Seedance 2

Overview

Seedance 2 is ByteDance’s latest video generation model. In varg you address it by name and varg picks the best provider route for your input:

varg name	What it is	Credits (5s @ 720p)
`seedance_2`	Canonical — best quality. Auto-routes: text-to-video by default (up to 4k), image-to-video when you send files	160
`seedance_2_mini`	Cheapest tier (ByteDance mini)	74
`seedance_2_less_restriction`	Permissive content review — explicit opt-in by choosing this name, +10% over strict tiers	116

All three share the same input schema. Under the hood each name spans one or more provider routes — see Advanced: exact route addressing if you need to pin a specific route.

The legacy seedance-2-preview and seedance-2-fast-preview names still work but are deprecated — use the names above.

Quick start

import { createVarg } from "vargai/ai"

const varg = createVarg({ apiKey: process.env.VARG_API_KEY! })

const result = await varg.videoModel("seedance_2").generate({
  prompt: "a golden retriever running through autumn leaves in slow motion",
  duration: 5,
  aspectRatio: "16:9",
})

console.log(result.video.url)

# Submit generation
curl -X POST https://api.varg.ai/v2/video \
  -H "Authorization: Bearer $VARG_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "seedance_2",
    "prompt": "a golden retriever running through autumn leaves in slow motion",
    "duration": 5,
    "aspect_ratio": "16:9"
  }'

# Poll for result (use the `id` from the response)
curl https://api.varg.ai/v2/jobs/JOB_ID \
  -H "Authorization: Bearer $VARG_API_KEY"
# When completed, the video URL is in output.outputs[0].url

Capabilities

Text-to-video

Generate video from a text prompt alone. The default route supports resolutions up to 4k.

const result = await varg.videoModel("seedance_2").generate({
  prompt: "aerial shot of a futuristic city at sunset, flying cars, neon lights",
  duration: 10,
  aspectRatio: "16:9",
})

curl -X POST https://api.varg.ai/v2/video \
  -H "Authorization: Bearer $VARG_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "seedance_2",
    "prompt": "aerial shot of a futuristic city at sunset, flying cars, neon lights",
    "duration": 10,
    "aspect_ratio": "16:9"
  }'

Image-to-video

Animate a still image. The first image in files becomes the start frame. An optional end frame can be set via provider_options.

const result = await varg.videoModel("seedance_2").generate({
  prompt: "the woman turns to face the camera and smiles",
  imageUrls: ["https://example.com/portrait.jpg"],
  duration: 5,
})

curl -X POST https://api.varg.ai/v2/video \
  -H "Authorization: Bearer $VARG_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "seedance_2",
    "prompt": "the woman turns to face the camera and smiles",
    "duration": 5,
    "files": [{ "url": "https://example.com/portrait.jpg" }]
  }'

curl -X POST https://api.varg.ai/v2/video \
  -H "Authorization: Bearer $VARG_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "seedance_2",
    "prompt": "smooth transition from day to night",
    "duration": 5,
    "files": [{ "url": "https://example.com/day.jpg" }],
    "provider_options": { "fal": { "end_image_url": "https://example.com/night.jpg" } }
  }'

Reference-to-video

Compose a video from multiple references: up to 9 images, 3 videos, and 3 audio files. Reference them in your prompt with @Image1, @Video1, @Audio1, and so on. Reference mode is not reachable by auto-routing — it requires exact route addressing:

curl -X POST https://api.varg.ai/v2/video \
  -H "Authorization: Bearer $VARG_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "fal:bytedance/seedance-2.0/fast/reference-to-video",
    "prompt": "@Image1 dances in the style of @Video1, set to @Audio1",
    "files": [
      { "url": "https://example.com/character.jpg" },
      { "url": "https://example.com/dance-reference.mp4" },
      { "url": "https://example.com/music.mp3" }
    ],
    "duration": 5
  }'

Budget tier (seedance_2_mini)

seedance_2_mini is the cheapest Seedance tier. Same input schema as seedance_2 — integer duration 4 to 15 (no "auto"). 5s at 720p costs 74 credits; drop to 480p for 37 credits.

const result = await varg.videoModel("seedance_2_mini").generate({
  prompt: "a paper boat drifting down a rainy street",
  duration: 5,
  aspectRatio: "16:9",
})

curl -X POST https://api.varg.ai/v2/video \
  -H "Authorization: Bearer $VARG_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "seedance_2_mini",
    "prompt": "a paper boat drifting down a rainy street",
    "duration": 5,
    "aspect_ratio": "16:9"
  }'

Less-restriction variant (seedance_2_less_restriction)

seedance_2_less_restriction runs Seedance with a more permissive content review, at +10% over the matching strict tier (5s at 720p = 116 credits). Choosing this name is the explicit opt-in — auto-routing never selects a less-restriction route on your behalf.

const result = await varg.videoModel("seedance_2_less_restriction").generate({
  prompt: "your prompt",
  duration: 5,
  aspectRatio: "16:9",
})

curl -X POST https://api.varg.ai/v2/video \
  -H "Authorization: Bearer $VARG_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "seedance_2_less_restriction",
    "prompt": "your prompt",
    "duration": 5,
    "aspect_ratio": "16:9"
  }'

Behavior differences:

If the provider still rejects the content, the job is not retried on another route. Your credits are refunded.
The following remain blocked regardless of variant: content involving minors, real-person identity or faces, graphic violence, and hate content.

