Capabilities

Seedream 5.5

bytedance/seedream-5.5 makes one image from a prompt, edits an image, or combines up to ten reference images. It is a synchronous call to POST /api/v1/images/generations.

Seedream 5.5 is an image generation model. It takes the OpenAI-compatible request of every image model, with a text prompt and up to ten reference images, and returns one PNG. It can make an image from text alone, change one image you send, or combine several images into a new one, and it can edit the same image over several turns. The rules every image request follows are on the Image generation page.

A village of lit glass-dome cabins and floating lanterns on snow beside a frozen lake, among pine trees, under green and violet northern lights at night.
Glass-dome village: 16:9 at 2K (2816 × 1584). Generated through AI Hub on 2026-11-13. Compressed display copy; its prompt and request are under Text to image.

Capabilities and limits

PropertyValue
Modelbytedance/seedream-5.5
CallPOST /api/v1/images/generations, synchronous
ModesText to image, an edit of one image, generation from 2 to 10 reference images
Images per request1 (n: 1)
Reference images0 to 10, in input_references
Tiers1K and 2K, 8 sizes each
Ratios1:1, 4:3, 3:4, 16:9, 9:16, 3:2, 2:3, 21:9
Default size2048x2048
OutputOne PNG, as base64 in data[0].b64_json

Inputs

  • Each reference is an image_url item: {"type": "image_url", "image_url": {"url": "…"}}. The url is a public https URL of at most 2,048 characters, or a base64 PNG, JPEG or WebP data: URI.
  • From a URL the model reads JPEG, PNG, WebP, BMP, TIFF, GIF, HEIC and HEIF images, each up to 30 MB, with a width to height ratio from 1:16 to 16:1.
  • With no references the request is text to image. With one, the prompt says what to change in that image. With two to ten, the prompt says how to combine them.
  • Write what to change and what to keep the same. The edit examples below name both.

Not offered through AI Hub

  • Layer decomposition, which splits an image into a base image and editable layers, and the text recognition that comes with it.
  • Transparent backgrounds: the background field is accepted and ignored.
  • The 1.5K and 4K tiers. A request for one is served at the highest tier below it, 1K or 2K, and the size in the response says which. A size outside the table under Sizes and ratios is clamped the same way.
  • More than one image per request. Send one request for each image.

Pricing

A request is priced per image by its tier, plus a fixed amount for each reference image sent. These are list prices in USD. The rows are in the model's tokenhub_price_table on GET /api/v1/images/models, and the Models page shows the same prices.

CaseList price
One image at 1KUSD 0.020
One image at 2KUSD 0.035
Each reference image sentUSD 0.008 more
  • An edit of one image at 2K costs 0.035 + 0.008 = USD 0.043.
  • Ten reference images at 2K cost 0.035 + 10 × 0.008 = USD 0.115; at 1K, USD 0.100.
  • Each turn of a multi-turn edit is its own request and is priced as one.

Make a request

Send one POST /api/v1/images/generations request with the key in TOKENHUB_API_KEY. The response holds the image as base64 in data[0].b64_json; the examples decode it to a PNG file. Allow enough time for the image to be made before the client stops waiting.

FieldNotes
modelRequired: bytedance/seedream-5.5.
promptRequired. Sent to the model as written.
sizeOne of the sizes under Sizes and ratios, for example 2816x1584. Or send aspect_ratio (or auto, which is 16:9) and resolution (1K or 2K) instead. Omit all three for 2048x2048.
n1, the default.
input_referencesOptional. 0 to 10 image_url items.
response_formatb64_json, the default.
import { writeFile } from "node:fs/promises";

const body = {
  "model": "bytedance/seedream-5.5",
  "prompt": "A cozy futuristic log cabin with a glowing round window in snowy Finnish Lapland at blue hour, soft aurora beginning in the sky, minimal, photorealistic.",
  "n": 1,
  "size": "1024x1024"
};
const response = await fetch("https://tokenhub-api.belive.sg/api/v1/images/generations", {
  method: "POST",
  headers: {
    Authorization: "Bearer " + process.env.TOKENHUB_API_KEY,
    "Content-Type": "application/json",
  },
  body: JSON.stringify(body),
});
if (!response.ok) throw new Error("Image request failed: " + response.status);
const result = await response.json();
await writeFile(
  "cabin.png",
  Buffer.from(result.data[0].b64_json, "base64"),
);
json (truncated response excerpt)
{
  "created": 1794556800,
  "data": [{ "b64_json": "[base64 PNG data omitted]" }]
}

Text to image

These four images were made from a prompt alone, through AI Hub on 2026-11-13. Three are 16:9 at 2K and one is a 2K square. The images are compressed WebP copies of the PNG responses; each new generation produces a different result. The people are AI-generated. Each Request block is the body that made the image.

