Seedream 5.0 Pro is a model in ByteDance's Seedream series for single-image creation and editing, suitable for product visuals, promotional posters, and concept design. It can generate images from text, as well as perform edits using multiple reference images, and provides transparent-background editing and layer separation. Compared with versions focused on continuous image generation, Pro is better suited for creating, revising, and delivering subsequent design assets around a single work.
Submit requests to the public API at api.acedata.cloud using the documented parameters, then use the results in your application.
api.acedata.clouddoubao-seedream-5-0-pro-260628Input parameters and result formats vary by service. Use the public API for this model and follow its guide for generation, task retrieval and editing operations.
The above are the calling specifications for this full model on this platform. Standard generation and layer separation use different input and size rules.
Text prompts can independently drive image generation, or be combined with product, character, or style references for editing. Multi-image input is suitable for bringing different assets into the same creative task; when describing them, specify the purpose of each image separately and make clear which subjects, compositions, or colors need to be preserved, avoiding mutually contradictory requirements between references.
Layer splitting converts existing images into a base image and transparent PNG layers, and provides hierarchy, names, and bounding boxes for recomposition on a canvas. It can automatically identify elements, or use prompts to specify splitting targets. It is suitable for post-production adjustments to existing compositions, rather than automatically including layered files with ordinary generation.
Standard generation can choose JPEG or PNG, and retrieve results as links or Base64, making it easy to integrate with asset libraries and design applications. A single PNG with an existing alpha channel can also be edited with a transparent background. Asynchronous tasks and callbacks allow applications to record tasks first and then receive the completed images.
Input a product image, style references, and a scene description, specifying the product appearance, placement, lighting, and background requirements to generate a product visual for display or advertising. When further modifications are needed, the result can be used as editing input to focus on adjusting the background or visual direction without having to redescribe all assets.
Provide the event theme, main elements, color direction, and layout goals, and use prompts to organize a single promotional image. When choosing a resolution preset, describe horizontal or vertical composition in the text; when there are specific canvas requirements, use pixel dimensions. After delivery, text, logos, and subject details should still be checked before completing the final layout.
Input a PNG or JPEG poster and enable layer splitting to obtain a base image and independently processable transparent layers. Design tools can recompose the image according to z_index and bounding boxes, then adjust element positions or replace assets. This is suitable for moving flat visuals into a post-production workflow, rather than requiring the model to directly output design project files.
When the deliverable is a product image, cover, or design work that needs to be separated into layers, Pro's single-image, transparent-background editing, and decomposition operations are better suited to the task. If you need to generate a related group of images at once, receive results progressively, or use web search to assist creation, choose Seedream 5.0 Lite; these features cannot be enabled for Pro by adding shared parameters.
Seedream 4.5 and 4.0 can still be used for image workflows that support image groups and streaming; Pro adds transparent-background editing and layer decomposition for post-production design. Existing continuous image-generation workflows do not need to switch to Pro solely because its version number is higher. Conversely, when editable elements need to be extracted from a flat image, choosing Pro better meets delivery requirements.
For regular generation, prepare prompt; for editing, you can include image. For decomposition, provide a separate PNG/JPEG; multiple reference images cannot be treated as a single poster to be decomposed.
Explicitly specify model=doubao-seedream-5-0-pro-260628 for /seedream/images; start regular creation with size=2K. This model does not use image groups, streaming, or search; for decomposition, then set layer_decomposition=true.
For single-image tasks, save the image; for decomposition tasks, receive the base image and transparent layers one by one. For asynchronous tasks, save task_id, retrieve results through /seedream/tasks or a callback, and check the alpha channel.
Create a tea beverage poster with a transparent glass cup in the center, the title “A Refreshing Cup” in the upper-left corner, space for a description at the bottom, and clear layering among the cup, title, leaves, and background.
First generate and proofread a single poster, then use the finished image as input for an independent layer decomposition task; check each transparent PNG, base-image completion, and layer hierarchy individually.
You should pass the full model to POST /seedream/images: doubao-seedream-5-0-pro-260628. Standard generation and editing also require prompt, and assets are submitted through image when editing. Using the full string explicitly selects this version; do not replace the invocation ID with a display name or abbreviation.
No. Standard generation and editing can accept up to 10 reference images, but the output is still a single image. Reference images are used to guide the subject, style, or composition, and do not represent the number of outputs. To obtain a related image set, choose a Seedream version that supports sequential_image_generation.
For standard generation, you can set size to 1K, 1.5K, or 2K, and describe landscape, portrait, or the target aspect ratio in prompt; you can also use WIDTHxHEIGHT to specify pixel dimensions that meet the requirements. auto is used for layer decomposition and should not be used as a size option for standard Pro generation.
No. Submit a PNG/JPEG and set layer_decomposition to true for automatic decomposition; add prompt only when you need to focus on specific elements. The result includes a base image, transparent layers, and position information, which can be stacked from bottom to top according to z_index, but this is not equivalent to project files such as PSD.
You can set async to true, obtain task_id first, then query the final result through /seedream/tasks; you can also set callback_url to receive completion notifications. Applications should save the task ID to associate requests with images, then read image_url or b64_json after obtaining the result. Do not treat task creation as generation completion.
Choose a development approach or application suited to this model and follow its setup guide.
Python SDK Integration Guide
SDK installation, authentication and examplesTypeScript SDK Integration Guide
SDK installation, authentication and examples
