This document explains the -less-restriction model variants offered for Seedance 2.0 and Seedream 5.0 Lite, how they differ from the standard (strict) variants, when to use them, and best practices for getting good results — including the limits that still apply on the looser variant.

Overview

For both Seedance 2.0 (video) and Seedream 5.0 Lite (image), we offer two parallel sets of models:

Strict (default) — applies our standard content review. Recommended for general production use.

Less-restriction — applies a more permissive content review. Recommended when your prompts and references include legitimate adult, expressive, or stylistic content that the strict variant may incorrectly reject.

Both variants run on the same underlying generation engine and accept exactly the same input schema — the only differences are the content-review strictness, the price, and the behavior on a content rejection (see Behavioral Differences below).

Available Models

Family	Strict (default)	Less-restriction
Seedance 2.0 — Pro	`seedance-2`	`seedance-2-less-restriction`
Seedance 2.0 — Fast	`seedance-2-fast`	`seedance-2-fast-less-restriction`
Seedream 5.0 Lite	`seedream-5-lite`	`seedream-5-lite-less-restriction`

Pricing

The -less-restriction variants are priced at a +25% markup over the matching strict tier at the same resolution / size.

Seedance — price per second

Model	480p	720p	1080p
seedance-2	$0.10	$0.20	$0.50
seedance-2-less-restriction	$0.125	$0.25	$0.625
seedance-2-fast	$0.08	$0.16	—
seedance-2-fast-less-restriction	$0.10	$0.20	—

seedance-2-fast and seedance-2-fast-less-restriction do not support 1080p.

Seedream — price per image

Model	2K (default)	3K
seedream-5-lite	$0.028	$0.042
seedream-5-lite-less-restriction	$0.035	$0.0525

3K is 1.5× the 2K rate within the same tier.

Cost calculation (Seedance with input videos)

Without input videos: credits = unit_price × output_duration

With input videos: credits = unit_price × output_duration + (unit_price / 2) × total_input_video_duration

Behavioral Differences

Aspect	Strict variants	Less-restriction variants
Content review	Standard, conservative	More permissive
Input schema	Identical	Identical
Output quality	Identical	Identical
Pricing	Base	+25% over base
Behavior on content rejection	The task may be retried once via an alternate review path that typically resolves prompts flagged for style / advertising / character intent	No retry. A content rejection is treated as final, your credits are refunded, and the task ends with a content-violation error

When you explicitly choose a -less-restriction model, you have already opted into the looser content review. A rejection there is treated as final and is not retried.

When to Use Less-Restriction

Pick the -less-restriction variants when one of the following use cases applies. Each block below is a copy-pasteable request body for that scenario.

Use case 1 — Adult / suggestive text-to-video

Your prompt contains legitimate adult, suggestive, romantic, or expressive descriptive content that the strict variant consistently refuses, and you have no reference media (text-to-video only).

{
  "model": "seedance",
  "task_type": "seedance-2-less-restriction",
  "input": {
    "prompt": "A young couple in dim candlelight, intimate slow-dance, soft shoulder reveal, cinematic shallow depth of field, warm rim light, 35mm grain",
    "duration": 5,
    "aspect_ratio": "9:16",
    "resolution": "720p"
  }
}

Use case 2 — Anime / illustration character with image reference

Your character is anime / illustration / non-photorealistic, and you want the model to follow the character's appearance from a reference image. Anime / 2D characters are not subject to the face restriction described later in this guide.

{
  "model": "seedance",
  "task_type": "seedance-2-fast-less-restriction",
  "input": {
    "prompt": "The anime character in @image1 walks slowly across a moonlit rooftop, looks back over her shoulder, hair catches the breeze, soft cinematic light",
    "duration": 5,
    "aspect_ratio": "9:16",
    "resolution": "720p",
    "image_urls": [
      "https://example.com/anime-character.png"
    ]
  }
}

Use case 3 — Motion / camera reference with face-masked video

You want to borrow the movement, camera language, and rhythm from a real-world reference video without bringing in the original person's identity. Pre-mask any visible faces in the reference video before passing it in.

{
  "model": "seedance",
  "task_type": "seedance-2-fast-less-restriction",
  "input": {
    "prompt": "Use the camera motion and rhythm from @video1 with the anime character from @image1 as the subject, maintain the illustration art style and the lighting from @image1",
    "duration": 5,
    "aspect_ratio": "9:16",
    "resolution": "720p",
    "image_urls": [
      "https://example.com/anime-character.png"
    ],
    "video_urls": [
      "https://example.com/face-masked-reference.mp4"
    ]
  }
}

Use case 4 — Stylized / adult-themed image generation (Seedream)

You want a single still image with stylized or adult-themed content that the strict variant rejects.

{
  "model": "seedream",
  "task_type": "seedream-5-lite-less-restriction",
  "input": {
    "prompt": "A stylized illustration of two anime characters at sunset on a beach, dramatic backlight, painterly brushstrokes, warm color palette",
    "aspect_ratio": "16:9",
    "size": "2K",
    "output_format": "jpeg"
  }
}

Stick with the strict (default) variants when

Your content is family-safe / general-purpose.

You want the maximum chance of recovery on a borderline rejection (the strict variant has a retry path that the less-restriction variant does not).

You want the lower price.

What is Still Blocked (Even on Less-Restriction)

The -less-restriction tier is not unrestricted. The following categories are blocked by the upstream regardless of which variant you choose:

Minors in sexual or suggestive contexts — absolute hard block.

Real-person identity — see "Real / virtual human faces" below.

Graphic violence, gore, torture, self-harm.