Sizes and ratios

Send size as one of these sixteen values, or the matching aspect_ratio and resolution. Any other size is clamped to the nearest ratio at the tier its long edge implies, never to a higher tier. A size too small for 1K, or sent beside aspect_ratio or resolution, is refused with 400 before the model is called, and costs nothing. Every 1K size is priced at the 1K amount and every 2K size at the 2K amount.

Ratio1K2K
1:11024x10242048x2048 (default)
4:31152x8642368x1776
3:4864x11521776x2368
16:91424x8002816x1584
9:16800x14241584x2816
3:21248x8322496x1664
2:3832x12481664x2496
21:91568x6723136x1344

The cabin below is a 1K square, at USD 0.020. The family under the northern lights above is a 2K square (2048 × 2048), at USD 0.035. The observation tower is 9:16 at 2K.

Image edit

Send one image in input_references and say in the prompt what to change and what to keep. The output has the size you request. Replace the reference URL in the request with your own public https URL or a base64 data: URI.

Night to day

bytedance/seedream-5.5

Requested: size: "2816x1584" · 2K · one reference image · USD 0.043 at list

A village of lit glass-dome cabins and floating lanterns on snow beside a frozen lake, among pine trees, under green and violet northern lights at night.
Input: the glass-dome village from Text to image.
The glass-dome village by the lake in daylight: a clear blue sky, a low sun over the hills and bright snow, with the cabins, trees and lake in the same places.
Output.

Prompt: Turn this night scene into a bright sunny winter morning: clear blue sky, low golden sun, sparkling snow, no aurora. Keep the glass-dome cabins, the frozen lake, the trees and the composition unchanged.

Request
import { writeFile } from "node:fs/promises";

const body = {
  "model": "bytedance/seedream-5.5",
  "prompt": "Turn this night scene into a bright sunny winter morning: clear blue sky, low golden sun, sparkling snow, no aurora. Keep the glass-dome cabins, the frozen lake, the trees and the composition unchanged.",
  "n": 1,
  "size": "2816x1584",
  "input_references": [
    {
      "type": "image_url",
      "image_url": {
        "url": "https://aihub.belive.sg/docs/examples/seedream-5-5/aurora-glass-village.webp"
      }
    }
  ]
};
const response = await fetch("https://tokenhub-api.belive.sg/api/v1/images/generations", {
  method: "POST",
  headers: {
    Authorization: "Bearer " + process.env.TOKENHUB_API_KEY,
    "Content-Type": "application/json",
  },
  body: JSON.stringify(body),
});
if (!response.ok) throw new Error("Image request failed: " + response.status);
const result = await response.json();
await writeFile(
  "edit-day.png",
  Buffer.from(result.data[0].b64_json, "base64"),
);

Multi-turn editing

To edit an image in steps, send the output of one request as the only reference of the next, with a prompt for that step. Each request stands alone: AI Hub keeps no conversation, so the image you send is the whole context. The provider of Seedream 5.5 states that it keeps the subject, the composition and the style more consistent across turns than Seedream 5.0 does.

  • The output is a PNG in b64_json. Send it back as data:image/png;base64,…, or as a JPEG or WebP data: URI, or at a URL of yours. A 2K PNG is several MB; the run below re-encoded each output as a JPEG of 1,600 px on the long side before sending it back.
  • Each turn is priced as one request with one reference image: USD 0.043 at 2K.

Three edits of one image

bytedance/seedream-5.5

Requested: size: "2816x1584" for each turn. The start image is the maglev train from Text to image. AI Hub, 2026-11-13.

