Skip to main content
Image to Image

Seedream v4.5 Image Editor

Edit images using natural language instructions with ByteDance Seedream v4.5 model

View details

Inputs

Loading input fields...
Execution Steps

Loading workflow structure...

Loading curated examples...

Overview

Seedream v4.5 Image Editor edits one to ten source images from natural-language instructions and returns edited images. Use it for product, campaign, social, or brand-asset revisions when you need multi-reference image editing with up to six outputs, preset or custom sizing, safety-checker, and optional seed controls.

Use cases

  • Use several product or brand images to guide a single edited composition.
  • Revise campaign creative while preserving key product, style, or scene details.
  • Generate multiple edited directions from one brief for comparison.
  • Choose preset or custom sizing when the final layout needs a larger canvas.

Input tips

  • Provide 1-10 public image_urls that can be fetched without login.
  • Use a prompt of 3-5,000 characters that names desired changes and fixed details.
  • Set num_images and max_images from 1-6 when you want several options returned.
  • Choose an image_size preset or custom width and height from 1920 to 4096 pixels.
  • Leave the safety checker enabled unless your use case needs a reviewed exception.
  • Use seed for repeatable generation inputs, then review outputs visually.

Expected output

The AI Tool returns edited images with downloadable URLs and optional content type, file name, file size, width, and height, plus cost metadata. The shared Seedream output view renders a single image or gallery and shows image count and the Seedream v4.5 Image Editor model label. The v4.5 edit output does not include seed.

Caveats

  • Input image URLs must be public and reachable.
  • Multi-image edits depend on prompt clarity about each reference's role.
  • Custom image sizes must use width and height from 1920 to 4096 pixels.
  • The v4.5 edit result does not return seed even when seed is supplied.
  • Safety filtering can alter or block some requests.
  • Generated text, logos, product details, faces, and image edges should be reviewed before use.