Bytedance Seedream V4.5 Edit API Documentation

Bytedance Seedream V4.5 Edit API Documentation

Playground

Seedream 4.5 Edit preserves facial features, lighting, and color tone from reference images, delivering professional, high-fidelity edits up to 4K with strong prompt adherence. Ready-to-use REST inference API, best performance, no cold starts, affordable pricing.

Features

Seedream 4.5 Edit is ’s high-end image editing model. It preserves facial structure, lighting, and colour tone from your input while applying precise, prompt-driven changes that look like they were done by a professional retoucher—not an over-filtered AI.

Model highlights

• Reference-faithful editing – Keeps identity, pose, lighting, and palette close to the original image.
• Multi-image support – Upload 1–10 images and apply the same edit prompt across the whole batch.
• High-fidelity details – Crisp textures, clean edges, and minimal artifacts even on complex edits.
• Strong prompt adherence – Understands nuanced edit instructions for clothing, background, mood, and style.
• 4K-ready output – Supports resolutions up to 8192 × 8192 when you need extra detail.

Recommended use cases

• Portrait retouching and beauty / fashion edits
• Outfit, hairstyle, and accessory changes
• Background replacement or scene upgrades
• Product visual updates and colourway exploration
• Interior / architecture visual tweaks
• Brand KV and campaign asset iterations

Pricing

Flat, simple pricing:

• $0.04 per generated image

Each successfully edited output counts as one image.

How to use

1. Upload images Add 1–10 images under images. These are the photos that will be edited.

2. Write your edit prompt Clearly describe what should change and what must stay the same, e.g. “Change jacket to red leather, keep pose and background, cinematic lighting.”

3. (Optional) Set size

• Choose width and height (e.g. 1024×1024, 2048×2048, up to 8192×8192).
• If you leave size empty, the model uses its default resolution.

4. Run Submit the job. All uploaded images will be edited using the same instructions.

Prompt patterns (copy & adapt)

Outfit / accessory change

> Change outfit to {DESCRIPTION}, keep pose and lighting, preserve skin tone, clean edges, no distortions.

Background replacement

> Replace background with {SCENE}, match original light direction, natural shadows, no haloing around subject.

Style / mood change

> Apply {STYLE} colour grading, keep subject identity and composition, subtle filmic contrast, rich details.

Product recolour

> Change product colour to {HEX / COLOUR NAME}, keep material and reflections, no change to logo.

Notes

• Higher input resolution generally yields better detail; for heavy crops, consider upscaling first.
• For best results, keep edit prompts specific but focused—avoid mixing too many unrelated changes at once.
• All image URLs must be publicly accessible for the model to load them.

Model comparison on WaveSpeedAI

Use Seedream 4.5 Edit together with other editing models depending on what you’re optimising for:

• /seedream-v4.5/edit-sequential – multi-image editing that keeps the same character and style across a whole batch; best when you need a series of consistent portraits, product shots or KV variations.

• wavespeed-ai/flux-2-dev/edit – lightweight FLUX.2 edit model for fast, low-cost batch updates; great for everyday asset tweaks where ultimate fidelity is less critical.

• flux-2-flex/edit – more style-rich and adventurous editing; good for creative restyling, mood shifts and art-direction heavy work rather than strict realism.

• flux-2-pro/edit – production-grade FLUX.2 editing with multi-reference and context-aware control, suited to complex commercial composites and e-commerce pipelines.

• qwen-image/edit-plus – strong semantic understanding and local control, ideal for masking-style edits, UI mockups, and precise object-level changes.

Authentication

For authentication details, please refer to the Authentication Guide.

API Endpoints

Submit Task & Query Result

set -euo pipefail

export WAVESPEED_API_KEY="your-api-key"

REQUEST_BODY=$(cat <<'JSON'
{
  "prompt": "A cinematic ocean wave at sunrise, highly detailed",
  "images": [
    "https://interactive-examples.mdn.mozilla.net/media/cc0-images/painted-hand-298-332.jpg"
  ]
}
JSON
)

# 1. Submit the prediction once. Never retry this request automatically:
# a retry after an unclear failure can create a second, billed task.
CHECK='The task may already exist. Check History or the predictions list before submitting again.'
if ! SUBMIT_RESPONSE=$(curl --silent --show-error --fail-with-body \
  --connect-timeout 10 --max-time 60 \
  -X POST "https://api.wavespeed.ai/api/v3/bytedance/seedream-v4.5/edit" \
  -H "Authorization: Bearer ${WAVESPEED_API_KEY}" \
  -H "Content-Type: application/json" \
  -d "${REQUEST_BODY}"); then
  printf 'Submit failed. %s\n' "${CHECK}" >&2
  exit 1
fi

TASK=$(printf '%s' "${SUBMIT_RESPONSE}" | jq 'if type == "object" and has("data") then .data else . end')
PREDICTION_ID=$(printf '%s' "${TASK}" | jq -r '.id // empty')
if [ -z "${PREDICTION_ID}" ]; then
  printf 'Submission response did not contain a prediction id. %s\n' "${CHECK}" >&2
  exit 1
fi
RESULT_URL="https://api.wavespeed.ai/api/v3/predictions/${PREDICTION_ID}/result"

# 2. Poll until the prediction reaches a final status.
BODY_FILE=$(mktemp)
trap 'rm -f "${BODY_FILE}"' EXIT
while true; do
  # Transient errors (429, 5xx) are retried, honoring Retry-After.
  HTTP_CODE=$(curl --silent --show-error --fail-with-body \
    --connect-timeout 10 --max-time 60 --retry 5 --retry-delay 2 \
    --output "${BODY_FILE}" --write-out '%{http_code}' \
    "${RESULT_URL}" \
    -H "Authorization: Bearer ${WAVESPEED_API_KEY}" || true)
  case "${HTTP_CODE}" in
    2??) ;;
    000|429|5??)  # dropped connection or temporary error: keep polling
      printf 'Task %s: HTTP %s, polling again in 10s\n' "${PREDICTION_ID}" "${HTTP_CODE}" >&2
      sleep 10; continue ;;
    *) printf 'Task %s: HTTP %s\n' "${PREDICTION_ID}" "${HTTP_CODE}" >&2
      cat "${BODY_FILE}" >&2; exit 1 ;;
  esac
  RESULT=$(jq 'if type == "object" and has("data") then .data else . end' "${BODY_FILE}")
  STATUS=$(printf '%s' "${RESULT}" | jq -r '.status // empty')

  case "${STATUS}" in
    completed) printf '%s\n' "${RESULT}" | jq '.outputs'; break ;;
    failed|cancelled|timeout|deleted)
      printf 'Task %s ended with status %s\n' "${PREDICTION_ID}" "${STATUS}" >&2
      printf '%s\n' "${RESULT}" | jq . >&2; exit 1 ;;
    *) sleep 2 ;;
  esac
done

Parameters

Task Submission Parameters

Request Parameters

Every model also accepts the global request options enable_sync_mode and enable_base64_output; see their pages for details.

Response Parameters

Result Request Parameters

Result Response Parameters