ByteDance Seedream 5.5 Layers
Seedream 5.5 splits one picture into a base image and transparent layers, each with its name and position. From $0.02 per image made.
Seedream 5.5 Layers takes one picture and splits it into a base image and independent layers: the subjects, the text and the other pieces of the picture, each a PNG that is transparent around the piece. The pieces are taken out of the base image, which keeps the rest of the picture with the background filled in behind them. Use it to turn a poster, a banner, a product shot or an illustration into pieces you can move, restyle or replace one at a time, to localize the text of a design, or to cut the objects out for a collage or a sticker pack.
You get image, a list of files with the base image first and then the layers in stacking order, and layers, a list in the same order that gives each image's z_index (0 for the base image), its background (opaque for the base image, transparent for a layer) and, for a layer, its name, a description and its bounding_box on the base image: absolute pixels [left, top, right, bottom] and normalized 0-1000. Scale a layer to its box, place it there and stack the layers by z_index to put the picture back together. With text_recognition on, a text layer also carries its text, its lines with their positions and the approximate style of each run (font, colour, bold, italic).
Inputs: image (required: one PNG or JPEG, 512x512 to 6000x6000 pixels, aspect ratio between 1:16 and 16:1, up to 30 MB; a link or an upload), prompt (optional: name what to split out, such as "each animal on its own layer" or "the headline, the date and the saxophone"; <bbox>left top right bottom</bbox> with values 0-1000 marks a region; leave it out to let the model decide), max_images (12 by default, 2-100: the most images a run may make, the base image included; a larger number splits the picture finer), size (1K by default, 1.5K, 2K or auto: the resolution of the base image and of each layer), text_recognition (off by default) and watermark (off by default).
The model decides how many layers a picture has, up to max_images: a picture with a few pieces comes back with a few layers, and you cannot ask for an exact number. It does not generate or edit pictures (use Seedream 5.5 for that), takes one picture and returns PNG layers, not a layered file. If one layer fails, the run fails as a whole and costs nothing. A run takes about 15-25 seconds for a handful of layers.
Price: every image made is billed, the base image and each layer, at $0.02 at 1K or 1.5K and $0.035 at 2K, and the image you give adds $0.008. A base image and 3 layers at 1K cost $0.088; at 2K, $0.148. Your balance must cover the most a run can cost before it starts: max_images images at the rate of the size (12 images at 1K: $0.248). With size auto each image is billed at the rate of its own size, which follows the size of your picture. A refused or failed run costs nothing.
- Image to Image
Price: From $0.02 per image, depending on size; every image is billed, the base image and each layer, and the image you give adds $0.008: a base image and 3 layers at 1K cost $0.088.
Run ByteDance Seedream 5.5 Layers via API
curl -X POST https://api.flymy.ai/api/v1/flymyai/bytedance-seedream-5_5-layers/predict \
-H "X-API-KEY: $FLYMYAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{"image":"https://images.unsplash.com/photo-1470770841072-f978cf4d019e?w=1024&q=80&fm=jpg","prompt":"The mountains, the lake and the sky, each on its own layer","max_images":4,"size":"1K"}'API documentation · Create a free account · Pricing