Hate / discriminatory content targeting protected groups.

Explicit drug use depicted in detail.

Bestiality, incest, and other categories prohibited by the upstream's usage policy.

A rejection on any of these categories cannot be recovered by switching variants, rephrasing the prompt, or retrying.

Real / Virtual Human Faces (Important)

Independent of the content review setting, the upstream's vision pipeline currently rejects reference images and reference videos that contain identifiable real human faces or photorealistic virtual human faces, regardless of the prompt's content. This restriction applies to both strict and less-restriction variants.

What works

Anime / illustration / 2D / non-photorealistic characters are not subject to the face restriction. You may pass these as image_urls directly.

Descriptive text prompts without any human-face reference image (e.g. "a 30-year-old Asian woman, soft natural light, cinematic") are not subject to the restriction.

What does not work

Uploading a clear photo of a real or photorealistic-virtual person as a reference image will be rejected during pre-processing.

A reference video where a recognizable real face is visible may also be rejected even if the rest of the prompt is acceptable.

Workaround: face-masking in reference videos

If you want to use a reference video for motion, camera language, rhythm, or composition (rather than for the person's identity), pre-process the video to:

Blur the face,

Pixelate / mosaic the face,

Mask the face with a solid block, or

Otherwise make the face unidentifiable.

This preserves the parts of the reference you actually want (movement, framing, atmosphere) while substantially reducing rejection risk.

Material Specifications

The same limits apply to both strict and less-restriction variants.

Reference images (Seedance + Seedream)

Formats: jpeg, png, webp, bmp, tiff, gif

Maximum size per image: 30 MB

Pixel dimensions per side: 300 – 6000 px

Seedance (omni_reference mode): up to 9 images; total references (images + videos + audios) must be 1–12

Seedream: up to 10 images

Reference videos (Seedance only)

Formats: mp4, mov

Duration per video: 2 – 15 seconds

Maximum size per video: 50 MB

Maximum number of videos: 3

Total duration across all videos: ≤ 15 seconds (the engine pre-validates this)

Reference audios (Seedance `omni_reference` only)

Formats: mp3, wav

Maximum number of audios: 3

Total duration: ≤ 15 seconds

Audio-only submissions are not allowed — at least one image or video reference must accompany audio

API Usage

The request body is identical to the strict variant — only the task_type field changes. All other parameters (prompt, mode, duration, resolution, aspect_ratio, image_urls, video_urls, audio_urls, size, output_format, etc.) work exactly the same way.

For copy-pasteable request bodies covering the most common scenarios, see the four use-case blocks in "When to Use Less-Restriction" above. For the full parameter reference, see the individual model docs:

Seedance 2 — seedance-2.txt

Seedream 5 Lite — seedream-5.txt

Best Practices

For prompts

Be specific. Describe lighting, framing, mood, and intent. Vague prompts produce inconsistent results across both tiers.

When using reference images and videos, reference them in the prompt using @image1, @video1, @audio1 (or the Chinese variants @图片1, @视频1, @音频1). The engine maps these to the upstream reference syntax automatically.

Make the role of each reference explicit in the prompt — for example, "use the style of @image1 and the motion of @video1, with the character described as ...".

For anime / illustration NSFW

Pass the character reference directly as an image_url.

Pass the motion reference as a video_url after face-masking if it contains real faces.

Use prompts that explicitly preserve the 2D / illustrated style ("maintain the anime style of @image1", "keep the hairstyle, clothing, and overall illustration aesthetic").

For real-person-look NSFW

Do not upload a photo of the person as a reference image — it will be rejected.

Describe the person in text (age range, ethnicity, hair, clothing, lighting) and let the model generate the appearance from the description.

If using a motion reference video that contains real faces, mask them first.

Testing and iteration

Start with short outputs (5 seconds) to validate the moderation pass and the overall composition cheaply, then scale up to longer durations once the prompt is stable.

For 720p / 1080p Seedance output you pay per second — running 5-second tests during prompt iteration keeps costs down.

Peak hours

From 09:00 to 15:00 GMT, Seedance experiences high traffic on both strict and less-restriction tiers. Queue times may extend to several hours during this period.

Summary

-less-restriction variants are a price-for-permissiveness trade — same engine, +25% cost, looser content review, no retry on rejection.

They are not unrestricted: minors, real-person identity, graphic violence, and other policy categories remain blocked.

The biggest practical limitation is the face restriction on reference images — anime works, photorealistic real or virtual faces do not. Workaround: describe the person in text and use only motion / style references with faces masked.

Input schema, output quality, and parameter behavior are identical to the strict variants. Only the task_type field changes.

Less-Restriction Models Guide

Overview#

Available Models#

Pricing#

Seedance — price per second#

Seedream — price per image#

Cost calculation (Seedance with input videos)#

Behavioral Differences#

When to Use Less-Restriction#

Use case 1 — Adult / suggestive text-to-video#

Use case 2 — Anime / illustration character with image reference#

Use case 3 — Motion / camera reference with face-masked video#

Use case 4 — Stylized / adult-themed image generation (Seedream)#

Stick with the strict (default) variants when#

What is Still Blocked (Even on Less-Restriction)#

Real / Virtual Human Faces (Important)#

What works#

What does not work#

Workaround: face-masking in reference videos#

Material Specifications#

Reference images (Seedance + Seedream)#

Reference videos (Seedance only)#

Reference audios (Seedance omni_reference only)#

API Usage#

Best Practices#

For prompts#

For anime / illustration NSFW#

For real-person-look NSFW#

Testing and iteration#

Peak hours#

Summary#