Seedance 2.0: modes, 4K, limits and pricing

Seedance 2.0 (seedance-2.0) is ByteDance’s video model, served on Playcut direct through BytePlus ModelArk. It is the cheapest per-second video model in the Playcut catalog — 10 credits/second at 480p, 21 at 720p — and it generates synchronized audio in the same pass (sound effects, ambience, speech). Pick it when you need long clips (up to 15 seconds), flexible aspect ratios including 21:9, or the lowest cost per finished second. Do not pick it if your input image contains a human face — see Limitations before you spend a credit.

ByteDance rejects input images containing human faces

Any image you send as a first frame or a reference is rejected at submit time with InputImageSensitiveContentDetected.PrivacyInformation if it may contain a real person. This includes AI-generated faces — the filter cannot tell synthetic from real. Text-to-video is unaffected. Image-to-video and reference-to-video with any human subject will fail. Full detail below.

Specs at a glance

	Seedance 2.0
Model id	seedance-2.0
Provider	ByteDance (BytePlus ModelArk)
Modes	text-to-video, image-to-video, reference-to-video (with optional audio reference), transition (start + end frame)
Aspect ratios	21:9 16:9 4:3 1:1 3:4 9:16 (default 16:9)
Resolutions	480p 720p 1080p 4k (default 720p; 4K added July 2026)
Duration	4–15 seconds, whole seconds (Playcut default 5)
Max reference images	7
Audio	Always on — every render has a soundtrack
Frame rate	24 fps
Credits	10 / 21 / 60 / 108 per second at 480p / 720p / 1080p / 4K
Not supported	Interpolation, video extension, video editing

Pixel dimensions at 16:9: 480p = 864×480 · 720p = 1248×704 · 1080p = 1920×1088.

Modes

Each mode maps to a different Playcut tool.

Mode	Tool	Required input
Text-to-video	generate-video-from-text	prompt
Image-to-video	generate-video-from-image	prompt + imageAssetId (used as the first frame)
Reference-to-video	generate-video-from-reference	prompt + referenceImageAssetIds (1–7)

Interpolation, extension, and video-edit are not available on Seedance 2.0. For interpolation and extension use Veo 3.1; for video editing use Wan 2.7.

Text-to-video

No image input, so the human-face restriction does not apply. Full range of aspect ratios, resolutions, and durations.

Image-to-video

Send one still image as imageAssetId. Seedance uses it as the first frame and generates motion forward from there. Audio is still generated.

Reference-to-video

Send up to 7 reference images in referenceImageAssetIds. References are addressed positionally inside your prompt text as @Image1, @Image2, and so on. A reference your prompt never names is effectively ignored by the model — always mention each one.

A cinematic dolly shot of the sneaker from @Image1 resting on wet asphalt,
neon reflections in the style of @Image2, rain audible in the background.

Optionally add an audio reference (audioAssetId, 15 seconds or shorter) to condition the generated soundtrack or voice on a real clip. Audio references are only available on this mode — the provider rejects audio combined with first/last frames.

Caution

First-frame input and reference media cannot be combined in the same request. “Use this scene as the first frame and these references” is structurally impossible on Seedance — pick one mode.

Transition (start + end frame)

Tool: generate-video-interpolation with model: "seedance-2.0". Provide startImageAssetId + endImageAssetId and a prompt describing the motion between them — Seedance bridges the two frames.

Unlike Veo interpolation (locked to 8 seconds), Seedance transitions take the full 4–15 second range at any resolution up to 4K. The two frames follow the same face rule as every other Seedance image input — no humans. Reference media cannot be added on this mode.

Pricing

Credits are charged per second of output.

Resolution	Credits / second	5 s clip	10 s clip	15 s clip
480p	10	50	100	150
720p	21	105	210	315
1080p	60	300	600	900
4K	108	540	1,080	1,620

For comparison at 720p: Seedance 2.0 is 21 cr/s versus Grok Imagine Video at 20 cr/s, Wan 2.7 at 40 cr/s, HappyHorse 1.1 at 56 cr/s, and Veo 3.1 at 160 cr/s. At 1080p Seedance matches Wan 2.7 (60 cr/s) well below HappyHorse (72) and Veo (160), and its 4K — the only non-Veo 4K in the catalog — is 108 cr/s versus Veo’s 240, with no 8-second duration lock.

