Updated
What is SeeDream 5.5?
SeeDream 5.5 is ByteDance's newest image model, released November 9, 2026. It generates images from a prompt, edits images you send as references, and is the first SeeDream model on our API that returns a PNG with a real transparent background.
ByteDance built this release around edits that stay where you put them. A change lands on the part you describe, and the rest of the picture keeps its pixels and colors, so an image can go through many rounds without drifting. Realism moved as well. Lighting, materials, and skin look less glossy and less overexposed than before, and small, dense text comes out cleaner.
The request is short: a prompt, up to 10 reference images, one of eight aspect ratios from 1:1 to 21:9, and output at 1K, 1.5K, or 2K in png or jpeg. One image comes back per call, typically in under a minute.
What's new in SeeDream 5.5
Transparent backgrounds in one call
Set transparent_background to true and the result is a PNG with a real alpha channel. There is no cutout step and no background remover to run afterwards, and the option adds nothing to the price. It needs png output, so a jpeg request with transparency is rejected.
Edits that leave the rest alone
Describe one change and the model keeps the pixels outside it: same face, same colors, same background. That is what makes long edit chains workable. Each round is its own request, so send the last output back in image_urls with the next instruction.
More natural light, materials, and skin
Photoreal scenes lose the plastic shine. Skin has texture, metal and fabric reflect the way they should, and highlights hold detail without blowing out.
Small, dense text that reads
Labels, menus, packaging copy, and poster body text come out cleaner than on earlier SeeDream models, in several languages. Layouts keep their spacing when the text gets long.
Up to 10 reference images
Send up to 10 images in image_urls to swap a product, restyle a photo, or merge several subjects into one scene with matching scale and light. Leave the field out and the same route does plain text-to-image.
Eight aspect ratios, 1K to 2K
Pick 1K, 1.5K, or 2K and a ratio from 1:1 to 21:9. The two set the pixel size together: 2K at 1:1 renders 2048x2048, 2K at 21:9 renders 3072x1312, and 1.5K at 16:9 renders 1968x1104.
What SeeDream 5.5 can do
A cutout you can drop on any background
A cut-glass perfume bottle with smoke rising from the cap, returned with a real transparent background. The glass edges and the thin smoke keep their soft transparency, so the bottle sits cleanly on the plain blue backdrop we placed behind it for this page.
One change, everything else untouched
The portrait below went back in as a reference with one instruction: swap the blue sweater for a red wool coat. Her face, her laugh, the apron, the fish on the ice and the market behind her all stay as they were.
Label text down to the ingredient list
A hot-sauce label with a brand name, an English and a Japanese product line, a nine-item ingredient list and a five-row nutrition table. Every word and number came out as written in the prompt, at the size real packaging print uses.
Skin and light that look photographed
A fishmonger laughing behind an ice counter at dawn. The window light, the lines around her eyes, the wool of the sweater and the wet scales on the ice all read as one real moment, with no waxy skin.
What SeeDream 5.5 is best for
Design assets with no background
Stickers, icons, product cutouts, logos, and game sprites that drop into a layout or an app without a masking pass.
Step-by-step editing tools
Editors where a user makes one change at a time and expects everything else in the image to stay as it was.
Product and storefront images
Background swaps, restyles, and composites built from a product shot, a logo, and a scene reference.
Realistic people and lifestyle shots
Portraits, fashion, and interiors where skin, fabric, and light have to pass as a photo.
Posters, packaging, and labels
Layouts with a headline plus small supporting text that still has to be readable.
SeeDream 5.5 use cases
Build a sticker or icon generator that returns transparent PNGs a user can drag straight onto a canvas. Add a chat-style editor to a photo app: every message becomes one request with the last image attached, and the user watches a picture change one step at a time. Run a storefront flow where a merchant uploads a product shot and gets it back on a clean transparent background, then placed into three lifestyle scenes from the same references. Or produce packaging mockups where the ingredient list and the fine print stay readable at 2K.
SeeDream 5.5 limitations
Output stops at 2K on this route, with no 3K or 4K option. Each request returns one flat image: there is no sequential set mode, no web search, and no layer separation, and extra fields in the request are rejected. Transparency needs png output. The API keeps no memory between requests, so multi-round editing means sending the previous output back as a reference each time. Reference images are billed per image, so a ten-image composite costs more than a text-to-image call.
When to use SeeDream 5.5
Use SeeDream 5.5 for new work that involves editing, transparent assets, or realistic people and products. It is the newest model in the family and the only one here with alpha output. Use SeeDream 5.0 Flash when cost per image decides the job: it charges one flat rate at every resolution and is still the cheapest SeeDream model per image. SeeDream 5.0 Pro remains a sound choice for pipelines already tuned on it, and SeeDream 5.0 Lite covers sequential sets and web-search references.
SeeDream 5.5 API examples
Call SeeDream 5.5 from any language by POSTing to /v1/tasks. Full parameter docs live at docs.unifically.com/models/image/bytedance/seedream-5.5.
curl -X POST https://api.unifically.com/v1/tasks \
-H "Content-Type: application/json" \
-H "Authorization: Bearer YOUR_API_KEY" \
-d '{
"model": "bytedance/seedream-5.5",
"input": {
"prompt": "A glass perfume bottle with a gold cap, studio product shot",
"aspect_ratio": "1:1",
"resolution": "2K",
"output_format": "png",
"transparent_background": true
}
}'
Successful submission returns a task_id. Poll GET /v1/tasks/<task_id> or set a callback_url on the request to receive the finished image.
FAQs
People also ask about SeeDream 5.5
SeeDream 5.5 is ByteDance's newest image model, released November 9, 2026. It generates images from a text prompt, edits images you send as references, and can return a PNG with a transparent background. All three run on the same route.
Three changes stand out. Edits stay on the part you describe while the rest of the image keeps its pixels and colors. The model can output a transparent background as a PNG with a real alpha channel. Lighting, materials, skin, and small text all look more natural.
Yes. Set transparent_background to true and keep output_format on png. The image comes back as a PNG with a real alpha channel, ready to place over any background. A jpeg request with transparency is rejected, and the option adds nothing to the price.
$0.018 per image at 1K and 1.5K, and $0.0315 at 2K. Each reference image you attach adds a small charge, listed on the pricing page. A transparent background costs nothing extra. Billing is per finished image with no subscription.
Three resolution options, 1K, 1.5K, and 2K (the default), across eight aspect ratios, 1:1, 4:3, 3:4, 16:9, 9:16, 3:2, 2:3, and 21:9. The two combine into the final pixel size, so 2K at 16:9 renders 2624x1472 and 2K at 21:9 renders 3072x1312.
Send each round as its own request. Put the previous output in image_urls, describe the next change in the prompt, and repeat. The API keeps no memory between requests, so the image you send is the full context.
Most images finish in 15 to 30 seconds, measured on the UnificAlly API. A 2K image or a busy period can stretch that to about two minutes, so set your timeout with that in mind.
No. Output stops at 2K, each request returns one flat image, and layer separation is not available. Transparent PNG output is the way to get an asset you can place into a layered design.