Seedream 5.0 API, billed per image
Pro, Lite and Flash, from $0.018 an image — 20% below the model vendor’s own published price on Pro and Lite, at every size. Three request shapes reach them: the OpenAI Images API, the Seedream API format, and generateContent. Change a base URL and a key and keep the client code you have.
- API key in 60 seconds
- Billed per image, not per token
- No watermark unless you ask
- Links kept 7 days, or base64
- Pay by card or USDT/USDC
What each model costs
Per image. Where a model charges two rates the size decides which applies, and the price of a size is never a surprise: the reply tells you the size it produced and usage.cost what it cost. The list price beside ours is the model vendor’s own published rate.
| Model | Per image | Larger images | Image you send | Sizes |
|---|---|---|---|---|
Seedream 5.0 Pro20% off bytedance/seedream-5.0-pro | $0.045$0.036 | $0.090$0.072 from 2.61 megapixels | $0.0024 first one free | 1024 × 1024, 1024 × 1536, 1536 × 1024, 1920 × 1920, 2048 × 2048 |
Seedream 5.0 Lite20% off bytedance/seedream-5.0-lite | $0.035$0.028 | same price | free | 1920 × 1920, 2048 × 2048 |
Seedream 5.0 Flash bytedance/seedream-5.0-flash | $0.018 | same price | free | 1024 × 1024, 1024 × 1536, 1536 × 1024, 1920 × 1920, 2048 × 2048 |
Last updated 2026-10-12.
Three ways to call the same models
Pick whichever matches the code you already have. They reach the same models at the same prices; what differs is the request shape and the defaults, which are documented per format.
from openai import OpenAI
client = OpenAI(base_url="https://api.tokenify.dev/v1", api_key=os.environ["TOKENIFY_API_KEY"])
res = client.images.generate(
model="bytedance/seedream-5.0-pro",
prompt="A still life of three pears, soft window light",
size="2K",
)Each format has its own page: what to change, every parameter, which ones are accepted and ignored, and where we answer differently from the API you are coming from. OpenAI · Seedream format · google-genai
Tokenify serves Seedream through these API formats. We are not affiliated with OpenAI or Google, and we do not serve their models through these endpoints.
Sizes and shapes
Ask for a tier and the model picks the shape from your prompt; ask for an exact width and height and you get exactly that. Both are priced by the pixels produced.
| Model | Tiers | Any size between | Reference images |
|---|---|---|---|
| Seedream 5.0 Pro | 1K, 1.5K, 2K | 0.92 megapixels – 4.62 megapixels | 10 |
| Seedream 5.0 Lite | 2K, 3K, 4K | 3.69 megapixels – 16.78 megapixels | 14 |
| Seedream 5.0 Flash | 1K, 1.5K, 2K | 0.92 megapixels – 4.62 megapixels | 10 |
The exact size each tier produces at each aspect ratio is on the docs page.
Layer decomposition
Send one picture to Seedream 5.0 Pro and get the background with its objects removed, plus each object as its own transparent PNG — with the box it came from, a name, a description and a z-index. Sort by z-index, draw each layer into its box, and you have the original back with every piece movable.
Billed per output at $0.0180 each up to 2.61 megapixels and $0.036 above it — cheaper than generating an image of the same size. Credit is held for the worst case, 17 outputs, and the difference is refunded when the model says how many layers it found.
The request, the response and how to recompose it →Links or base64
response_format decides where the image lives, not what it costs. Base64 puts the bytes in the reply and stores nothing. A link puts them on a host of ours, served from the edge for 7 days and then deleted, with expires_at in the reply so you can plan around it. The link needs no API key, so it can go straight into an <img> tag.
What you never get is a link to whoever generated the image. Their reply carries one; we fetch the bytes and publish our own.
Questions people ask before signing up
What does the Seedream API cost here?
Per image, from $0.018 on Seedream 5.0 Flash. On Seedream 5.0 Pro and Seedream 5.0 Lite that is 20% below the model vendor's own published price per image, at every size; the others are sold at it. There is no subscription and no minimum: you add credit and each image is billed when it is generated.
Do I have to change my code?
Three request shapes reach the same models, so usually one line. The OpenAI Images API works with the official openai SDK, images.generate and images.edit both; the Seedream API format works with code written against that API; and generateContent works with the google-genai SDK. You change the base URL, the key and the model name.
Can it edit an image I already have?
Yes. Send the image alongside the prompt and the model edits it instead of starting from nothing — as a file part to /v1/images/edits, which is what the OpenAI SDK sends, or as base64 in the JSON body. Several reference images are allowed, in the order you list them.
What sizes can I ask for?
A tier name, which lets the model choose the shape from your prompt, or any width by height inside the model's range — 0.92 megapixels to 4.62 megapixels on Seedream 5.0 Pro. The old fixed list of five sizes is gone; anything inside the range works.
How do I get the image back?
Base64 in the response, or a link on a host of ours that we serve for 7 days and then delete. The reply carries expires_at so you know when. You never get a link to whoever generated it.
Is there a rate limit?
60 images a minute per key by default, counted in images rather than tokens. Sets count as the number of images they produce.
What is layer decomposition?
Send one picture and get the background with its objects removed, plus each object as its own transparent PNG with the box it came from, a name and a z-index — so you can move the pieces. On Seedream 5.0 Pro, billed per output at $0.0180 each up to 2.61 megapixels, which is cheaper than generating an image of the same size.