Photorealistic reference faces are rejected even on less-restriction routes — the model refuses to animate real-person likenesses. Anime and 2D character references work. If you need a specific-looking person, describe them in the text prompt instead of providing a photo.

seedance_2_less_restriction auto-routes to the pro less-restriction tier. Fast and mini less-restriction tiers are reachable only via exact route addressing.

Parameters

string

required

Text description of the video to generate. In reference mode, address inputs with @Image1, @Video1, @Audio1, etc.

integer | 'auto'

default:"5"

Video duration in seconds, 4 to 15. The fal routes also accept "auto" (the model picks a natural length). PiAPI routes (including seedance_2_mini and seedance_2_less_restriction) accept integers only.

string

default:"16:9"

Output aspect ratio. Supported values: auto, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16.

array

Array of { url: string } objects. On seedance_2, an image triggers image-to-video (start frame). In reference mode, up to 9 images, 3 videos, and 3 audio files.

object

Provider-native overrides. PiAPI routes default to 720p — override with {"piapi": {"input": {"resolution": "480p"}}}. fal routes: {"fal": {"resolution": "480p"}}.

{
  "model": "seedance_2_mini",
  "prompt": "a paper boat drifting down a rainy street",
  "duration": 6,
  "provider_options": { "piapi": { "input": { "resolution": "480p" } } }
}

Pricing

All Seedance 2 routes bill per second of output video:

Model	Route detail	480p	720p	1080p	4k
`seedance_2`	fal text-to-video (default)	$0.135/s	$0.3034/s	$0.682/s	$1.556/s
`seedance_2`	fal fast i2v / r2v (with files)	$0.108/s	$0.2419/s	—	—
`seedance_2_mini`	PiAPI mini	$0.07/s	$0.14/s	—	—
`seedance_2_less_restriction`	PiAPI pro, permissive review	+10% over pro	+10% over pro	+10% over pro	—

Worked examples: 5s at 720p costs 160 credits on seedance_2 (text-to-video), 74 credits on seedance_2_mini, and 116 credits on seedance_2_less_restriction. At 480p, seedance_2_mini drops to 37 credits. Additional routes are reachable via exact addressing: piapi:seedance-2 (pro:

0.10 /

0.20 /

0.50 per second at 480p / 720p / 1080p), `piapi:seedance-2-fast` (

0.08 / $0.16 per second), and the fast/mini less-restriction variants at +10% over their strict tiers. Estimates are exact when you pass a numeric duration. One exception: duration: "auto" (fal routes) makes the length unknowable upfront, so the estimate reserves the worst case (~479 credits at 720p) and the charge equals the reservation. Pass an explicit duration when cost matters. Check the price and route before submitting:

curl -s -X POST https://api.varg.ai/v2/estimate \
  -H "Authorization: Bearer $VARG_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "seedance_2", "prompt": "a cat", "duration": 5}'
# {"valid": true, "model": "seedance_2", "tool": "video",
#  "resolved_model_key": "fal:bytedance/seedance-2.0/text-to-video",
#  "pricing": {"provider_cost_cents": 152, "user_price_cents": 160, ...}}

Migrating from preview names. The legacy seedance-2-preview and seedance-2-fast-preview names still work (they redirect upstream) but are deprecated — use seedance_2, seedance_2_mini, or seedance_2_less_restriction. The old flat pricing (250/150 credits per generation) no longer applies; all routes now bill per second as shown above.

Advanced: exact route addressing

Most requests should just use a varg name — each name auto-routes across one or more provider routes by priority and input shape. For seedance_2:

No files → fal:bytedance/seedance-2.0/text-to-video
Files present → fal:bytedance/seedance-2.0/fast/image-to-video (first image becomes the start frame)

Every route also has a permanent model_key, listed in GET /v2/models?tool=video (public, no auth). Passing a model_key in model pins that exact route. Three address forms, from loose to exact:

{ "model": "seedance_2" }                                          // canonical — varg picks the route
{ "model": "piapi:seedance_2" }                                    // provider prefix — restrict to PiAPI routes
{ "model": "fal:bytedance/seedance-2.0/fast/reference-to-video" }  // exact model_key — pin one route

The main reason to pin a route: reference-to-video is not reachable by auto-routing.

