Back to Models

Doubao Seedream 4.5

doubaodoubao/doubao-seedream-4-5

Doubao SeedDream 4.5 β€” text/image-to-image generation, Chinese-bilingual prompt support.

Doubao Seedream 4.5 is ByteDance's unified image generation and editing model, developed by Team Seedream and served through the Volcengine Ark platform. Unlike traditional pipelines that separate creation from editing, Seedream 4.5 merges both capabilities into a single architecture β€” you can generate images from text prompts and refine them with reference images in the same API call.

The model excels at Chinese-English bilingual prompt understanding, text rendering within images (ideal for posters, e-commerce banners, and UI mockups), multi-image sequential generation for maintaining character consistency across a set, and reference-based editing that preserves facial features, lighting, and color tone. Output resolution reaches up to 4 megapixels (2048Γ—2048) with flexible aspect ratios.

Best for
  • β€’ E-commerce product imagery with embedded text overlays and pricing badges
  • β€’ Marketing posters and brand visuals requiring accurate Chinese/English typography
  • β€’ Multi-image storyboards and sequential illustrations with consistent characters
  • β€’ Reference-based image editing β€” style transfer, outfit changes, background replacement while preserving identity
Reach for something else if
  • β€’ Photorealistic human portraits at extreme close-up β€” dedicated portrait models (e.g. doubao/doubao-seed-character) offer finer face control
  • β€’ Tasks requiring OpenAI-native image reasoning (think-then-draw) β€” route to openai/gpt-image-2 instead
  • β€’ Western-language-heavy dense text rendering where FLUX Kontext models may perform better
Context Length
--
Max Output
--
Request Priceper request
$0.0432/ request

Modalities

textimage→image

Capabilities

VisionImage Generation

Media Generation Capabilities

image_generation
sizes
  • 1024x1024
  • 864x1152
  • 1152x864
defaults
size
1024x1024

Pricing Breakdown

TypeRate
Request$0.0432 / request

Per-image output price. BytePlus ModelArk `seedream-4-5-251128`: input image free, output $0.04/image (docs.byteplus.com/en/docs/ModelArk/1544106, read 2026-07-29).

Supported Parameters

promptsizenresponse_formatuser

Specifications

Release date2025-12-03seed.bytedance.com β†—verified
Parameters~1.2Bdocs.apiyi.com β†—to verify
Max output resolution4MP (up to 2048Γ—2048)fal.ai β†—verified
ArchitectureUnified text-to-image + image editing (multimodal generation)seed.bytedance.com β†—verified
Batch capabilityUp to 6 images per request; sequential mode for consistent multi-image setsfal.ai β†—verified
Prompt languagesChinese and English (bilingual native support)seed.bytedance.com β†—verified
Model versiondoubao-seedream-4-5-251128www.volcengine.com β†—verified
LicenseCommercial use permitted via Volcengine Ark / TheRouter APIwww.volcengine.com β†—to verify

Benchmarks

BenchmarkDistributionScoreSource
GenEval
Compositional generation quality benchmark; reported in Seedream 4.0 paper (4.5 is the release build of 4.0 research).
0.76arxiv.org β†—
T2I-CompBench
Compositional text-to-image benchmark; paper reports state-of-the-art among Chinese-bilingual models.
Competitive with DALL-E 3 / SDXLarxiv.org β†—

API Usage Examples

Use the global api.therouter.ai endpoint shown below for new integrations; the legacy China accelerated endpoint is retired.

Recommended: use the async API

Image generation typically takes 30–180s, beyond the edge sync timeout. The examples below use the ?async=true submit + poll pattern. Read the full async image generation & edit guide β†’

cURL
# 1) Submit job (returns 202 immediately with a polling URL).
# Image generation takes 30-180s β€” always use the async path in production.
JOB=$(curl -s -X POST "https://api.therouter.ai/v1/images/generations?async=true"   -H "Content-Type: application/json"   -H "Authorization: Bearer $THE_ROUTER_API_KEY"   -d '{
    "model": "doubao/doubao-seedream-4-5",
    "prompt": "A cinematic product render with soft studio lighting"
  }' | python3 -c "import sys,json;print(json.load(sys.stdin)['id'])")
echo "submitted: $JOB"

# 2) Poll until terminal (succeeded / failed / cancelled / expired).
while :; do
  R=$(curl -s "https://api.therouter.ai/v1/jobs/$JOB"     -H "Authorization: Bearer $THE_ROUTER_API_KEY")
  S=$(echo "$R" | python3 -c "import sys,json;print(json.load(sys.stdin)['status'])")
  echo "status: $S"
  case "$S" in
    succeeded) echo "$R" | python3 -c "import sys,json;print(json.load(sys.stdin)['unsigned_urls'][0])"; break ;;
    failed|cancelled|expired) echo "$R"; exit 1 ;;
  esac
  sleep 5
done

