Seedream 5.5 Layerize (ByteDance)
Decompose one image into a base composite plus up to 99 transparent PNG layers. Includes stacking order, bounding boxes, and optional OCR metadata for text layers.
## Overview
Seedream 5.5 Layerize by ByteDance splits one finished image into a base image and separate transparent layers. Each layer keeps its original position and stacking order, so the image can be edited, animated or rebuilt one element at a time. A prompt can name the objects to separate; without one, the model splits out every element it finds. Optional text recognition reads the printed text in text layers.
## What you can build
- Product ad variants: lift products, props and text off a finished ad and rearrange them for new sizes
- Motion graphics: animate each element of a still image separately
- Parallax scenes: move foreground and background layers at different speeds
- Editable designs: turn a flat poster or banner back into a layer stack
- Text extraction: read the wording, position and font style of printed text in a design
## Inputs
- Input image: one image, at least 262,144 pixels in total (for example 512x512), with an aspect ratio between 1:16 and 16:1, up to 30 MB
- Max layers: the most images returned, counting the base image, from 2 to 100
- Prompt (optional): which objects to put on their own layers
- Resolution: 1K, 1.5K or 2K size for the returned images
- Text recognition (optional): reads the text in text layers
- Output format (optional): png or jpeg for the base image
- Watermark (optional): adds an "AI generated" mark
## Outputs
- A base image with the separated elements removed
- One transparent png per separated element
- A layers file listing each layer's order, size, position, name and description, plus the recognized text when text recognition is on
## Recommended settings
- Name the objects you want in the prompt and set max layers to their count plus one for the base image
- Use 1K for quick tests and 2K when the layers will be used at full size
- Turn on text recognition only for designs with printed text
## Limitations
- The model may return fewer layers than max layers
- Fine edges such as hair, glass, smoke and motion blur may not separate cleanly
- Text recognition works best on clear printed text
- Low-quality or very small images give rougher layers
## Safety & compliance
- ByteDance's content checks run on every request and may block some images or prompts
- Only upload images you have the rights to use
Example prompts
Great starting points for seedream 5.5 layerize.
API quick start
Run seedream 5.5 layerize with a single API call.
{
"prompt": "Separate the strawberry, the lemon, the k…",
"inputImage": "https://your-cdn.com/input.png",
"maxLayers": 4,
"resolution": "1k"
}curl -X POST "https://api.wiro.ai/v1/Run/bytedance/seedream-5-5-layerize" \
-H "Content-Type: application/json" \
-H "x-api-key: YOUR_WIRO_API_KEY" \
--data-binary @- <<'JSON'
{
"prompt": "Separate the strawberry, the lemon, the k…",
"inputImage": "https://your-cdn.com/input.png",
"maxLayers": 4,
"resolution": "1k"
}
JSONOther models you might like
Explore similar image generation models.