  1. A silver high-speed train on a curving track with light trails across snowy hills at night, a reindeer lying in the snow in front, under green and magenta northern lights.
    Start: the image made from text.
  2. The train and reindeer scene with a snowman in a red scarf added at the front left.
    Turn 1: Add a friendly snowman wearing a red scarf in the foreground on the left. Keep everything else unchanged.
  3. The same scene with the snowman, train and reindeer, now under pink and gold northern lights.
    Turn 2: Change the northern lights to deep pink and gold. Keep the snowman, the train, the reindeer and everything else unchanged.
  4. The same scene under pink and gold northern lights, with snowflakes falling across it.
    Turn 3: Add gentle falling snowflakes across the whole scene. Keep the snowman, the pink and gold aurora, the train, the reindeer and the composition unchanged.
Requests

Turn 1: add a snowman

import { writeFile } from "node:fs/promises";

const body = {
  "model": "bytedance/seedream-5.5",
  "prompt": "Add a friendly snowman wearing a red scarf in the foreground on the left. Keep everything else unchanged.",
  "n": 1,
  "size": "2816x1584",
  "input_references": [
    {
      "type": "image_url",
      "image_url": {
        "url": "https://aihub.belive.sg/docs/examples/seedream-5-5/aurora-maglev-reindeer.webp"
      }
    }
  ]
};
const response = await fetch("https://tokenhub-api.belive.sg/api/v1/images/generations", {
  method: "POST",
  headers: {
    Authorization: "Bearer " + process.env.TOKENHUB_API_KEY,
    "Content-Type": "application/json",
  },
  body: JSON.stringify(body),
});
if (!response.ok) throw new Error("Image request failed: " + response.status);
const result = await response.json();
await writeFile(
  "turn1-snowman.png",
  Buffer.from(result.data[0].b64_json, "base64"),
);

Turn 2: pink and gold sky

import { writeFile } from "node:fs/promises";

const body = {
  "model": "bytedance/seedream-5.5",
  "prompt": "Change the northern lights to deep pink and gold. Keep the snowman, the train, the reindeer and everything else unchanged.",
  "n": 1,
  "size": "2816x1584",
  "input_references": [
    {
      "type": "image_url",
      "image_url": {
        "url": "https://aihub.belive.sg/docs/examples/seedream-5-5/turn1-snowman.webp"
      }
    }
  ]
};
const response = await fetch("https://tokenhub-api.belive.sg/api/v1/images/generations", {
  method: "POST",
  headers: {
    Authorization: "Bearer " + process.env.TOKENHUB_API_KEY,
    "Content-Type": "application/json",
  },
  body: JSON.stringify(body),
});
if (!response.ok) throw new Error("Image request failed: " + response.status);
const result = await response.json();
await writeFile(
  "turn2-pink-aurora.png",
  Buffer.from(result.data[0].b64_json, "base64"),
);

Turn 3: falling snow

import { writeFile } from "node:fs/promises";

const body = {
  "model": "bytedance/seedream-5.5",
  "prompt": "Add gentle falling snowflakes across the whole scene. Keep the snowman, the pink and gold aurora, the train, the reindeer and the composition unchanged.",
  "n": 1,
  "size": "2816x1584",
  "input_references": [
    {
      "type": "image_url",
      "image_url": {
        "url": "https://aihub.belive.sg/docs/examples/seedream-5-5/turn2-pink-aurora.webp"
      }
    }
  ]
};
const response = await fetch("https://tokenhub-api.belive.sg/api/v1/images/generations", {
  method: "POST",
  headers: {
    Authorization: "Bearer " + process.env.TOKENHUB_API_KEY,
    "Content-Type": "application/json",
  },
  body: JSON.stringify(body),
});
if (!response.ok) throw new Error("Image request failed: " + response.status);
const result = await response.json();
await writeFile(
  "turn3-snowfall.png",
  Buffer.from(result.data[0].b64_json, "base64"),
);

Ten reference images

Send 2 to 10 images in input_references and say in the prompt how to combine them. This image was made through AI Hub on 2026-11-13 from the maximum of ten reference images, at size: "2048x2048". At list it costs USD 0.035 for the 2K image plus USD 0.008 for each of the ten references, USD 0.115 in all. The original response is a 2048 × 2048 PNG; the image here is a 1024 × 1024 WebP copy. The ten inputs are earlier AI Hub outputs on this site.

