Coming soon to Runware

Seedream 5.5 is coming to Runware

Seedream 5.5 is ByteDance's new image model for realistic generation and consistent multi-turn editing, with native transparent backgrounds and layer decomposition. It's coming to the Runware API soon.

A black and gold race car with Seedream livery throwing spray down a wet track

Generate, edit, and take apart.

Every workflow will run through one imageInference request, with the same key and billing as every other model on Runware.

  • Text-to-image
  • Up to 10 reference images
  • Multi-turn editing
  • Transparent backgrounds
  • Layer decomposition
  • Text recognition

Edit again and again, and keep the rest

Make one targeted change per turn. Seedream 5.5 applies it and keeps every unedited region as it was, so a scene can evolve over many edits without drifting. These are ByteDance's own demos.

  • 0:00

    A study, refurnished one request at a time

    Rug, shelves, colors, and decor change across turns while the room's layout and light stay put.

  • 0:00

    One face, six looks

    Hair, makeup, and wardrobe change with each edit, and the face stays the same person.

  • 0:00

    A still life, built up from a stone

    Each turn adds one element and leaves everything already placed where it was.

Transparent from the start

Generate artwork on a transparent or semi-transparent background, then reuse it on any product or scene without cutting it out first.

  • 0:00

    A character on a transparent layer

    Generated as a cutout, then reused as a keychain, a towel, and a sticker.

  • 0:00

    One flamingo, many products

    A transparent illustration placed on luggage, a robe, a patch, and a pin.

  • 0:00

    Whole sticker packs

    Dozens of separate elements on one transparent sheet, ready to cut and arrange.

Take a flat image apart

Set settings.layerDecomposition and Seedream 5.5 splits a finished image into editable layers, so a generated poster can go straight into a design tool.

0:00

ByteDance's demo: a finished snowy owl infographic comes apart into the owl, its callouts, and each text block as separate transparent layers.

Input
One reference image of at least 0.26 megapixels
Output
A base image plus up to 99 transparent PNG layers
Per layer
A name, a description, and a bounding box in pixels
Text layers
The text with its position, font, and style, when text recognition is on

Lifelike by default

ByteDance's own Seedream 5.5 stills: natural light, fine materials, and refined skin textures, from street portraits to landscapes.

  • A laptop on a blue rug showing a portrait, with a lime scarf, a pink gerbera, and a chrome flower brooch across the keysA portrait on screen, styled in front of it
  • A weathered stone bust against a blue sky with white flowers growing from the top of its headA stone bust in bloom
  • A skateboarder in midair over the lip of a concrete ramp, shot from below against the sunAirborne over the ramp
  • Jagged mountain peaks lit orange above a glacier and a still blue lakeA glacier lake under red peaks
  • A long-haired white and ginger cat sitting on a bar counter among bottles and glassesA cat among the bottles
  • A father asleep on a sofa with a sleeping baby on his chest, in soft window lightAsleep on the sofa
  • A woman with windswept hair checking her phone on a busy city streetA street portrait in the wind
  • A donkey looking into the camera through a gap between two weathered wooden boardsA donkey between the boards
  • Commuters walking along a crowded subway platform under a lit departure signRush hour on the platform

Set text, legible layouts

Posters, character sheets, and infographics with readable headlines and labels, in English and Chinese. All of them are ByteDance's own Seedream 5.5 outputs.

A green and purple screen-print poster reading Print Play, with smaller slogans around a printing press
A screen-print event poster
A character sheet titled Berry Club showing a blue mascot in many poses, expressions, and accessories
A mascot character sheet
An illustrated infographic titled From Seed to Salad, showing sow, sprout, tend, grow, harvest, and share steps
A step-by-step infographic
A bilingual Chinese and English infographic of a blue morpho butterfly with eight labeled close-ups
A bilingual nature infographic

What one request can do

Seedream 5.5 on Runware, from its schema.

01

Text-to-image and editing

Generate from a prompt of up to 10,000 characters, or edit with 1 to 10 reference images.

02

1K, 1.5K, or 2K

Or any width and height from 0.92 to 4.62 megapixels, at aspect ratios from 1:16 to 16:1.

03

Transparent output

Set settings.background to transparent for a PNG or WEBP cutout, or leave it on auto.

04

Layer decomposition

One image in, a base image plus up to 99 PNG layers out, each with a name, a description, and its position.

05

Text recognition

For layers recognized as text, get the text itself with its position, font, and style.

