Qwen Image 3.0 Edit API by Alibaba

Qwen-Image 3.0 (Edit)

Qwen-Image 3.0 Edit is a high-fidelity image editing and image-to-image model. Supply one to three reference images plus an instruction to edit, restyle, or drive subject-consistent generation — preserving identity while changing clothing, scene, or composition. Built-in prompt expansion (DPE/APE), flexible resolution, and reproducible seeds support both rapid exploration and controlled iteration.

Invite-beta model for interface testing. Output does not represent final quality.

🌟 Key Features

🖌 Image Editing

Provide one reference image plus an instruction to edit or restyle it, preserving subject identity.

🧩 Multi-Image Reference

Pass up to three reference images for subject-driven personalization and compositing across inputs.

🧠 Prompt Expansion (DPE / APE)

prompt_extend intelligently rewrites short instructions. Choose direct (DPE, single-pass) or agent (APE, agentic pipeline).

📐 Flexible Resolution

Specify size as width*height (up to 1440×1440), or omit it to let the model pick automatically.

🔁 Reproducible Results

Pin a seed value to reproduce or iterate on a specific edit.

⚙️ Parameters

Parameter	Required	Description
prompt	✅	Editing instruction describing how to transform the reference image(s).
reference_image_urls	✅	1 image for single-image editing; up to 3 for multi-image reference. Public URL, OSS temporary URL, or Base64 data URI. Recommended 384–2048 px, ≤10MB each.
size	❌	Output resolution as width*height (e.g. 1440*1440). Omit for automatic. Range: 512×512–1440×1440.
n	❌	Number of images to generate (default: 1).
negative_prompt	❌	Content that should NOT appear in the image.
seed	❌	Seed for reproducible output, range [0, 2147483647]. Omit for random.
prompt_extend	❌	Enable intelligent prompt rewriting (default: true).
prompt_extend_mode	❌	direct (DPE, default).

💡 Prompt Tips

Keep the woman's face and long black hair; change her knit top to a champagne silk shirt with a dark-grey blazer; place her in a modern minimalist café with soft afternoon light through floor-to-ceiling windows.

• Describe explicitly what to keep (identity, hair, pose) and what to change (clothing, scene, lighting).
• With multiple reference images, state the role of each (e.g. "person from image 1 wearing the outfit from image 2").
• Fix seed when iterating to isolate the effect of each change.
• Keep prompt_extend on for short instructions; turn it off to follow your exact wording.

🎯 Use Cases

• Portrait & Fashion Editing — Restyle clothing, backgrounds, and lighting while preserving identity.
• Subject-Driven Generation — Combine multiple reference images for consistent characters or products.
• Product Restaging — Place a product into new scenes and lighting conditions.
• Creative Compositing — Merge elements from several references into one cohesive image.
• Design Exploration — Iterate on edits at scale to compare directions.

📝 Notes

• prompt and reference_image_urls are required (besides model); provide 1–3 images.
• The platform model name is qwen-image-3.0-edit; requests are dispatched upstream as qwen-image-3.0.
• size uses a width*height format (asterisk); valid range is 512×512–1440×1440.
• Task status values: created, processing, completed, failed.
• Generated image URLs are returned in outputs once status is completed, and are valid for about 24 hours — download promptly.