Seedance 2.5 video
Seedance 2.5 is a ByteDance video model served through the BytePlus ModelArk API. On Sogni its model id is seedance-2-5.
Seedance 2.5 is the latest Seedance generation and the default for current storyboard workflows. It generates longer clips, accepts a wider set of multimodal references, and renders up to 4K. Seedance 2.0 remains available for existing workflows.
#Seedance 2.5 or Seedance 2.0?
Seedance 2.5 renders up to 4K. Select 4K explicitly for a new video, an edit, an extension, or a final made from a 480p draft. Existing resolution defaults are unchanged.
| Seedance 2.5 | Seedance 2.0 | |
|---|---|---|
| Resolutions | 480p, 720p, 1080p, 4K | 480p, 720p, 1080p, 4K |
| Clip length | 4–30 seconds | 4–15 seconds |
| Frame rate | 24 fps | 24 fps |
| Native audio | Yes | Yes |
| Reference workflows | Reference-to-video, video editing, video extension | Reference-to-video, video editing, video extension |
| Reference images | Up to 30 | Up to 9 |
| Reference videos | Up to 10, 30 seconds in total | Up to 3, 15 seconds in total |
| Reference audio clips | Up to 10, 30 seconds in total | Up to 3, 15 seconds in total |
| Total media per request | 50 items | 12 assets |
| Current storyboard choice | Yes | Available for existing workflows |
Choose Seedance 2.5 for current storyboard workflows, output up to 4K, clips longer than 15 seconds, first-and-last-frame transitions, or a set of references larger than Seedance 2.0 can accept.
Choose Seedance 2.0 to continue an existing 2.0 workflow. It keeps its existing resolution choices and limits.
#Generation limits
- Resolutions: 480p, 720p, 1080p, and 4K.
- Duration: 4 to 30 seconds.
- Frame rate: fixed 24 fps.
- Output: MP4.
- Audio: generated natively together with the video.
#Capabilities
| Mode | Inputs |
|---|---|
| Text to video | Prompt |
| Image to video | Prompt + first frame |
| First and last frame | Prompt + first and last frames |
| Reference to video | Prompt + reference media |
| Video editing | Prompt + source video |
| Video extension | Prompt + source video |
Reference-to-video, video editing, and video extension are all served by Seedance 2.5's multimodal reference path. Seedance 2.0 exposes the same modes; the difference is scale, not the mode list — 2.5 takes a much larger reference set and renders up to 30 seconds per call.
Keyframes. In Sogni Web's Reference to Video on Seedance 2.5 and Seedance 2.5 Uncensored, turn on Add keyframes to choose attached reference images and the whole second, from 0 to the clip's length, where each should appear. Sogni sends them in order with their timing alongside your prompt. Seedance uses them as visual guidance, so exact images and timing may vary.
#Reference limits
A single Seedance 2.5 request accepts up to 30 reference images, 10 reference videos, and 10 reference audio clips, for a maximum of 50 media items in total. The per-modality caps apply on top of that total, so 50 items is only reachable by combining all three kinds. The text prompt does not count toward the cap. The reference videos can add up to 30 seconds in total, and the reference audio clips, separately, to 30 seconds.
That is a large increase over Seedance 2.0, Fast, and Mini, which accept up to 9 reference images, 3 reference videos, and 3 reference audio clips, for 12 assets in total; there, the reference videos, and separately the reference audio clips, can add up to 15 seconds. If a job needs more reference material than 2.0 can take, that is the clearest reason to pick 2.5.
#Pricing
The prices below cover output without a video reference. R2V with video references charges for their combined input seconds plus output seconds on every generated clip, using the model's with-video rate. Reusing a reference on another clip adds its input charge again. See how R2V video charges work for examples and quote reservations.
Seedance 2.5's price depends on the resolution you select. The 480p–1080p rates below are per second. Values use the standard 1 Spark = $0.005 peg, before subscriber discounts.
| Resolution | Per second | 5 seconds | 10 seconds | 30 seconds |
|---|---|---|---|---|
| 1080p | $0.739206 | $3.70 | $7.39 | $22.18 |
| 720p | $0.300456 | $1.50 | $3.00 | $9.01 |
| 480p | $0.133640 | $0.67 | $1.34 | $4.01 |
Totals are computed from the exact per-second rates above, so a long clip does not match a rounded rate multiplied out by hand — 30 seconds at 720p is $9.01, not $9.00.
