image-generator

name	image-generator
description	Generate image assets from text prompts or create reference-image variants and edit-style regenerated images for article illustrations, poster concepts, product concept art, marketing materials, ecommerce product scenes, illustrations, icons, and cover drafts. Use when the user asks for 生成图片, 图片生成, 文生图, 文章配图, 活动海报草图, 产品概念图, 营销素材, 插画生成, 图标草案, 封面草案, 电商商品场景图, 参考图生成, 图片变体, 换背景, 改颜色, 加元素, 去元素, generate image, text to image, image variant, image edit, background change, color change, add element, or remove element.
when_to_use	Use this skill when the answer should include generated image files from a prompt or reference-image edit request. Use generate_image for text-to-image and prompt-based candidates. Use edit_image for reference-image variants, background changes, color changes, adding elements, removing elements, and style transfer requests. Do not use this skill for OCR, image understanding, table extraction, screenshot diagnosis, or image-to-text analysis. For casual, vague, incomplete, or non-professional image generation requests, first route through prompt-professionalizer to optimize the prompt, then call this skill with the optimized prompt. For generic image generation requests, call request_user_input before generating.
provider_type	builtin
provider_id	image_generator
runtime_type	tool
tools	["generate_image","edit_image"]
max_calls_per_turn	5
timeout_seconds	120
display	{"icon":"image-plus","category":"content_creation","scenarios":["content_creation","office_collaboration","business_operations"],"label":{"en_US":"Image Generator","zh_Hans":"图片生成"},"description":{"en_US":"Designed for illustrations, marketing visuals, concept art, and image revisions; generates new image assets from prompts or edits and creates variants from reference images.","zh_Hans":"适用于插画、营销素材、概念图和图片修改，可根据提示词生成新图片，也可基于参考图编辑并创建变体。"},"when_to_use":{"en_US":"Use when the answer should include generated image files.","zh_Hans":"当回答需要生成图片文件时使用。"},"tags":{"en_US":["Image","Generation","Creative"],"zh_Hans":["图片","生成","设计"]}}

Image Generator Skill

Use this skill to generate image assets from user descriptions and to create reference-URL-guided variants or edit-style regenerated images. This skill creates image files; it does not analyze images, OCR text, extract tables, or diagnose screenshots.

Supported Tasks

• generate_image: text-to-image generation for article illustrations, marketing assets, poster concepts, product concepts, ecommerce scenes, illustrations, icons, and cover drafts.
• edit_image: reference-URL-guided variants and edit-style regeneration such as background change, color change, adding elements, removing elements, and style transfer.

Important: the current backend does not pass reference images as structured image-edit inputs through adapter.ImageRequest; it passes a signed reference image URL inside the prompt. Treat edit_image as prompt-plus-reference-URL regeneration, not precise in-place editing or guaranteed reference adherence.

Workflow

1. Determine whether the user wants text-to-image generation or reference-image editing/variants.
2. If the user request is casual, vague, incomplete, or not already written as a professional image prompt, first use prompt-professionalizer to produce an optimized image prompt and parameter suggestions.
3. If required image decisions are still missing after prompt professionalization, call request_user_input before calling any image tool.
4. Read exactly one reference document for the selected task or use case.
5. Build a concise generation prompt that includes subject, scene, composition, style, color, lighting, intended use, and prohibited elements when provided.
6. Choose or confirm aspect ratio and number of candidates.
7. Call call_skill_tool with tool_name set to generate_image or edit_image.
8. In the final answer, mention the generated file names, usage notes, and any copyright or brand compliance reminder. Do not paste raw image bytes or base64.

Clarification Workflow

Call request_user_input instead of writing a plain clarification when a missing decision blocks useful generation. Ask 1-4 focused questions. Options must be concrete answers that can be used directly. A single question may contain at most 5 options.

Ask when:

• The user only says "生成图片", "做张图", "generate an image", or similar without a usable prompt.
• The intended use is unclear and affects composition, such as article image versus poster versus ecommerce scene.