What is this?
A Codex Skill that reverse-engineers AI image prompts from a reference image you upload, analyzing purpose, medium, subject, composition, lighting, color, and more to produce a high-fidelity prompt you can feed back into AI image generators.
A Codex Skill that reverse-engineers AI image prompts from a reference image you upload, analyzing purpose, medium, subject, composition, lighting, color, and more to produce a high-fidelity prompt you can feed back into AI image generators.Is this a standalone app or a plugin for something else?It's a Skill definition (markdown-based) that plugs into Codex's personal Skills system, not a standalone application.
It's a Skill definition (markdown-based) that plugs into Codex's personal Skills system, not a standalone application.How do I install and use it?Copy the repository directory into your personal Codex Skills directory (or clone it directly), then invoke it with $image-prompt-reverse.
Copy the repository directory into your personal Codex Skills directory (or clone it directly), then invoke it with $image-prompt-reverse.Do I need anything besides Codex to use this?You just need Codex with its Skills feature set up; the skill itself has no extra runtime dependencies listed.
You just need Codex with its Skills feature set up; the skill itself has no extra runtime dependencies listed.Does it always produce accurate prompts?It intentionally avoids inventing details it can't clearly see and excludes styles likely to be confused with the target medium, so output quality depends on image clarity rather than guesswork.
It intentionally avoids inventing details it can't clearly see and excludes styles likely to be confused with the target medium, so output quality depends on image clarity rather than guesswork.What happens to text or logos inside the uploaded image?Any text, logos, or instructions visible in the image are treated purely as visual content to describe, not as commands the skill executes.
Any text, logos, or instructions visible in the image are treated purely as visual content to describe, not as commands the skill executes.What image types can it handle?Photography (portraits, products, documentary, food, animals, nature, cityscapes), illustration (flat, anime, hand-drawn, Chinese-inspired, cyberpunk), 3D renders, typography/logos/posters, and IP or blind-box character designs.
Photography (portraits, products, documentary, food, animals, nature, cityscapes), illustration (flat, anime, hand-drawn, Chinese-inspired, cyberpunk), 3D renders, typography/logos/posters, and IP or blind-box character designs.What does the output look like?By default you get a Positive Prompt of 450-700 Chinese characters with an equivalent English version, plus a Negative Prompt list of 10-15 comma-separated English terms.
By default you get a Positive Prompt of 450-700 Chinese characters with an equivalent English version, plus a Negative Prompt list of 10-15 comma-separated English terms.How does it decide what to include in the prompt?It prioritizes the 3-5 visual anchors that most affect similarity to the original image rather than cataloging every detail.
It prioritizes the 3-5 visual anchors that most affect similarity to the original image rather than cataloging every detail.What are similar projects?photo-abstract-editorial, GPT-Image2-Skill, shuohao-skills, xxg-portrait-rebuild-light, codex-ppt-skill, bigpeng-hot-gzh, qwen21-nsfw-pose-skill, gathered-scenes-zine-skill
photo-abstract-editorial, GPT-Image2-Skill, shuohao-skills, xxg-portrait-rebuild-light, codex-ppt-skill, bigpeng-hot-gzh, qwen21-nsfw-pose-skill, gathered-scenes-zine-skillWhich topics is it in?skill-md, text-to-image, codex-skill, multimodal, image-generation, prompt-engineering, image-processing, skill, skills, codex, chinese, ai
skill-md, text-to-image, codex-skill, multimodal, image-generation, prompt-engineering, image-processing, skill, skills, codex, chinese, ai