4K examples: 3840 × 2160, 16:9, 24 fps, no video reference. The quote includes the output's endpoint frame, so use a complete job quote rather than multiplying a rounded per-second rate. Other aspect ratios and video references can change the total.
| Duration | Output frames | Spark | USD |
|---|---|---|---|
| 5 seconds | 121 | 1,197.68 | $5.9884 |
| 10 seconds | 241 | 2,385.47 | $11.9273 |
| 30 seconds | 721 | 7,136.60 | $35.6830 |
In Spark, at 1080p that is about 147.8 Spark per second — 739.2 Spark for 5 seconds, 1,478.4 Spark for 10 seconds, and 4,435.2 Spark for 30 seconds. At 720p it is about 60.1 Spark per second — 300.5 Spark for 5 seconds, 600.9 Spark for 10 seconds, and 1,802.7 Spark for a full 30-second clip. At 480p it is about 26.7 Spark per second, so 133.6 Spark for 5 seconds, 267.3 Spark for 10 seconds, and 801.8 Spark for 30 seconds.
- Duration and resolution set the price. At 720p a 30-second clip costs six times a 5-second one, and 1080p costs about 2.5 times 720p.
- A reference video adds its seconds to the bill. Its input seconds are billed like output seconds, at the with-video rate of $0.179712 per second at 720p: a 15-second reference plus 15 seconds of output bills as 30 seconds, $5.39 (1,078.3 Spark), against $4.51 (901.4 Spark) for 15 seconds with no reference video. Previews use combined input-video duration when available, or a conservative 30-second input allowance when clip timing is unavailable. The server measures the combined input duration before reserving funds. Final charges can settle down to provider-confirmed usage and never exceed the reservation; provider minimum billable usage still applies. Input counts as at least two-thirds of the output duration, rounded up to a whole second.
- 480p is a cheaper iteration path. At about 44% of the 720p rate, it is a sensible place to test prompts and timing before a higher-resolution render.
For an exact per-job quote at any duration, use the job estimate calculator, which quotes live from the current Sogni pricing endpoints.
#Draft: 480p first, then 1080p or 4K
Turn on Draft in Sogni Web to render a Seedance 2.5 video as a 480p draft, priced like any 480p video. When you like it, choose Render final, then 1080p or 4K. The final keeps the draft's prompt, references, length, aspect ratio and seed. The dialog starts at 1080p and shows a quote for the resolution you choose before you submit it.
- Use the time shown on the draft. Each draft shows its expiry and time left to become a final video.
- Every final is priced separately, for its selected resolution and the draft's length, and at the with-video rate when the draft used a reference video. Rendering the same draft again is charged again.
- Seedance 2.5 Uncensored has Draft too, at its own rates.
- In the SDK, set
seedanceDraft: trueon a Seedance 2.5 video project. To render its final, create a project withseedanceDraftJobIdset to the draft's job id. AddseedanceDraftFinalResolution: '4k'for 4K, or'1080p'for 1080p; omitting it keeps the existing 1080p behavior. Send the same fields toestimateVideoCostfor the matching quote. Use a current SDK version that exposesseedanceDraftFinalResolution.
#Access
Seedance 2.5 is a Premium Spark model like the other Seedance tiers, so Free Spark can't pay for it. Like Seedance 2.0, Fast, and Mini, it is a third-party vendor API model and is not covered by any Sogni subscription tier — see Unlimited plan details.
Seedance renders on the external ByteDance/BytePlus vendor path, so a Seedance job does not consume Sogni GPU worker capacity.
#Where you can use it
Seedance 2.5 is available in Sogni Web through the ByteDance video flow with the required Premium Spark balance.
Seedance 2.5, Seedance 2.0, and Mini accept photos and clips of real people once your account accepts a one-time likeness agreement. All three decline mature prompts; for mature themes, use Seedance 2.5 Uncensored or Seedance 2.0 Mini Uncensored.
