Skip to main content

Seedream 5.0 Flash layer decomposition splits one finished image into a clean base plate plus up to sixteen named, transparent PNG layers, each with its position. Resolution is auto, 1K, 1.5K or 2K, and unfiltered source images are handled like any other.

From $0.026/image

Workbench

Input

Layer decomposition
0/5000
Optional, but at least 3 characters when you write one. Leave it empty and the model picks the main elements itself; name elements and it splits those. You can pin an element to a place with `<bbox>x1 y1 x2 y2</bbox>`, using coordinates normalized to 0-1000 with the top-left corner as the origin. Example: `Separate the title text <bbox>179 58 809 197</bbox> and the parrot <bbox>330 274 641 991</bbox> into independent layers`. No effect on price. at least 3 characters · at most 5000 characters
1/1

Drop files or media links, paste, or browse · 1 file · Max 10 MB · JPG · PNG · WEBP · GIF

  • 1
1 / 1 files
One image to split: png, jpeg, webp, bmp, tiff or gif, up to 30 MB, 262,144 to 36,000,000 total pixels, aspect ratio between 1:16 and 16:1. HEIC is not supported. You must hold the rights to the image. No effect on price. at least 1 characters
auto, 1k, 1.5k or 2k. The base plate keeps the input's aspect ratio and each layer keeps its element's proportions. The price is per output image and does not depend on the tier.
Output format
JPEGPNG
jpeg or png for the base plate only. Layers are always transparent PNG whichever you choose. It changes the delivered base-plate file, not the price.
Safety checkerOff

SpicyAPI does not run a safety checker on this model. Your prompt and media go to the model as written, and the model itself decides what comes back.

Checking your account before enabling submission and pricing.

Output

Ready for a new run
flash-layers-late-shiftExample result

Related models and next tasks

Missing the model you need?

Tell us what to add and where you would use it. Requests go to the people who decide what gets onboarded next.

Pull a finished image apart into a clean plate and transparent cut-outs

Seedream 5.0 Flash Layer Decomposition takes one finished image and returns a base plate, with the lifted elements removed and the background filled in, plus up to sixteen transparent PNG layers, one per element. Leave the prompt empty and it picks the elements itself; our retro poster came back as eight images. Every layer comes with a name and a position box, so the pieces stack back exactly where they sat.

Open the playground
Pull a finished image apart into a clean plate and transparent cut-outs
Input
1 image up to 30 MB; HEIC not supported
Output
Base plate + up to 16 transparent PNG layers (17 images max)
Layer data
A name and a position box in base-plate pixels for every layer
Prompt
Optional; at least 3 characters when written; <bbox> targeting
Billing
Per output image, base plate included; 17 reserved, difference refunded
Content policy
Sources at every content level accepted, Very strong included
Automatic split
An empty prompt still finds the piecesWith no prompt the model chose eight images from our film poster: the repaired base plate plus seven named layers, the diner, a distant car, the red neon sign, the woman in the red trench coat, the tagline, the title and the ticket stub.
Positions and names
Every layer knows where it belongsEach layer returns a name and a boundingBox of x1, y1, x2, y2 in base-plate pixels. On the web the layers stack back over the plate at their positions, each downloadable, with one button for the whole set.
Pay per piece
The base plate counts as one imageA request reserves the cost of seventeen images and refunds the difference once the real count is known. Naming only the elements you need gives fewer pieces and a smaller bill: two named elements on a ballroom scene returned three images.

What the Seedream 5.0 Flash layer decomposition API gives you

A repaired plate and one PNG per element

A repaired plate and one PNG per element

The base plate is the picture with every lifted element removed and the gap filled in: our poster's plate was an empty rainy street, with the diner, the woman and all the lettering gone. Each element arrives as its own transparent PNG. Layers are always PNG whatever output_format says; that setting only changes the plate.

