9 down to 9:16 and say in the prompt how the subjects should sit inside it. No effect on what you are billed.
affects price
Output resolution: 480p, or 720p by default. It selects a rate tier and will not sharpen low-resolution references. Check that the references combine at a low tier, then raise it for the final take.
Durationaffects price
Length of the finished clip in whole seconds, 4 to 15, default 5; -1 lets the model choose, with the estimate assuming 15 seconds. Together with the seconds of your reference clips it is the billed quantity. The more references a shot has to reconcile, the less 4 seconds can show them all — but confirm the combination short before paying for 15. no lower than 4 · no higher than 15
Generate audioaffects price
Whether a native soundtrack is rendered in sync with the picture. On by default; switching it off does not lower the rate in this tier. Turn it off for silent loops and for material you will score in post.
Safety checkerOff

SpicyAPI does not run a safety checker on this model. Your prompt and media go to the model as written, and the model itself decides what comes back.

Checking your account before enabling submission and pricing.

Output

Ready for a new run
Example result
Extend
Video editing
Video upscaling

Related models and next tasks

Missing the model you need?

Tell us what to add and where you would use it. Requests go to the people who decide what gets onboarded next.

Cast a shot from your own images, clips and audio

The Seedance 2.0 Mini reference-to-video API builds one shot out of material you supply: up to 9 reference images, 3 clips and 3 audio tracks in a single call, with prompt required and at least one image or clip among the references. Output runs 4 to 15 seconds at 480p or 720p, and the seconds of any reference clips are billed on top of the output length.

Open the playground
Cast a shot from your own images, clips and audio
Required input
prompt, with at least one reference image or clip
Reference caps
9 images, 3 clips, 3 audio tracks per call
Reference clips
counted from 2 seconds each, 15 seconds combined, added to the bill
Duration
4 to 15 whole seconds, default 5, or -1 quoted as 15
Resolution
480p or 720p, 720p by default
Not accepted
image_url, which belongs to image-to-video
Billing
Reference clips add seconds to the quantitySeedance 2.0 Mini reference-to-video counts every reference video for at least 2 seconds, caps their combined length at 15 seconds, and adds that figure to the output seconds when the call is priced. Reference images and audio tracks leave the quantity alone. Trimming clips before upload is a direct saving once the same call runs many times.
Positioning
Testing pairings takes more than one callFinding out which combination of references reads correctly means composing it several times, which is a run rather than a single render. ByteDance names A/B testing and rapid iteration as the stages the 720p models are priced for, and this tier has the lowest rate per second of the three published for Seedance 2.0.
No repair
A near miss is composed again from scratchSeedance 2.0 Mini publishes no edit endpoint, so a composite that gets one detail wrong is resubmitted as a whole call. At volume that makes the cheap 480p pass, where the role of each reference gets settled, the step that decides what the rest of the run costs.

What the Seedance 2.0 Mini API does well

A partial reference set is a normal call

A partial reference set is a normal call

reference_image_urls, reference_video_urls and reference_audio_urls are each optional on their own, as long as one image or one clip is present. A composition that needs two stills sends two stills, and the unused arrays stay out of the body entirely. The ceilings are 9 images, 3 clips and 3 audio tracks in a single request.

Extra stills do not move the billed quantity

Extra stills do not move the billed quantity

Only the seconds of reference clips join the quantity on Seedance 2.0 Mini reference-to-video. Adding a fourth still or a second audio track changes what the model has to reconcile, not what the call is quoted at, which makes reference images the inexpensive way to give a shot more to work from.

  • One approved asset set, many compositions

    A single character still can anchor every shot in a series while the location and motion references change around it. The request body stays a template with URLs substituted per call, so scaling up is a matter of bookkeeping rather than of rewriting anything in the schema.

  • Smart duration across a set of compositions

    duration_seconds accepts -1 here as well, and the model then settles a length in the 4 to 15 second range. The quote assumes 15 output seconds plus the reference-clip seconds until the result lands, so a fixed number keeps the estimates for a large run much closer to what it actually costs.