It is also selectable in the Creative Agent tool surfaces as the videoModel value seedance2-5, accepted by generate_video, sound_to_video, and video_to_video (with controlMode="seedance-v2v"). animate_photo does not take any Seedance model — use generate_video with reference indices instead. Seedance 2.0 and Mini remain available as seedance2 and seedance2-mini; the legacy seedance2-fast selector now routes to Mini, which is faster and cheaper.
#Real people
Seedance 2.5, Seedance 2.0, and Mini accept your own likeness, or that of people who gave you explicit, legal consent, once your account accepts the one-time likeness and consent agreement. It is the same agreement the uncensored models use, so if you already accepted it, you are covered. Without it, they refuse a photo or video that may contain a real person. Mature prompts are still declined.
- Best likeness. Likeness holds best in Reference to Video: attach the person's photo as a reference and give it a role in the prompt, for example
Use @Image1 as the presenter. Image to Video, with the photo as the first frame, can drift away from the face on big head turns. - API and SDKs. The agreement belongs to the account, so once it is accepted in a Sogni app,
seedance-2-5,seedance-2-0, andseedance-2-0-minijobs (Creative Agentseedance2-5,seedance2, andseedance2-mini) from that account accept real people too. They never fail with4103; from an account without the agreement, a real-person input is refused with a message that names the agreement page for that model. On Seedance 2.5 it reads:Seedance rejected the input image because it may contain a real person. Try again after accepting the one-time likeness agreement at https://app.sogni.ai/account/model-consents/seedance-2-5, or try Seedance 2.5 Uncensored or MiniMax H3 instead.
#Uncensored
Seedance 2.5, Seedance 2.0, and Mini decline mature prompts. Two uncensored versions run on a more permissive content policy for mature themes: Seedance 2.5 Uncensored (seedance-2-5-uncensored), also known as Seedance 2.5 Spicy, and Seedance 2.0 Mini Uncensored (seedance-2-0-mini-uncensored). Each is a separate model from its standard version; selecting one never runs the other.
#Seedance 2.5 Uncensored
Model id seedance-2-5-uncensored. The Creative Agent videoModel value seedance2-5-uncensored is accepted wherever seedance2-5 is: generate_video, sound_to_video, and video_to_video (with controlMode="seedance-v2v"). It has every Seedance 2.5 capability and limit on this page: 480p, 720p, 1080p, or 4K; 4–30 seconds at 24 fps with native audio; text, first and last frames, references, video editing, and extension; up to 30 images, 10 videos, and 10 audio clips (50 items). Sogni Chat resolves "Seedance 2.5 Uncensored" and "Seedance Uncensored" to it. Open it in Sogni Web.
#Seedance 2.0 Mini Uncensored
Model id seedance-2-0-mini-uncensored. The Creative Agent videoModel value seedance2-mini-uncensored is accepted wherever seedance2-mini is. It has every Seedance 2.0 Mini capability and limit: 480p or 720p; 4–15 seconds at 24 fps with native audio; 16:9, 9:16, 1:1, 4:3, 3:4, and 21:9; text, a first frame, an image with audio, or references, up to 9 images, 3 videos, and 3 audio clips (12 items, and audio needs an image or video reference). Sogni Chat resolves "Seedance 2.0 Mini Uncensored" and "Seedance Mini Uncensored" to it. Full Seedance 2.0 has no uncensored version; for mature themes, use Seedance 2.5 Uncensored. Open it in Sogni Web.
#Likeness and consent agreement
Seedance video generation models share one one-time agreement; accepting it once covers them all, and earlier acceptances count. The uncensored models require it: Sogni refuses every job for either from an account until that account has accepted the current version. Seedance 2.5, Seedance 2.0, and Mini never require it; without it, they refuse real people.
In the agreement you confirm that:
- the likeness in any media you upload, and in any video Sogni generates from it, is your own, or you have explicit, legal consent from every person shown;
- you understand Sogni or the model provider may run a real-person verification check to confirm this; and
- you agree to the Sogni.ai Terms of Service and Privacy Policy, and that violations can lead to account suspension and legal liability.
