Seedream

POST /v1/images/generations

The Seedream family is a synchronous image API: one request returns the finished images; there is no task to poll. All four models share the same endpoint and request shape.

Model IDs

Model ID	Sizes	Reference images	Notes
`dola-seedream-5-0-pro-260628`	1K, 1.5K, 2K (or explicit W×H)	up to 10	Best single-image quality; interactive editing; text rendering; standard content moderation
`dola-seedream-5-0-pro-260628-ep`	1K, 1.5K, 2K (or explicit W×H)	up to 10	Same capability as the pro model, without NSFW filtering
`seedream-5-0-lite-260128`	2K, 3K, 4K (or explicit W×H)	up to 14	Group generation (sequential images) and streaming-capable upstream
`seedream-4-5-251128`	2K, 4K (or explicit W×H)	up to 14	Previous generation; group generation; JPEG output only

Supported parameters

API field	Type	Required	Values / limit
`model`	string	Yes	One of the model IDs above
`prompt`	string	Yes	Nonempty; recommended ≤300 Chinese characters or 600 English words
`image`	string or array	No	Reference images: public HTTPS URLs or `data:image/…;base64` strings; up to 10 (pro) / 14 (lite, 4-5)
`size`	string	No	Resolution tier or explicit `W×H`; default `2K`
`response_format`	string	No	`url` (default) or `b64_json`
`watermark`	boolean	No	Default `true`; set `false` for no “AI-generated” watermark
`output_format`	string	No	`png` or `jpeg`; 5.0-series models only (rejected by `seedream-4-5`)
`optimize_prompt_options`	object or string	No	`{mode: "fast"}` / `"fast"` shortens latency; pro models only
`sequential_image_generation` + `_options.max_images`	string + integer	No	Group generation, `lite`/`4-5` only; up to 15 images total
`n`	integer	No	Playground shorthand: mapped to group generation on `lite`/`4-5`; values >1 are rejected by pro models
`stream`	boolean	No	Streaming is not exposed on this integration and is rejected

Explicit W×H sizes must satisfy the model’s pixel-count range and 1:16–16:1 aspect ratio at the same time. Group generation is constrained to inputs + outputs ≤ 15 images.

Reference images

Reference images are public HTTPS URLs or base64 data URLs; each ≤30 MB, dimensions >14 px, total pixels ≤6000×6000, aspect ratio 1:16–16:1.
Prompt-referencing: refer to them by order (“Image 1”, “Image 2”).
Pro models support interactive editing — hand-drawn markers described in the prompt, or <bbox>x1 y1 x2 y2</bbox> coordinates.

Request example

curl https://api.apimart.cc/v1/images/generations \
  -H "Authorization: Bearer $HIPIX_API_KEY" \
  -H "Content-Type: application/json" \
  --data-raw '{"model": "dola-seedream-5-0-pro-260628", "prompt": "Minimalist product poster, a perfume bottle centered on a white background, headline text \"AURORA\"", "size": "2K", "watermark": false}'

With a reference image:

curl https://api.apimart.cc/v1/images/generations \
  -H "Authorization: Bearer $HIPIX_API_KEY" \
  -H "Content-Type: application/json" \
  --data-raw '{"model": "dola-seedream-5-0-pro-260628", "prompt": "Replace the jacket with a red leather one; keep the face, pose and background unchanged", "image": "https://example.com/photo.jpg", "size": "2K"}'

Results

The response contains data[] with url (valid for 24 hours — download or re-host immediately) or b64_json, plus usage.generated_images, which is the billing basis. usage.output_tokens is a usage statistic and is not the billing basis.

Billing rules

Billed per generated image: usage.generated_images — the actual number of successfully generated images. Group generation bills the actual count, not max_images.
Images blocked by moderation or lost to server errors are not billed; they carry an error entry in data[] and are excluded from the count.
Pro models (dola-seedream-5-0-pro-*) price by output pixel count per image: at most 2.61 megapixels uses the standard rate, anything larger the high rate. The tier is judged by the exact dimensions the vendor returns in data[].size.
Pro reference images: the first reference image per request is free; each additional one is billed. lite and 4-5 reference images are free.
lite / 4-5 bill a flat per-image rate regardless of output size.
Failed tasks are not billed.

Current per-image rates are shown on the model square and in the pricing details.