Alibaba Qwen-Image Edit

An advanced image editing model from Alibaba Cloud, offering precise control and high-quality results. Qwen-Image Edit is designed to handle common editing tasks, allowing users to modify images with natural language prompts. It supports single-image editing and multi-image blending.

Overview

• Purpose: Perform image edits using text instructions.
• Core Capability: Supports single-image editing and multi-image blending.
• Foundation: Powered by Alibaba's advanced multi-modal generative AI technology.
• Typical Output: High-quality edited image (1 per request) that blends changes with the original content.
• Use Cases: Social media content adjustment, basic photo retouching, and creative experimentation.

Key Features

• Multi-image Blending:
  • Example: Combine a girl from Image 1, wearing a skirt from Image 2, sitting in a pose from Image 3.
  • Example: Combine a girl from Image 1, a necklace from Image 2, and a bag from Image 3.
• Single-image Editing:
  • Generate depth-compliant images.
  • Replace text (e.g., "HEALTH INSURANCE" -> "明天会更好").
  • Replace shirt color.
  • Change background (e.g., to Antarctica).
• High Fidelity: Preserves the quality of the original image while applying edits.
• Smart Inpainting: Fills gaps or replaces objects based on the surrounding context.

Designed For

• Content Creators: Quickly adjust visuals for social platforms.
• General Users: Easily modify personal photos or creative projects.
• Developers: Integrate image editing capabilities into applications.

Input Requirements

To achieve the best results, follow these guidelines:

Inputs

• Structure:
  • messages array with role: user.
  • content array: 1-3 images ({"image": "..."}) + 1 text instruction ({"text": "..."}).
• Image Format: JPG, JPEG, PNG, BMP, TIFF, WEBP, GIF (first frame).
• Resolution: Recommended 384px - 3072px.
• Size Limit: Max 10MB per image.
• Text Limit: Max 800 characters.

Parameters

• n: Number of output images. Fixed at 1.
• negative_prompt: Description of content to exclude (max 500 characters).
• size: Not customizable. Output resolution maintains the aspect ratio of the input (or last input image), typically around 1024x1024.
• watermark: Boolean to add "Qwen-Image" watermark. Default is false.
• seed: Integer for reproducibility.

Pricing

• Billing Logic: Pay-as-you-go based on the number of successful output images.
• Tier: Standard tier offering a balance of performance and cost.

Limitations & FAQ

• Conversation: Does not support multi-turn conversation (single turn only).
• Languages: Chinese and English are supported; other languages are unverified.
• Aspect Ratio: Output follows the aspect ratio of the input image (or the last image if multiple are provided).
• Resolution: Does not support custom output resolution.
• Output Quantity: Only supports generating 1 image per request.

Version

• Model: Alibaba Qwen-Image Edit
• Family: Qwen-Image
• Technical Context: Standard version with essential editing capabilities.