Sogni Web and Sogni Chat show the agreement the first time you pick or submit either uncensored Seedance model. For Seedance 2.5, Seedance 2.0, and Mini, Sogni Web offers it when a real-person input is refused, and at app.sogni.ai/account/model-consents/seedance-2-5. Tick I agree to continue. You will not be asked again unless the agreement's terms change, in which case every account is asked to accept the new version once.
The agreement can only be accepted in a signed-in Sogni app session, not with an API key. API and SDK users accept it once in Sogni Web with the same account; their jobs run after that. Price estimates are not gated, so you can quote a job before accepting.
#Error 4103
A job for either uncensored model from an account that has not accepted the current agreement fails before it is charged, with error code 4103:
{
"error": "4103",
"isFromWorker": false,
"jobID": "…",
"error_message": "Seedance 2.5 Uncensored requires a one-time likeness and consent agreement. Review and accept it in the Sogni app, then try again.",
"consentRequired": { "key": "seedance-2-5-uncensored", "version": 2, "modelId": "seedance-2-5-uncensored" }
}
Retrying does not help. Accept the agreement in the Sogni app, then submit the job again. Both models' entries in /api/v1/models/tiers carry consentRequired: { "key": "seedance-2-5-uncensored", "version": 2 }, so a client can tell in advance that a model needs the agreement; in a 4103 error, modelId names the model you requested. GET /v1/account/model-consents/seedance-2-5-uncensored on the Sogni API reports whether the authenticated account has accepted it (accepted, acceptedVersion).
#Uncensored pricing
Both models come at a premium rate above their standard versions at every resolution. They are Premium Spark models (Free Spark can't be used) and are not covered by any Sogni subscription tier; Unlimited members save 5% and Unlimited Pro members 10%. Values below use the standard 1 Spark = $0.005 peg, before subscriber discounts.
Seedance 2.5 Uncensored
| Resolution | Per second | 5 seconds | 10 seconds | 30 seconds |
|---|---|---|---|---|
| 1080p | $0.852930 | $4.26 | $8.53 | $25.59 |
| 720p | $0.346680 | $1.73 | $3.47 | $10.40 |
| 480p | $0.154200 | $0.77 | $1.54 | $4.63 |
In Spark, that is about 170.6 Spark per second at 1080p, 69.3 at 720p, and 30.8 at 480p; a 5-second clip is 852.9, 346.7, or 154.2 Spark.
4K examples: 3840 × 2160, 16:9, 24 fps, no video reference, including the output's endpoint frame.
| Duration | Output frames | Spark | USD |
|---|---|---|---|
| 5 seconds | 121 | 1,381.94 | $6.9097 |
| 10 seconds | 241 | 2,752.46 | $13.7623 |
| 30 seconds | 721 | 8,234.54 | $41.1727 |
Seedance 2.0 Mini Uncensored
| Resolution | Per second | 5 seconds | 10 seconds | 15 seconds |
|---|---|---|---|---|
| 720p | $0.189000 | $0.95 | $1.89 | $2.84 |
| 480p | $0.084066 | $0.42 | $0.84 | $1.26 |
In Spark, that is about 37.8 Spark per second at 720p and 16.8 at 480p; a 5-second clip is 189 or 84.1 Spark.
- A reference video adds its seconds to the bill. Its input seconds are billed like output seconds: 15 seconds of reference plus 15 seconds of output at 720p bills as 30 seconds, $6.22 (1,244.2 Spark) on Seedance 2.5 Uncensored and $3.40 (680.4 Spark) on Seedance 2.0 Mini Uncensored. Previews use combined input-video duration when available. Without clip timing, they allow 30 seconds of input on 2.5 Uncensored or 15 on Mini Uncensored. The server measures combined input duration before reserving funds, including provider minimum billable usage. Final charges can settle down to provider-confirmed usage and never exceed the reservation.
- 480p is the cheaper iteration path for testing prompts and timing before a longer or higher-resolution render.
For an exact per-job quote, use the job estimate calculator.
Need help? Ask in our Discord!
For questions about the model lineup, requests for new models, or tips — join the Sogni Discord ✨