方舟官方产物验证 API
火山方舟(火山引擎)Upload image or video artifacts and use the official API query to check whether they were generated by the Ark platform and the corresponding model version.

Text-to-image and image-to-image models invoked via API, supporting interactive editing, layer splitting, and native text rendering in 14 languages.
Doubao Seedream 5.0 pro and flash are image generation models on the Volcano Ark platform, taking prompts and optional reference images as input and outputting generated images in png/jpeg format. Both models support text-to-image, single- and multi-image-to-image generation, interactive editing that specifies positions via box selection/coordinates/arbitrary markers, and splitting a single image into 1 base images plus up to 16 independent layers with alpha channels. pro targets higher image quality, while flash targets faster speed and lower cost. The tool is provided in API form; callers need to prepare an ARK_API_KEY and pass in the Model ID via curl, the Python/Java/Go SDKs, or an OpenAI-compatible interface. Output image URLs are retained for only 24 hours and should be saved promptly.
The Model IDs of the two models are doubao-seedream-5-0-pro-260628 and doubao-seedream-5-0-flash-260915 respectively; basic usage is consistent with other Seedream models—simply replace the model parameter. Resolution tiers support 1K, 1.5K, and 2K, and you may also specify width/height pixel values within a total pixel range of [921600, 4624220] and an aspect ratio range of [1/16, 16]. Interactive editing supports two forms: arbitrary markers plus natural-language positioning, or precise positioning via <point>/<bbox> coordinate tags. Layer splitting is enabled via layer_decomposition=true and only supports a single input image; the response returns the base image (z_index=0) and each layer's url, z_index, bounding_box (absolute/normalized), and name/description, making it easy to reconstruct, recompose, or edit layers individually in design tools or front-end canvases; when editing a single transparent layer, you can set background=transparent with output_format=png to preserve the transparent background. For image generation scenarios, up to 10 reference images may be passed in (jpeg/png/webp/bmp/tiff/gif/heic/heif, each no larger than 30MB); layer splitting only supports a single png/jpeg input. pro supports both standard and fast prompt optimization modes, while flash supports only standard mode. Both models have an IPM rate limit of 500 images per minute; layer splitting pre-deducts 17 IPM per request and refunds based on the actual quantity after generation. The Java SDK requires com.volcengine:ark-runtime 0.6.0 or higher.
Developers who need to integrate image generation and editing capabilities into applications; as well as design and operations teams that need to generate images in batches, make precise local edits, or split design drafts into editable layers.
Obtain an ARK_API_KEY in the Volcano Ark console and confirm the target Model ID (pro is doubao-seedream-5-0-pro-260628, flash is doubao-seedream-5-0-flash-260915).
Prepare inputs: text-to-image requires only a prompt; image-to-image and interactive editing require attaching the image to edit (URL or Base64) plus a prompt containing positioning information; layer splitting requires setting layer_decomposition=true and passing in only 1 image.
Call the https://ark.cn-beijing.volces.com/api/v3/images/generations endpoint via curl or the Python/Java/Go SDKs, configuring parameters such as size, output_format, and watermark.
Parse the response: for normal generation, take the image URL or b64_json from data; for layer splitting, reconstruct or recompose layers by z_index and bounding_box.
Download and save output images promptly; image URLs are retained for only 24 hours.
For latency-sensitive scenarios, set pro's optimize_prompt_options.mode to fast, or switch to the flash model.
Invocation requires a Volcano Ark account and an ARK_API_KEY; the endpoint is in the Beijing region at ark.cn-beijing.volces.com. Image URLs are retained for only 24 hours; the per-model IPM rate limit under the account is 500 images/minute; layer splitting pre-deducts 17 IPM per request. Input images must not exceed 30MB, and layer splitting only supports a single png/jpeg image. The page does not state free quotas or billing standards; check the official website for details.
费用与套餐:The source mentions that Seedream 5.0 flash is priced lower, but the page does not provide specific plans or pricing figures; costs should be confirmed on the official Volcano Ark website.