A poster from ten images

bytedance/seedream-5.5
Earbuds, an iced coffee, a skincare bottle and jar, a running shoe and a red apple on a floating stone island with waterfalls, under paper lanterns and hot-air balloons.
Prompt: A playful premium product poster that brings together all ten reference images: the white wireless earbuds, the iced coffee with tropical leaves, the citrus skincare bottle and cream jar, the orange-and-teal running shoe, and the red apple on a white plate, arranged on a floating island like the fantasy landscape, under a sky of glowing paper lanterns and hot-air balloons from the lantern-city scenes. Bright, cohesive studio lighting, clean composition, no text.
Request
import { writeFile } from "node:fs/promises";

const body = {
  "model": "bytedance/seedream-5.5",
  "prompt": "A playful premium product poster that brings together all ten reference images: the white wireless earbuds, the iced coffee with tropical leaves, the citrus skincare bottle and cream jar, the orange-and-teal running shoe, and the red apple on a white plate, arranged on a floating island like the fantasy landscape, under a sky of glowing paper lanterns and hot-air balloons from the lantern-city scenes. Bright, cohesive studio lighting, clean composition, no text.",
  "n": 1,
  "size": "2048x2048",
  "input_references": [
    {
      "type": "image_url",
      "image_url": {
        "url": "https://aihub.belive.sg/docs/examples/earbuds.jpg"
      }
    },
    {
      "type": "image_url",
      "image_url": {
        "url": "https://aihub.belive.sg/docs/examples/seedream-5-0-pro-coffee-product.webp"
      }
    },
    {
      "type": "image_url",
      "image_url": {
        "url": "https://aihub.belive.sg/docs/examples/seedream-5-0-pro-skincare-flatlay.webp"
      }
    },
    {
      "type": "image_url",
      "image_url": {
        "url": "https://aihub.belive.sg/docs/examples/seedream-5-0-pro-sneaker-hero.webp"
      }
    },
    {
      "type": "image_url",
      "image_url": {
        "url": "https://aihub.belive.sg/docs/examples/hy-image-3-5-preview-sky-island.webp"
      }
    },
    {
      "type": "image_url",
      "image_url": {
        "url": "https://aihub.belive.sg/docs/examples/qwen-image-2.1-pro.webp"
      }
    },
    {
      "type": "image_url",
      "image_url": {
        "url": "https://aihub.belive.sg/docs/examples/qwen-image-2.1-turbo.webp"
      }
    },
    {
      "type": "image_url",
      "image_url": {
        "url": "https://aihub.belive.sg/docs/examples/vidu-image-q2-text.webp"
      }
    },
    {
      "type": "image_url",
      "image_url": {
        "url": "https://aihub.belive.sg/docs/examples/background.jpg"
      }
    },
    {
      "type": "image_url",
      "image_url": {
        "url": "https://aihub.belive.sg/docs/examples/vidu-image-q2-reference.webp"
      }
    }
  ]
};
const response = await fetch("https://tokenhub-api.belive.sg/api/v1/images/generations", {
  method: "POST",
  headers: {
    Authorization: "Bearer " + process.env.TOKENHUB_API_KEY,
    "Content-Type": "application/json",
  },
  body: JSON.stringify(body),
});
if (!response.ok) throw new Error("Image request failed: " + response.status);
const result = await response.json();
await writeFile(
  "seedream-5-5.png",
  Buffer.from(result.data[0].b64_json, "base64"),
);

Errors

These are checked before the model is called, so each one costs nothing. See Errors and recovery for the envelope and the retry rules.

Status and typeWhen
400
invalid_request
A size that is not WIDTHxHEIGHT, is sent beside aspect_ratio or resolution, or is too small for 1K; n other than 1, more than 10 reference images, or a reference that is not a public https URL or a base64 PNG, JPEG or WebP image.
400
unsupported_parameter
A resolution that is not a tier name or is below 1K, an aspect_ratio that is not W:H or auto, a table size that disagrees with them, or a field the model does not take: negative_prompt, style, watermark or stream: true. The message names the field.

A ratio or tier the model does not offer, such as 5:4 or 4K, is not an error: it is clamped as described under Sizes and ratios. The fields quality, seed, output_format, output_compression, background and moderation are accepted and ignored.

If your client stops waiting, the image may still be made. Set a generous timeout before you send the request again.