Failed and moderation-rejected submissions are not billed by the provider. Playcut holds credits at submit and refunds them when a task ends FAILED or CANCELLED.

Note

Provider list pricing for Seedance 2.0 is taken from a BytePlus ModelArk pricing page capture and our own billed test runs, not from a live vendor page fetch. Playcut credit prices above are exact and come from our own pricing constants.

Limitations & gotchas

🔴 Human faces in input images are rejected

BytePlus states it in their own API reference: “Seedance 2.0 series models do not support direct upload of reference images or videos containing real human faces.”

What this means in practice:

• Any input image that may contain a real person is rejected at submit with HTTP 400 InputImageSensitiveContentDetected.PrivacyInformation.
• It applies to both the first-frame role (image-to-video) and the reference role (reference-to-video).
• AI-generated faces are also rejected. We verified this against a portrait generated by our own actor pipeline — the filter does not distinguish synthetic from real.
• It is not a toggleable content filter. Disabling the provider’s optional pre-filter does not change the outcome.
• The same wall exists on third-party resellers of the model, so switching routes does not help.
• Non-person imagery — scenery, products, packaging, vehicles, illustration — passes without issue.

Playcut deliberately leaves the model ungated in the picker and surfaces the provider’s error verbatim, because the restriction is content-dependent and we cannot reliably pre-detect it. If you need people in a video from an image, use Veo 3.1, Grok Imagine Video, HappyHorse 1.1, or Wan 2.7.

Other things to know

• Audio is always generated. There is no toggle to disable it. Every Seedance render ships with a soundtrack.
• Lip-sync quality is variable. Seedance’s audio is strongest on sound effects and ambience. Treat spoken dialogue as best-effort, not production-ready.
• 4k is real (since July 2026). All modes accept it at 108 credits/second, with the full 4–15 second range — no Veo-style 8-second lock. Renders are slower at 4K; long clips are likelier to hit the generation timeout.
• Aspect ratios are validated before submit. An unsupported ratio is rejected with a 400 before any credits are held — only the six ratios listed above are accepted.
• Long renders can time out. Generation is asynchronous with a 9-minute ceiling on our side. A 15-second 1080p render is the most likely to hit it.
• Up to 9 reference images — the provider maximum. Video-clip references exist upstream but are not offered on Playcut; a single audio reference (≤15 s) is supported on reference mode.

How to generate

Same tools-registry shape as everything else in the API — set model to seedance-2.0.

Text-to-video

curl -X POST https://api.playcut.ai/api/v1/tools/generate-video-from-text/execute \
  -H "Authorization: Bearer $PLAYCUT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Aerial shot over a neon-lit rainy city at night, slow push in",
    "model": "seedance-2.0",
    "aspectRatio": "21:9",
    "resolution": "720p",
    "durationSeconds": 10
  }'

Image-to-video

curl -X POST https://api.playcut.ai/api/v1/tools/generate-video-from-image/execute \
  -H "Authorization: Bearer $PLAYCUT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Camera orbits slowly around the product, steam rising",
    "imageAssetId": "65f3a1b2c3d4e5f600000003",
    "model": "seedance-2.0",
    "resolution": "1080p",
    "durationSeconds": 6
  }'

Reference-to-video

curl -X POST https://api.playcut.ai/api/v1/tools/generate-video-from-reference/execute \
  -H "Authorization: Bearer $PLAYCUT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "The sneaker from @Image1 on wet asphalt, lit like @Image2",
    "referenceImageAssetIds": ["65f3a1b2c3d4e5f600000003", "65f3a1b2c3d4e5f600000004"],
    "model": "seedance-2.0",
    "aspectRatio": "16:9",
    "resolution": "720p",
    "durationSeconds": 8
  }'

Every generation tool returns a task acknowledgement. Poll with get-task-status or block with wait-for-task, then fetch a signed URL with download-asset — see the Quickstart for the full loop.

The same tools are available over MCP under the identical names (generate-video-from-text, generate-video-from-image, generate-video-from-reference). See MCP Install.

Related

All models The full Playcut catalog — three image models, seven video models, music and voice — side by side.

Aspect ratios by model Which ratios every Playcut image and video model accepts, side by side.

Resolutions and pricing Full credit cost comparison across the video catalog.

Veo 3.1 Highest quality, interpolation and extension, 4k output.

Wan 2.7 Audio-driven image-to-video and video editing.