Access ByteDance Seedream 5.5 through EvoLink's unified API to create and edit images, generate transparent assets or split an existing design into layers. Evaluate results for your task before choosing the model and integration path.
Test Seedream 5.5 output quality, transparent backgrounds and layer splitting, estimate credit usage, and generate a sample before you formally integrate.
Create image task
Fixed to doubao-seedream-5.5Text-to-image, image-to-image and editing. One image per request: 1K and 1.5K share one rate, 2K is billed higher, and each input image is billed.
Upload source or reference images. Each input image is billed at the input-image rate.
1K and 1.5K share one rate; 2K is billed at the higher rate.
Auto lets the model decide; the task result reports whether the image came back transparent or opaque.
Generated images will appear here.
Results are kept for 24 hours — download anything you want to keep.
Generation bills one output image plus each reference image. Layer decomposition bills one input plus every returned base image and layer. 1K and 1.5K share one output rate; 2K costs more. Use the estimates and rates below to price your request.
First Test Cost
Seedream 5.5 - Text-to-Image - 1K - 1 imageOne image at the 1K or 1.5K tier, no input images
About 597 tests with $10 in credits.
In Playground, keep Seedream 5.5 selected and generate one 1K or 1.5K image without input images. 1.5K costs the same as 1K, so test at 1.5K; switch to 2K or add reference images only when the layout needs them.
Seedream 5.5 testing budget guide
Pick an amount based on how many tests you expect.Good for a first validation.
Good for prompt iteration.
Good before production integration.
Output images are billed by tier and input images per image; Layer Split bills the input image once plus every image produced.
doubao-seedream-5.5| Item | Rule | Rate | Billing |
|---|---|---|---|
| 1K / 1.5K output | Single generated image at the 1K or 1.5K tier, or a custom size up to 2,610,000 pixels. Both tiers cost the same, so 1.5K is the better pick. | $0.016/image-20% 1.088 cr/image$0.020 official price | Output image |
| 2K output | Single generated image at the 2K tier, or a custom size above 2,610,000 pixels. | $0.028/image-20% 1.904 cr/image$0.035 official price | Output image |
doubao-seedream-5.5-layerize| Item | Rule | Rate | Billing |
|---|---|---|---|
| 1K / 1.5K output | Each image produced by Layer Split - the base image and every layer - whose size is at or below 2,610,000 pixels. | $0.016/image-20% 1.088 cr/image$0.020 official price | Output image |
| 2K output | Each image produced by Layer Split above 2,610,000 pixels. A run yields one base image plus up to 99 layers, each priced by its own pixel count. | $0.028/image-20% 1.904 cr/image$0.035 official price | Output image |
| Item | Rule | Rate | Billing |
|---|---|---|---|
| Input images | Applied to each source or reference image in image_urls, up to 10 per request. Layer Split bills its single input image at the same rate. | $0.0064/image-20% 0.4352 cr/image$0.0080 official price | Input image |
| Layer Split reserve | One input image plus n output images at the 2K rate are reserved when the task starts (n defaults to 30). You pay only for the images actually produced, each at its own tier; the rest is refunded. | $0.847 · 57.5552 cr | Refunded after completion |
Seedream 5.5 is ByteDance's image generation and editing model for ongoing revisions and reusable assets. EvoLink's unified API connects generation, reference-guided editing and layer decomposition within the same production workflow.
Start with the deliverable: keep editing one image, generate transparent assets or split an existing design into separate layers.
Use the previous result as a reference to adjust the subject, background or colors. ByteDance highlights improved detail preservation and consistency across successive edits.
Official detailsFor product backgrounds, character styling and ad revisions
Generate images with an alpha channel from text or references, so product cutouts, stickers and icons can be placed on different backgrounds and layouts.
Official detailsDelivers one transparent image; layer decomposition is a different task
ByteDance describes improvements to lighting, materials, skin and dense small text. Evaluate the results for product visuals, posters and packaging drafts.
Official detailsProofread every word and check key details before publishing
The separate Layerize route returns a base image, transparent PNG layers, their positions and stacking order for new layouts and localized variants.
Official detailsOptional text recognition; no direct PSD delivery
Before delivery：Check product shape, brand marks and label text
Before delivery：Check alpha, edges and shadows on the intended background
Before delivery：Proofread and zoom in on small text and structural details
Before delivery：Check layer alignment; your application must still rebuild recognized text
Generation and reference-guided editing use the same model. Choose the separate Layerize model to decompose an existing design.
doubao-seedream-5.5Generate from text or continue editing with up to 10 reference images. Returns one image, with transparent PNG available.
Read the transparent PNG guidedoubao-seedream-5.5-layerizeProvide one PNG or JPEG to receive a base image and separate PNG layers. Optional text recognition; billing follows actual output.
