Coordinate-aware editing
Use normalized point or bounding-box markers in the prompt to identify exactly where an edit belongs.

A precision-focused Seedream model for text-to-image and single-output edits with up to ten references and coordinate-aware control.

Seedream 5.0 Pro is designed for controlled changes where placement, boundaries, and preserved regions matter.
Use normalized point or bounding-box markers in the prompt to identify exactly where an edit belongs.
Combine up to ten references for subject, material, layout, identity, or replacement cues.
Write generation and editing instructions naturally in supported languages without translating the creative brief first.
The strongest edit prompts identify the target, describe the replacement, and explicitly protect everything else.
For precision edits, express a normalized 0-999 point or bounding box around the area to change.
Say what to add, remove, replace, move, or repaint and which reference supplies the new element.
State which layout, identity, lighting, proportions, and surrounding details must remain unchanged.
Use these notes as production guidance inside Creativ Studio. Provider availability and exact limits can vary by account, region, and model rollout.
High-precision generation and interactive image edits
Text plus up to 10 reference images
1K and 2K
One PNG or JPEG image per request
Keep prompts below 600 English words
Use these as starting shapes, then swap in your own subject, references, and constraints.
Replace the chair inside <bbox>140 310 470 820</bbox> with the chair from image 2. Preserve the room layout, camera angle, shadows, and every object outside the marked area.
Place a small ceramic vase centered at <point>680 610</point>. Match the table perspective and warm window light. Keep the existing composition unchanged.
It adds precise spatial editing with points and bounding boxes and focuses on a single controlled output.
Use normalized values from 0 to 999 inside point or bounding-box tags in the prompt.
No. Seedream 5.0 Pro returns one image per request; use Lite for sequential image generation.