Create material-rich scenes from a written brief or refine a source image with focused editing instructions.
Generation time varies. The demo and examples are Seedream 5.5 outputs. README feature images are illustrative artwork.
Output image
One output image per task.
Playground Examples
An architectural photograph of a small oak and rammed-earth reading pavilion beside a quiet alpine lake, morning mist, warm recessed light, linen lounge chair, large clear windows, believable joinery and natural material texture, tranquil wide composition. No people, text, letters, signage, logos, or watermark.
An intricate handcrafted paper-cut illustration of a tiny kingfisher beside a miniature river garden, layered teal leaves, delicate amber flowers, curved cream paper ripples, soft real studio shadows between the paper layers, exquisite physical craft texture. Wide landscape composition. No text, letters, logos, labels, or watermark.
Keep the glass bird's exact silhouette, pose and position on the charcoal stone. Change only the bird glass from jade-green to amber, and replace the ivory plaster background with a softly lit indigo studio backdrop. Preserve the copper leaf, framing, realistic reflections and material detail. No text, letters, logos, labels, or watermark.
Similar Models
High-fidelity image generation and reference-guided editing with photorealistic detail, typography, and flexible framing.
Create expressive images with Midjourney 8.2, 8.1, 7, or Niji 7, with controls for composition, stylization, and repeating patterns.
ByteDance image generation and editing for cinematic visuals, product shots, precise local edits, and reference-guided creative work.
Model overview
Explore product imagery, architectural concepts and detailed illustrations from a written brief. For editing, use source images to guide the subject, material and composition while describing the specific change you want.
Start with the subject and its setting, then describe the lighting, surface texture and camera position. Glass, ceramic, cloth and natural materials make useful subjects for evaluating small visual details.
Choose 1K or 2K with auto, 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, 21:9 framing. Fixed ratios determine the image dimensions within the selected area tier. Auto leaves framing to the model.
Use Image Edit with 1–10 ordered references. Identify which image supplies the subject or styling, then state what should change and which details should stay consistent. Inspect each result before using it in the next edit.
Model characteristics
Text to Image uses a nonblank prompt to generate a new composition. Specify framing, subject placement and visual constraints together.
Image Edit adds ordered references to the prompt. Keep instructions focused on a clear change, such as a material, background or prop.
Each task returns one PNG image. Review fine edges, important object details and the requested edit before using the output.
Submit the request, save data.task_id, and poll the shared task endpoint or provide a top-level webhook for completion.
FAQ
Text to Image needs a prompt. Image Edit also requires 1–10 source or reference images.
Use PNG, JPG, JPEG or WebP in the Playground. API references use HTTP(S) URLs that the service can access.
Select 1K or 2K and an aspect ratio. Fixed ratios use the resolution as an image-area tier, with width and height rounded to multiples of eight pixels. Auto lets the model determine the frame.
Yes. Use the previous output URL as a reference in a new Image Edit request and describe the next change. Each request includes its own prompt and references.
Poll the task endpoint and read data.output_urls when data.status is 1, or receive the completed result through your top-level webhook.