Read the layer reconstruction guideEvoLink brings model access, cost planning and asynchronous task management into one workflow, helping you move from testing Seedream 5.5 to integrating it into your product.
EvoLink lets you evaluate and use Seedream 5.5 alongside Nano Banana, GPT Image and other models through a shared API integration workflow. Choose the model and route that suit transparent product assets, reference editing, poster text or layer splitting while keeping the same API key, balance and task-query flow. Model parameters and input rules still differ.
Compare the Seedream model familyCost planning can start with a concrete task. Review output tiers, reference-image charges and Layerize billing in the pricing section, then check the estimate for your current Playground settings before choosing a test budget.
View Seedream 5.5 pricing and billingTrack generation, editing and decomposition by task ID, using polling or HTTPS callbacks to connect downstream processing.
Read the Seedream 5.5 API referenceChoose models, manage your balance and track tasks through one API. Use the same model IDs when moving from this page's Playground to your product integration.
Create a key in the console and store it in a server-side environment variable.
Pass the model ID and assets that match generation, editing or decomposition.
Save the task ID and retrieve the final result through polling or an HTTPS callback.
A transparent PNG is still one flat image. Choose Layerize only when you need to move or reuse individual elements of a design.
The default maximum is 30 images including the base. Raising the limit allows up to one base image plus 99 layers; actual output count depends on the result.
OCR provides text, positions and approximate styling for clean text layers. PNGs do not become native text objects; your application handles layout and text editing.
Compare output limits, reference handling, background options and billing across Seedream models under one EvoLink account and API key.
Seedream 5.0 Pro
Compare Pro for its standard and fast prompt modes and discounted resolution-based pricing. Evaluate output quality using the same prompts and references.
View Model
Seedream 5.0 Flash
Compare Flash for flat per-image pricing with free reference images when speed and volume matter more than transparent output.
View Model
Seedream 5.0 Lite
Evaluate Lite when your workflow needs web-search-enhanced generation. Check its own resolution, reference limits and optional search charges.
View Model
GPT Image 2.5 Flare
GPT Image 2.5 Flare supports image generation and reference-based editing, with token-based billing.
View Model
GPT Image 2.5 Sunburst
GPT Image 2.5 Sunburst is another route for image creation and reference-based editing. Compare it using your own task.
View Model
Grok Imagine Image 2.0
xAI image model for text-to-image and editing with 1–3 reference images, with 1K/2K output.
View ModelByteDance announced Seedream 5.5 on November 9, 2026. The release emphasizes more stable successive edits, native RGBA assets, improved realism and clearer text. The announcement date is separate from access on an individual platform or API route.
Use doubao-seedream-5.5 for generation and reference-guided editing; each request returns one image. Use doubao-seedream-5.5-layerize to split an existing design: one PNG or JPEG input becomes a base image and separate PNG layers.
Use the current rates in this page's pricing section. Ordinary generation bills the output image; reference editing adds a charge for each input image. Layerize bills the input once plus each base image and layer actually produced.
A transparent PNG is a flat image with an alpha channel that can be placed on different backgrounds. Use Layerize when you need to move or reuse multiple elements separately. Transparent output requires background: transparent and PNG format.
Use each output as the next reference and give one focused instruction per round. Identify what must stay unchanged, including faces, product shape, logos and text. Save intermediate files and compare unedited areas after each step. If details drift, return to a saved version and narrow the next change.
Provide every required string, number and punctuation mark exactly, then specify hierarchy, alignment and placement. Review the downloaded image at delivery size and check each language, small label and repeated character. Improved text rendering does not remove the need for proofreading.
No. n caps total output images including the base: 2–100, with a default of 30, or at most one base plus 29 layers. Raising the limit allows up to one base plus 99 layers; actual count depends on decomposition results.
It does not directly return PSD files or native text objects. Output consists of a base image, separate PNG layers and metadata; output_format controls the base format. Optional OCR returns text, positions and approximate styling for clean text layers. Your application must rebuild and lay out the text.
No. Seedream 5.5 accepts only prompt_priority: standard. Check background, output format and input-image charges when switching; replacing the model ID alone is insufficient.
Create an API key and authenticate with Authorization: Bearer $EVOLINK_API_KEY. Submit the model and assets to POST /v1/images/generations and save the returned task ID. Retrieve results with GET /v1/tasks/{task_id} or an HTTPS callback. Consult the API docs for full fields, responses and errors.
Evaluate 5.5 for successive edits, transparent asset generation without pre-existing alpha input, and finer layer decomposition with OCR. Compare Pro and Flash against your existing workflow, prompt modes and actual task costs. Use the same assets; prices and version names are not quality rankings.
Current tiers are 1K, 1.5K and 2K, with no 4K tier. Ordinary generation supports aspect ratios or custom pixels within documented limits. Layerize uses tiers, not aspect ratios or custom pixel sizes.