06

Up to 20 images per request

Set numberResults from 1 to 20. JPG, PNG, or WEBP output.

Built for design and production work

Where Seedream 5.5's editing consistency, transparency, and layers save the most time.

Posters and campaign assets

Layouts with legible headlines and copy, then split into layers for your design tool.

Packaging and product mockups

Transparent artwork you can place on any product, and edits that keep the product untouched.

Iterative creative direction

Adjust one thing per turn, such as color, props, or wardrobe, and keep everything else stable.

Sticker packs and brand kits

Sets of separate elements on transparent backgrounds, ready for apps, merch, and social.

Editable text from flat images

Decompose a finished poster and get its text back as text, with the font, style, and position.

Photoreal scenes

Natural light, fine materials, and realistic skin for editorial, lifestyle, and product photography.

Get ready to build

When Seedream 5.5 launches, it's one API call in whichever language you're already using, with the same key and billing as every other model on Runware.

curl https://api.runware.ai/v1 \
  -H "Authorization: Bearer $RUNWARE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '[
    {
      "taskType": "imageInference",
      "taskUUID": "<uuid-v4>",
      "model": "bytedance:[email protected]",
      "positivePrompt": "A cyclist waiting at a crossing in Lisbon at dusk, warm backlight through her hair, natural skin texture, 50mm lens",
      "resolution": "2K"
    }
  ]'

Response

[
  {
    "taskType": "imageInference",
    "taskUUID": "<uuid-v4>",
    "imageUUID": "b7e21f04-93c6-4d58-8a1e-5f0c2d9e6a31",
    "imageURL": "https://im.runware.ai/image/ws/2/ii/b7e21f04-93c6-4d58-8a1e-5f0c2d9e6a31.jpg"
  }
]

Frequently asked questions

What is Seedream 5.5?

ByteDance's new image generation and editing model, coming soon to Runware's API. It focuses on realism, aesthetics, and editing consistency: text-to-image, editing with up to 10 reference images, multi-turn edits that keep unedited areas stable, transparent output, and layer decomposition.

When will Seedream 5.5 be available on Runware?

Soon. Leave your details at the bottom of this page and we'll let you know as soon as it's live. It will run through the same API as every other model on Runware: one endpoint, one API key, and the same request shape.

What's new compared with Seedream 5.0?

Multi-turn editing that keeps unedited regions intact from one edit to the next, native transparent and semi-transparent backgrounds, and layer decomposition with text recognition. ByteDance also reports more natural lighting, finer materials, and more refined skin textures.

How does multi-turn editing work?

Send the image you want to change as a reference image with an instruction that names the change. To keep going, send the latest result back with the next instruction. Seedream 5.5 changes what you ask for and keeps the rest of the image as it was.

How do I get a transparent background?

Set settings.background to transparent. The result comes back as a PNG by default, or as a WEBP. JPG output is always opaque, so the auto setting picks an opaque background for it.

What does layer decomposition return?

Send one reference image with settings.layerDecomposition set to true. You get a base image plus up to 99 layers in outputs.layers. Each layer is a transparent PNG with a name, a description, and its bounding box. Turn on settings.layerDecompositionTextRecognition to also get the text, font, and style of each text layer.

What sizes can it output?

Set resolution to 1K, 1.5K, or 2K, or set width and height yourself, anywhere from 0.92 to 4.62 megapixels at an aspect ratio between 1:16 and 16:1. Use one or the other in a request, not both.

How many reference images can I use?

Up to 10 per request, each up to 30 MB and 36 megapixels. Layer decomposition takes exactly one.

How much will Seedream 5.5 cost on Runware?

Runware will publish pricing for Seedream 5.5 when it launches here. As with every model on Runware, you'll pay only for what you generate, with no subscription or minimums.

Can I use Seedream on Runware today?

Yes. Seedream 5.0 Pro, Seedream 5.0 Lite, Seedream 5.0 Flash, Seedream 4.5, and Seedream 4.0 are live on Runware now, under the same key and billing as every other model on the platform.

Already integrated with Seedream somewhere else?

Our Migration Skills and docs help map an existing integration over. It's the same request shape as every other Runware model, so the change is usually just field names and auth. See the migration guide for details.

Get notified when Seedream 5.5 launches

Leave your details and we'll let you know as soon as Seedream 5.5 is live on Runware. We can also talk through enterprise access, including volume discounts and dedicated RPM.