doubao-seedream-4-0-250828 is the date-fixed version of ByteDance Seedream 4.0, bringing text-to-image generation, reference image creation, and instruction-based editing into a single creative workflow. It is suitable for poster revisions, product scene composition, illustration series, and knowledge infographics, supports output up to 4K, and focuses on understanding complex visual requirements, using reference materials, and maintaining visual continuity before and after editing.
Submit requests to the public API at api.acedata.cloud using the documented parameters, then use the results in your application.
Input 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.
Clarify capacity, inputs and outputs, and invocation methods before selecting a model.
Up to 4K is a publicly available model capability; the number of inputs, image group parameters, and task delivery methods are within this platform's supported usage scope.
Learn what doubao-seedream-4-0-250828 can bring to your work.
It can both create new images from text and continue modifying existing images. When editing, you can explicitly specify objects to replace, adjust lighting, repair scratches, or change poster text, while indicating which composition, fonts, and color schemes should be preserved. This is suitable for iterating on existing designs without having to start from a blank canvas every time.
Multi-image input is not only used to provide style, but can also help combine people, objects, and scenes. When creating, specify the elements each reference image is responsible for and the spatial relationships in the final image; combined with related image sets, you can create consecutive visuals around the same theme, reducing the work of redescribing visual settings for each image.
Seedream 4.0's creative scope includes educational illustrations, step-by-step tutorials, timelines, and professional comparison charts, and it can handle the relationship between visual content and textual explanations. It is suitable for turning structured knowledge into visual drafts and can also produce stylized expressions such as watercolor and cyberpunk; factual accuracy should still be checked before delivery.
Start with specific tasks to find where the model can be most effective.
Input an existing poster, provide the new event name, date, and revision requirements, while specifying that the original font style, text colors, and alignment should be retained. The deliverable is an edited poster image, suitable for event version updates and visual concept drafts; small text, dates, and punctuation should be checked item by item before publishing.
Provide product photos, scene references, and style examples, describe the product's placement, lighting, and relationship to the background, and generate showcase images or related series images. This is suitable for e-commerce scene ideation and brand content production; prompts should emphasize requirements for preserving packaging, logos, and appearance, and product details in the results should be reviewed.
Write steps, categories, or comparison points as clear text, and request tutorial illustrations, knowledge posters, or comparison charts. For example, separately specify the title of each column, illustrated objects, and arrangement order to obtain a visual draft that is easy to continue laying out. When formulas, historical sequences, or technical terms are involved, prepare accurate content before generating.
Choose based on task complexity, input materials, and expected results.
If your workflow requires first testing compositions at 1K, then producing final work at 2K or 4K, while retaining linked image sets and streaming delivery, 4.0 is a suitable choice. Compared with the 2K/4K presets in 4.5, 4.0 adds a 1K option; prompts and editing workflows already based on this fixed version can also continue to be iterated around it.
4.0 is suitable for text, reference images, and linked image creation, but it does not provide web search or layer decomposition. If the task requires search assistance, consider 5.0 Lite; if it requires transparent-background editing or independently composable layers, consider 5.0 Pro. The latter does not support image sets or streaming, so choose based on delivery method rather than version number alone.
From a small-scale task to formal integration.
Define the objectives, required inputs, and output requirements, using real business examples as a starting point.
Open the trial page, confirm the parameters supported by this entry point, then submit a small-scale task to review the results.
Retain the complete model ID, use the request format specified in the documentation, and confirm billing rules on the Pricing page.
Before formal use, understand the output quality and capability scope.
Answers to common questions about using doubao-seedream-4-0-250828.
Send a request to POST /seedream/images, set model exactly to doubao-seedream-4-0-250828, and provide a prompt. Text-to-image does not require a reference image; add image when editing or creating with multiple images, and read the returned result from data.
Yes. image accepts an image URL or Base64, and can also receive an array of images. prompt should specify the changes and the parts to preserve, for example, replace only the poster title while keeping the original color scheme and layout. The output is an edited image, not a design file that can be modified layer by layer.
Preset sizes include 1K, 2K, and 4K, and you can describe landscape, portrait, or square composition in the prompt; you can also use width x height pixel values supported by this version. It is recommended to determine the use case and layout first, then choose the resolution. Do not treat resolution tiers as fixed aspect ratios.
Set sequential_image_generation to auto, and specify the series theme, content of each image, and shared visual settings in the prompt; max_images can control the maximum number of outputs. If you only need a single image, set it to disabled. Series images still require checking the consistency of character and product details.
No. Streaming mode uses stream: true and application/x-ndjson to receive results line by line; asynchronous mode uses async: true to obtain task_id, then queries through /seedream/tasks, and can also use a completion callback. Streaming cannot be used together with async or callback_url.
Model information · Updated: 2026-10-01. For call parameters and billing rules, see the API and pricing sections.
Choose a development approach or application suited to this model and follow its setup guide.
↗Python SDK Integration Guide
SDK installation, authentication and examples↗TypeScript SDK Integration Guide
SDK installation, authentication and examples
↗
↗
↗Pricing
$0.0323· per generation
You pay$13,584for 142,800 credits · $0.0951 / credits46% OFF
You're billed in credits. Pick a top-up tier to see its per-credit rate — the larger the top-up, the cheaper each credit.
This is only a basic call example. See the full docs for more parameters and advanced usage.
View Docs