  • The same ceiling as the rest of the family

    Adding references narrows nothing else: resolution still takes 480p or 720p, duration still runs 4 to 15 seconds, and the task still has 1200 seconds to finish. A composition carrying nine references is priced on the same two fields as a bare prompt call, plus whatever its reference clips contribute.

  • Every call stands alone in a run

    Nothing is shared between Seedance 2.0 Mini reference-to-video calls. The same reference URL sent in twenty requests produces twenty independent renders, each priced on its own inputs, so a run has no warm-up period and nothing to reset when one composition has to be sent again.

Who Seedance 2.0 Mini is for

  • One character reference across many shots

    Series with a fixed cast

    One character reference across many shots

    The same entry in reference_image_urls anchors identity while the location and motion references vary per composition. Because the prompt does the casting, the sentence that assigns roles is written once and reused, and each shot in the series is an ordinary call against the same template.

  • Recombine approved material into many clips

    Asset libraries

    Recombine approved material into many clips

    A small set of cleared images, clips and audio can be paired in different combinations across a run, which turns a library that took real effort to license into a much larger set of finished shots. Each pairing is a separate call, so the rate per second decides how many pairings are worth trying.

  • Find the combination that reads, cheaply

    Teams testing reference pairings

    Find the combination that reads, cheaply

    Which reference governs which detail is learned by composing, and that means several calls before one is right. Running those at 480p on the lowest-priced Seedance 2.0 tier keeps the discovery cost low enough that three or four attempts per shot is a normal step.

  • Settle the roles at 480p, then commit

    Previsualisation

    Settle the roles at 480p, then commit

    A 480p pass across a handful of reference pairings answers which combination the prompt actually controls. Once the wording holds, the same composition is rerun at 720p, and the reference set that survived the pass becomes the template the rest of the run is built on.

How to call the Seedance 2.0 Mini API

  1. Write the casting once, substitute the URLs

    The prompt is what sorts the references, so write the division of labour once, such as identity from the first image and setting from the second, then reuse that sentence with different asset URLs for every item in the run.

  2. Trim reference clips before uploading

    Each reference video counts for at least 2 seconds, their combined length is capped at 15, and those seconds join the billed quantity. Cutting a 30 second source down to the part worth citing changes both the quote and what the model has to read.

  3. Group a run by asset set, not by wording

    Calls that share references also share their failure modes. Grouping a run that way means a replaced asset touches one known block of items, and the rest of the run needs no attention when something has to be swapped out.

  4. Prove the roles on two or three items

    Which reference governs which detail is learned by composing, not by reading the schema, and at 480p three test compositions cost a fraction of a full run. The wording that survives them is usually the wording for the rest of the batch.

  5. Refresh the estimate after touching references

    The quote follows duration_seconds, resolution and the length of the reference clips, so it moves when a clip is swapped for a longer one. Adding or removing reference images and audio tracks leaves the number where it was.

Seedance 2.0 Mini worked examples

  • One character across three settings

    The same character still in reference_image_urls, three different setting references, sent as three separate calls from one prompt template.

    Place the person from the first reference image into the setting shown in the second reference image, walking at a relaxed pace, late afternoon light

  • One audio reference against three stills

    The same reference_audio_urls entry tested with three different image references, to find which visual pairing sits best against the track.

    Match the quiet ambience of the reference audio while the scene from the first reference image plays out at a calm, unhurried pace

  • A 480p pass to settle the reference roles

    resolution at 480p and duration_seconds at 5, run across a few pairings to see which reference the prompt is actually controlling.

    Place the figure from the first reference image at the market stall shown in the second reference image, browsing unhurried among the produce

Seedance 2.0 Mini API questions

Write the casting sentence before you collect the assets, because references are read in the order they are sent and the prompt is the only thing that assigns roles to them. Settle that wording on two or three 480p compositions, then reuse it across the run with the URLs substituted per item. Trim every reference clip to the seconds worth citing, since each one counts from 2 seconds upward toward a 15 second combined cap and those seconds are added to the output length when the call is priced.

All Seedance models10