qwen-image-2-pro_skill

Overview

This skill connects agents to Alibaba Qwen-Image-2.0-Pro via the inference.sh CLI for high-quality image generation. It focuses on professional text rendering, fine-grained realism, and stronger semantic adherence. Use it when you need precise, text-heavy designs like posters, banners, or marketing assets.

How this skill works

The skill invokes the alibaba/qwen-image-2-pro app through infsh to send a prompt and receive PNG images and metadata. It supports text-aware prompts, reference images for editing, and options like width, height, seed, negative_prompt, and prompt_extend to control output. Outputs include an images array and output_meta for dimensions and count.

When to use it

• Designing text-heavy posters, event flyers, and banners that require crisp multi-line text
• Creating marketing assets and product banners with precise typography and layout
• Generating photorealistic or fine-textured scenes that must follow complex prompts
• Editing images by applying styles or swapping elements between reference images
• Producing multiple variations or reproducible outputs using num_images and seed

Best practices

• Wrap exact visible text in quotes and provide exact spelling and punctuation
• Specify font details (color, size, style, position) and set prompt_extend:false for precise text control
• Use negative_prompt to avoid blurry, distorted, or low-quality text and artifacts
• Keep total pixel count between 512×512 and 2048×2048 to meet size constraints
• Provide 1–3 clear reference_images when doing edits or style transfers

Example use cases

• Create a large-format poster: "GRAND OPENING" title in red serif at top, date and details below
• Produce a marketing banner: "SUMMER SALE" in white on a sunset gradient with "50% OFF" below
• Generate multiple logo concepts: request 3–4 minimalist variations for A/B testing
• Edit photos: transfer outfit or style from one image onto a subject in another image
• Reproducible art: generate abstract geometric art with a fixed seed for consistent outputs