API guide

Text-to-image generation

Generate images using the OpenAI-compatible images endpoint. Specify prompt, size, and format.

cURL
curl https://api.therouter.ai/v1/images/generations \
  -H "Authorization: Bearer $THEROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "doubao/doubao-seedream-4-5",
    "prompt": "A modern minimalist cafe interior, warm lighting, Japanese aesthetic, 4K detail",
    "size": "1024x1024",
    "n": 1,
    "response_format": "url"
  }'

Image Editing Examples

Upload an image and describe the edit you want with a text prompt; the model returns the edited image as base64.

cURL
# Same async submit + poll pattern as /v1/images/generations.
JOB=$(curl -s -X POST "https://api.therouter.ai/v1/images/edits?async=true"   -H "Authorization: Bearer $THE_ROUTER_API_KEY"   -F "model=doubao/doubao-seedream-4-5"   -F "prompt=Turn this scene into a watercolor painting"   -F "size=1024x1024"   -F "image=@input.png" | python3 -c "import sys,json;print(json.load(sys.stdin)['id'])")
echo "submitted: $JOB"

while :; do
  R=$(curl -s "https://api.therouter.ai/v1/jobs/$JOB"     -H "Authorization: Bearer $THE_ROUTER_API_KEY")
  S=$(echo "$R" | python3 -c "import sys,json;print(json.load(sys.stdin)['status'])")
  echo "status: $S"
  case "$S" in
    succeeded) echo "$R" | python3 -c "import sys,json;print(json.load(sys.stdin)['unsigned_urls'][0])"; break ;;
    failed|cancelled|expired) echo "$R"; exit 1 ;;
  esac
  sleep 5
done

More from doubao

Similar models

Cross-provider sibling models

News & changes

2025-12-03

Seedream 4.5 launches on Volcengine Ark with unified generation + editing

ByteDance's Team Seedream released version 4.5 of their image model, consolidating text-to-image and multi-reference editing into one architecture. The model supports up to 4MP output, sequential image generation for multi-frame consistency, and improved Chinese/English text rendering β€” making it especially relevant for e-commerce and marketing creative workflows routed through TheRouter.

re-authored by TheRouterByteDance Seed β†—

Frequently asked

What is the difference between Seedream 4.5 and Seedream 5.0?

Seedream 5.0 is the direct successor with improved image fidelity and detail preservation. Seedream 4.5 remains available at a slightly higher per-image price ($0.0347 vs $0.0306) and is still a strong choice for workflows that rely on its specific sequential generation and editing capabilities.

re-authored by TheRouterseed.bytedance.com β†—
Does Seedream 4.5 support image editing (img2img)?

Yes. Seedream 4.5's unified architecture handles both text-to-image generation and reference-based image editing in the same model. You can pass input images alongside text prompts to perform style transfer, element replacement, or multi-image fusion while preserving subject identity.

re-authored by TheRouterseed.bytedance.com β†—
What resolution and aspect ratios does Seedream 4.5 support?

The model generates images up to 4 megapixels total (max 2048Γ—2048). Flexible aspect ratios are supported β€” common options include 1:1, 16:9, 9:16, 3:4, and 4:3. Specify desired dimensions via the 'size' parameter in your API call.

re-authored by TheRouterfal.ai β†—
How does text rendering in Seedream 4.5 compare to other image models?

Seedream 4.5 is specifically optimized for Chinese and English typography within generated images β€” posters, product cards, and banners with embedded text are a core strength. For Western-language-only dense text scenarios, FLUX Kontext models may edge ahead, but Seedream 4.5 leads for CJK text rendering quality.

re-authored by TheRouterseed.bytedance.com β†—
Fact ledger β€” every claim on this page traces here
sourceURLretrieved
Release dateseed.bytedance.com β†—2026-05-26verified
Parametersdocs.apiyi.com β†—2026-05-26to verify
Max output resolutionfal.ai β†—2026-05-26verified
Architectureseed.bytedance.com β†—2026-05-26verified
Batch capabilityfal.ai β†—2026-05-26verified
Prompt languagesseed.bytedance.com β†—2026-05-26verified
Model versionwww.volcengine.com β†—2026-05-26verified
Licensewww.volcengine.com β†—2026-05-26to verify
GenEvalarxiv.org β†—2026-05-26to verify
T2I-CompBencharxiv.org β†—2026-05-26to verify
Seedream 4.5 launches on Volcengine Ark with unified generation + editingByteDance Seed β†—2026-05-26verified
What is the difference between Seedream 4.5 and Seedream 5.0?seed.bytedance.com β†—2026-05-26to verify
Does Seedream 4.5 support image editing (img2img)?seed.bytedance.com β†—2026-05-26to verify
What resolution and aspect ratios does Seedream 4.5 support?fal.ai β†—2026-05-26to verify
How does text rendering in Seedream 4.5 compare to other image models?seed.bytedance.com β†—2026-05-26to verify
Help & contact