curl -X POST https://api.varg.ai/v2/video \
  -H "Authorization: Bearer $VARG_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "fal:bytedance/seedance-2.0/fast/reference-to-video",
    "prompt": "@Image1 walks through the scene from @Image2 while @Audio1 plays",
    "files": [
      { "url": "https://example.com/character.jpg" },
      { "url": "https://example.com/background.jpg" },
      { "url": "https://example.com/theme.mp3" }
    ],
    "duration": 5
  }'

Full route list:

Priority	model_key	Mode	Pricing (per second)
1	`fal:bytedance/seedance-2.0/text-to-video`	Text-to-video, standard tier, up to 4k	$0.135 (480p),$ 0.3034 (720p), $0.682 (1080p),$ 1.556 (4k)
2	`fal:bytedance/seedance-2.0/fast/image-to-video`	Image-to-video (start + optional end frame), fast tier	$0.108 (480p),$ 0.2419 (720p)
3	`fal:bytedance/seedance-2.0/fast/reference-to-video`	Reference-to-video (up to 9 images, 3 videos, 3 audio refs)	$0.108 (480p),$ 0.2419 (720p)
10	`piapi:seedance-2`	Pro tier, t2v / i2v / omni	$0.10 (480p),$ 0.20 (720p), $0.50 (1080p)
11	`piapi:seedance-2-fast`	Fast tier	$0.08 (480p),$ 0.16 (720p)
12	`piapi:seedance-2-mini`	Mini tier (served by `seedance_2_mini`)	$0.07 (480p),$ 0.14 (720p)
20	`piapi:seedance-2-less-restriction`	Pro, permissive review (served by `seedance_2_less_restriction`)	Pro +10%
21	`piapi:seedance-2-fast-less-restriction`	Fast, permissive review	Fast +10%
22	`piapi:seedance-2-mini-less-restriction`	Mini, permissive review	Mini +10%

model_key values are a permanent public contract — safe to hardcode.

Not sure which route your request will hit? Send the same body to POST /v2/estimate — the response includes resolved_model_key and the exact price, without creating a job.

Tips

Cost-sensitive? Use seedance_2_mini. 5s at 480p is 37 credits — the cheapest Seedance path (provider_options: {"piapi": {"input": {"resolution": "480p"}}}).
Run POST /v2/estimate first — it shows the resolved_model_key and price without creating a job or reserving credits.
Image-to-video produces better results than text-to-video. Generate a reference image first with nano-banana-pro, then animate it.
Use @Image1 / @Video1 / @Audio1 syntax in reference mode to control which input maps to which element of the scene.
Be specific about motion — describe camera movements (“slow pan left”, “zoom in”) and subject actions (“walks toward camera”, “turns around”) explicitly.
Aspect ratios matter — 9:16 for TikTok/Reels/Shorts, 16:9 for YouTube/landscape, 21:9 for cinematic widescreen, or auto to let the model decide.

Composition example

Use Seedance in a full video composition with the SDK:

/** @jsxImportSource vargai */
import { Render, Clip, Image, Video, Speech, Captions } from "vargai"
import { createVarg } from "vargai/ai"

const varg = createVarg({ apiKey: process.env.VARG_API_KEY! })

const scene = Image({
  model: varg.imageModel("nano-banana-pro"),
  prompt: "professional woman in modern office, confident pose",
  aspectRatio: "9:16",
})

export default (
  <Render width={1080} height={1920} fps={30}>
    <Clip duration={5}>
      <Video
        model={varg.videoModel("seedance_2")}
        prompt={{ text: "she turns and smiles at the camera", images: [scene] }}
        duration={5}
      />
    </Clip>
  </Render>
)

Kling V3

Best overall video model. Supports 3-15s flexible duration.

Sora 2

OpenAI’s video model. Good for cinematic shots.

Wan 2.5

Budget-friendly video generation with good character motion.

Nano Banana Pro

Generate reference images to use with Seedance image-to-video.

Get Started

Models

SDK

Templates

MCP

AI Agents

Comparisons

Overview

Quick start

Capabilities

Text-to-video

Image-to-video

Reference-to-video

Budget tier (seedance_2_mini)

Less-restriction variant (seedance_2_less_restriction)

Parameters

Pricing

Advanced: exact route addressing

Tips

Composition example

Kling V3

Sora 2

Wan 2.5

Nano Banana Pro

​Overview

​Quick start

​Capabilities

​Text-to-video

​Image-to-video

​Reference-to-video

​Budget tier (seedance_2_mini)

​Less-restriction variant (seedance_2_less_restriction)

​Parameters

​Pricing

​Advanced: exact route addressing

​Tips

​Composition example

​Related models

Kling V3

Sora 2

Wan 2.5

Nano Banana Pro

Overview

Quick start

Capabilities

Text-to-video

Image-to-video

Reference-to-video

Budget tier (seedance_2_mini)

Less-restriction variant (seedance_2_less_restriction)

Parameters

Pricing

Advanced: exact route addressing

Tips

Composition example

Related models