Ask for exactly the pieces you want

Ask for exactly the pieces you want

Name elements in plain words and the split follows the list. "The dancing couple and the chandelier" on the ballroom scene returned three images: the empty hall, the couple as one layer and the chandelier as another. Fewer, deliberate pieces are easier to reuse and cost less than an automatic split.

  • Point at an element with coordinates

    Add <bbox>x1 y1 x2 y2</bbox> after an element's name to say which one you mean, using 0–1000 coordinates from the top-left corner, so the numbers do not depend on the image's pixel size. It helps when two similar objects share a frame and a name alone is ambiguous.

  • Stack them back at their boxes

    The first image is always the plate and the rest follow from bottom to top. Scale each layer into its boundingBox over the plate and the original returns; or use a layer where position does not matter, such as a product on a new background or a subject in a new template.

  • Every content level splits too

    Posters, ads, illustrations and portraits are split as supplied at every content level we tried, including the Very strong level that Seedream 5.0 Pro Layer Decomposition refuses. SpicyAPI adds no content filter here; the model's own limits still apply.

  • One click from a generated image

    On the web, a finished image from the playground can go straight into a split, and the layers open as checkerboard tiles with a download button on each and one for the whole set. Through the API the same result is an ordered list of assets.

Where Seedream 5.0 Flash layer decomposition fits

  • Rewrite the headline, keep the artwork

    Poster and ad refresh

    Rewrite the headline, keep the artwork

    Lift the title layer, set new copy in your own typeface and place it over the base plate. The picture underneath is untouched, so one render can carry a campaign across several languages or dates.

  • Drop a subject into another layout

    Template reuse

    Drop a subject into another layout

    The woman from the poster arrives as a transparent PNG with clean edges, ready for a flyer, a social tile or a storyboard. Clear the rights to anything you reuse, especially people.

  • Separate planes for parallax

    Motion and depth

    Separate planes for parallax

    A flat illustration becomes ordered pieces: background plate, midground props, foreground figure. Offset them at different speeds for a parallax move, or hand them to an animator as the starting point for a rig.

  • An empty set to build on

    Compositing

    An empty set to build on

    Keep only the base plate and add your own elements. The poster split left a clean, rain-soaked street; the ballroom split left an empty hall under its lights.

How to call the Seedream 5.0 Flash layer decomposition API

  1. Send one finished image

    image_url takes a single image up to 30 MB, between 262,144 and 36 million pixels and no more extreme than 16:1. HEIC is not supported, so convert phone photos to JPEG first.

  2. Try it with no prompt first

    An automatic split shows how the model reads the picture. If you do write a prompt, it needs at least three characters.

  3. Then name only what you need

    Use plain names, adding <bbox> coordinates when two similar objects share the frame. Fewer named elements mean fewer images billed.

  4. Submit and poll for a terminal state

    Flash splits took 45 to 98 seconds in our runs. The quote reserves seventeen images, and the rest is refunded when the real count is known.

  5. Read the assets in order

    The first asset is the plate (zIndex 0) and the layers follow from bottom to top, each with a name and a boundingBox. Scale each layer into its box to rebuild the image.

Seedream 5.0 Flash layer decomposition worked examples

  • An automatic split of a film poster

    No prompt and resolution auto: eight images, with the street repaired behind everything that was lifted out. The title, the tagline and the ticket stub each came back as a separate named layer with its own box.

  • Two named elements from a ballroom

    A prompt naming just these two, at 2k, returned three images: the empty hall, the couple and the chandelier.

    Separate the dancing couple and the chandelier into independent layers
  • Pinning elements with coordinates

    Coordinates are 0–1000 from the top-left corner and go straight after the element's name.

    Separate the title text <bbox>179 58 809 197</bbox> and the parrot <bbox>330 274 641 991</bbox> into independent layers

Seedream 5.0 Flash layer decomposition questions

Start with no prompt to see how the model reads the picture, then name only the elements you need, since every returned image, the plate included, is billed. Every layer carries a name and a box, so rebuild by zIndex, scaling each PNG into its box, or lift a single piece as a cut-out. Sources at every content level are split as supplied, Very strong included.