image-to-prompt
Skill18 starsChanged 3 months agoMITView on GitHub
Analyze one or more reference images and reconstruct faithful, reusable image-generation prompts in Chinese and English. Use when users want to reverse-engineer, infer, restore, imitate, or write prompts from visual references.
1 other project uses a version of this.
Adopted by
Most recently taken up by
- canyexuanfan/17-Skillstheir own version, first seen Oct 11, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- canyexuanfan/17-Skills★ 22same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
The file
Show its rules and text
We split this skill into the rules it gives an agent
16 rules. About 0 are common, 16 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈16
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 22
Default: produce a model-neutral reconstruction prompt.
- Distinctive: line 23
If a model is named, adapt syntax and parameter suggestions to that model.
- Distinctive: line 24
If the user wants only the prompt, keep analysis brief but still perform it internally.
- Distinctive: line 43
Use spatially explicit phrases such as “subject in the lower-right third” or “soft backlight from frame left.” Prefer concrete visual …
- Distinctive: line 48
Do not add narrative, symbolism, props, logos, text, anatomy, or scenery that is not visible.
- Distinctive: line 49
Do not identify a real person from appearance alone.
- Distinctive: line 50
Do not assert an artist, brand, camera, lens, model, LoRA, checkpoint, seed, or parameter unless supplied by the user or visibly documented.
- Distinctive: line 51
If style attribution is uncertain, describe the visual characteristics instead of guessing a creator's name.
- Distinctive: line 52
Treat focal length, aperture, film stock, render engine, and generation parameters as reproduction suggestions, not recovered facts.
- Distinctive: line 55
Keep the prompt internally consistent.
- Distinctive: line 63
Write natural, information-dense phrases.
- Distinctive: line 67
Midjourney: place the natural-language prompt first.
- Distinctive: line 69
DALL-E or model-neutral: use coherent natural language with explicit composition and constraints.
- