偷师

Learn by Stealth is a browser extension positioned to break down web images into reusable AI painting prompts, helping users move from "seeing an image they like" into the creative workflow of "breaking it down into executable prompts."

[Purpose] Learn by Stealth comes from the Chinese concept of "stealth learning" (偷师学艺), advocating not copying an image, but taking apart its subject, composition, lighting, text, materials, and style methods. Users can click a floating button on a web image, or use the right-click menu, to send the current image for analysis and obtain prompt assets usable for AI painting.

[Main Capabilities]

1. Image-to-prompt reverse inference: recognizes subject, style, lighting, composition, and aspect ratio, and outputs Chinese, English, and JSON structured prompts.
2. Adaptive analysis by image type: portraits are supplemented with facial features, makeup, lighting, and lens; posters are supplemented with text, layout, quantities, and corner elements; product images are supplemented with materials, placement, selling points, and commercial texture.
3. Secondary check: infographics, collages, posters, and product images undergo a secondary check of text, quantities, layout, materials, and corner details.
4. Supported types: real-person portraits, illustrations, product images, infographics, graphic posters, collages, scene photography, Logos, and other common types.
5. One-click generation of the same style: automatically carries the current language prompt, aspect ratio, and reference information, reducing repeated copying, and jumps to PixPix to continue generation.
6. Result organization: supports sidebar display, history, and asset categorization. You can reverse-infer first, then copy, favorite, or take it to PixPix to continue generation.

[How to Use] There are two installation methods. It is recommended to prioritize installation from the Chrome Store: open the store page and click "Add to Chrome"; subsequent versions can be automatically updated through the browser. If store review, network, or account environment is temporarily unavailable, you can download the local installation package and load it manually; after downloading the zip, manually load it according to the steps. According to the official description, the local installation package and the store version have identical functionality; the store version is easier to install and supports automatic browser updates, while the local installation package is suitable as a backup option during review or for beta testing.

Basic process: open the extension settings, obtain and fill in the Zhipu API Key according to the tutorial; return to any web image, click the floating button or right-click menu; check the Chinese or English results, confirm the aspect ratio, style tags, and core elements; click one-click generation to jump to PixPix with the prompt and continue creating.

[Applicable Scenarios] Suitable for AI painting, e-commerce design, content operations, social media illustrations, and visual learning. It can be used to break down covers, outfits, posters, collages, and lifestyle images; identify the materials, display, lighting, selling point labels, and commercial photography style of product images; supplement the art style, layout, text, element quantitiesand compositional relationships of posters; and strengthen descriptions of facial features, makeup, clothing, posture, lens, and lighting for portraits.

[Limitations and Notes] The official description clearly states that Learn by Stealth will not restore the original author's real prompts; the results are inferred by AI and are for learning reference only, and do not represent the original work's real prompts. The extension uses a bring-your-own-Key approach to call the vision model, requiring users to obtain and fill in the Zhipu API Key themselves to avoid congestion from shared quotas; the Key is only stored locally. Regarding price, whether it is charged, specific model versions, and the full range of supported output languages, the materials do not currently provide this information.