Seedream 5.0 Flash and 5.0 Pro API launch

LaoZhang API now serves Seedream 5.0 Flash at $0.018 per call and Seedream 5.0 Pro at $0.12 per call. Both use the existing Seedream endpoint and a default-group token, so most integrations switch by changing model and removing two unsupported fields.

Published: September 25, 2026
Last verified: September 25, 2026
Status: Active. Both models are available in the default token group.

Key facts

Item	Details
Model IDs	`seedream-5-0-flash-260915` (5.0 Flash), `seedream-5-0-pro-260628` (5.0 Pro)
Price	5.0 Flash: $0.018 per call, the same as BytePlus’s list price. 5.0 Pro: $0.12 per call
Billing	Per-call: one charge for each output image, at any size
Reference images	Free on Flash. On Pro, the first is free and each additional one costs $0.003
Endpoint	`POST https://api.laozhang.ai/v1/images/generations` with a JSON body, for generation and editing
Token	`default` group, billing mode Usage first (按量优先, recommended) or Per-call (按次计费)
Reference images	1–10 in `image`, as public URLs or `data:` URLs
`size` levels	5.0 Flash: `1K`, `1.5K`, `2K`. 5.0 Pro: `1K`, `2K`
Unsupported fields	`sequential_image_generation` and `stream` return HTTP 400
Typical time	5.0 Flash: under 20 seconds. 5.0 Pro: about 2 minutes

Who should care

High-volume image work: ad creatives, e-commerce product images, posters, and marketing graphics with text. BytePlus positions 5.0 Flash for faster output, stronger layout and typography, and lower cost at scale.
Top-quality images: realistic portraits, fine material detail, professional scenes, and complex instructions. 5.0 Pro has the best image quality and instruction following of the Seedream models, and supports interactive edits, where you describe changes by position in the image.
Transparent assets: 5.0 Flash can split an existing image into a background and separate transparent layers.

Existing Seedream 5.0, 4.5, and 4.0 integrations aren’t affected and keep working without changes.

What to do

Check your token

Use an API key in the default group with the billing mode set to Usage first or Per-call. Confirm that both models appear on the console pricing page.

Change the model and remove unsupported fields

Set model to the 5.0 Flash or 5.0 Pro model ID. Delete sequential_image_generation, even when it’s "disabled", and stream from the request.

Set the size and client timeout

Pick a size the model accepts. Set the client timeout to at least 60 seconds for 5.0 Flash and at least 240 seconds for 5.0 Pro.

Compare before you switch batch jobs

Run a few of your own prompts on the new model and on your current one, then compare quality, time, and cost in call logs.

Generate an image with 5.0 Flash

This request creates one 2K image without a watermark, then downloads it. Replace the placeholder with your LaoZhang API key; the download step uses jq:

export LAOZHANG_API_KEY="YOUR_LAOZHANG_API_KEY"

curl --fail-with-body --max-time 120 "https://api.laozhang.ai/v1/images/generations" \
  -H "Authorization: Bearer $LAOZHANG_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "seedream-5-0-flash-260915",
    "prompt": "A summer sale poster for iced coffee, headline \"COOL DOWN\" in bold letters, clean layout",
    "size": "2K",
    "output_format": "jpeg",
    "response_format": "url",
    "watermark": false
  }' \
  -o seedream-flash.json

curl -sS -o seedream-flash.jpg "$(jq -r '.data[0].url' seedream-flash.json)"

The image URL is in data[0].url and stays valid for about 24 hours, so download it promptly. Without "watermark": false, the image gets an “AI generated” label. To use 5.0 Pro, change model to seedream-5-0-pro-260628 and raise --max-time to at least 240. For edits, Python, and the full parameter list, see the Seedream API guide.

Split an image into transparent layers

5.0 Flash can decompose one reference image into layers. Set layer_decomposition to true and send exactly one image:

curl --fail-with-body --max-time 300 "https://api.laozhang.ai/v1/images/generations" \
  -H "Authorization: Bearer $LAOZHANG_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "seedream-5-0-flash-260915",
    "prompt": "Split this image into separate layers.",
    "image": "https://ark-doc.tos-ap-southeast-1.bytepluses.com/doc_image/seedream4_imageToimage.png",
    "layer_decomposition": true,
    "size": "1K",
    "output_format": "png",
    "response_format": "url",
    "watermark": false
  }' \
  -o seedream-flash-layers.json

n=0
jq -r '.data[].url' seedream-flash-layers.json | while read -r url; do
  n=$((n + 1))
  curl -sS -o "layer-$n.png" "$url"
done

The response lists several images in data[]:

layer-1.png is the background, with the areas behind removed elements filled in by the model.
Each remaining file is one element as a transparent RGBA PNG.

Check the filled-in background before commercial use. Asking for a transparent background in a normal prompt doesn’t produce an alpha channel, so use layer decomposition when you need transparency.

Layer decomposition charges $0.018 for every image it returns.

The model decides how many layers to return, and the count can differ between runs of the same image. A dozen or so is common, and 16 images cost $0.288.
Use size: "1K" and a client timeout of at least 300 seconds. A 2K request can run for a long time without returning a result and is still charged.
If an image can’t be decomposed, the request returns HTTP 400 and isn’t charged. You can retry it.

Migration notes

When you move a Seedream 5.0 or 4.x request to 5.0 Flash or 5.0 Pro, check these fields:

sequential_image_generation and stream: remove both. Any value, including "disabled", returns HTTP 400.
n: ignored. Each request returns one image, except for layer decomposition, so send separate requests when you need several.
size levels: 5.0 Flash accepts 1K, 1.5K, and 2K, and uses 2K when size is omitted. 3K and 4K return HTTP 400. 5.0 Pro accepts 1K and 2K.
Aspect ratio: with a size level, 5.0 Flash chooses the aspect ratio from the content. For a fixed ratio, pass exact pixels such as 2560x1440.
Exact pixels on 5.0 Flash: 921,600 to 4,624,220 pixels in total, with the long side no more than 16 times the short side.
Exact pixels on 5.0 Pro: up to about 4.19 million pixels, for example 2048x2048, or 2720x1530 for 16:9.
Reference images: up to 10. An eleventh returns HTTP 400.
Output format: output_format accepts png or jpeg.
Response format: response_format accepts url (valid for about 24 hours) or b64_json.

Compare Seedream versions

Model	Price per call	`size` levels	Best for
5.0 Flash	$0.018	`1K`, `1.5K`, `2K`	Volume, speed, poster layouts, transparent layers
5.0 Pro	$0.12	`1K`, `2K`	Highest quality, realistic portraits, complex instructions
5.0	$0.035	`2K`, `3K`	3K output, sequential generation, streaming
4.5	$0.045	`2K`, `4K`	4K output
4.0	$0.035	`1K`, `2K`, `4K`	4K output

5.0 Pro costs about 3.4 times as much as 5.0 and is slower, so use 5.0 or 5.0 Flash for routine work. Prices are LaoZhang API public pricing as of September 25, 2026.

Frequently asked questions

How is layer decomposition billed?

Per returned image, at $0.018 each. The model decides how many layers to return, and you can’t set the count in advance, so a request that returns 16 images costs $0.288. A request that fails with HTTP 400 because the image can’t be decomposed isn’t charged.

Why does my request return HTTP 400 after switching from Seedream 5.0?

Check these causes, most common first:

The request still includes sequential_image_generation or stream. Remove both.
size is a level the model doesn’t accept, such as 3K or 4K.
The request sends more than 10 reference images.

Does 5.0 Pro cost more at 2K or with reference images?

No. LaoZhang API charges a flat $0.12 for each output image at any size. The first reference image is free, and each additional one costs $0.003, so ten references add at most $0.027. BytePlus’s own pricing for 5.0 Pro depends on output resolution and also charges for input reference images.

Can I get a transparent background by asking for one in the prompt?

No. The output has no alpha channel. Use layer decomposition with 5.0 Flash to get transparent PNG layers.

Do I need to migrate from Seedream 5.0?

No. seedream-5-0-260128 stays available and remains the model for 3K output, sequential generation, and streaming. For 4K, keep using 4.5 or 4.0.

Am I charged if my client times out?

A client timeout isn’t the same as a failure, because the server may still finish the image and charge for it. Check call logs before you resend. Requests that return an error without an image, such as a parameter error, aren’t charged.

BytePlus

BytePlus ModelArk image generation API: request parameters, size ranges, and URL lifetime
BytePlus image generation tutorial: model positioning, speed, and input image limits

LaoZhang API

Model IDs, prices, groups, and billing types come from LaoZhang API’s public pricing configuration as of September 25, 2026:

Model and pricing catalog: current prices for every Seedream version
Console pricing page: the prices that apply to your account
Token management: create an API key and set its group and billing mode
Call logs: check the model and charge for each request

​Key facts

​Who should care

​What to do

​Generate an image with 5.0 Flash

​Split an image into transparent layers

​Migration notes

​Compare Seedream versions

​Frequently asked questions

​How is layer decomposition billed?

​Why does my request return HTTP 400 after switching from Seedream 5.0?

​Does 5.0 Pro cost more at 2K or with reference images?

​Can I get a transparent background by asking for one in the prompt?

​Do I need to migrate from Seedream 5.0?

​Am I charged if my client times out?

​Sources and related links

​BytePlus

​LaoZhang API

​Related documentation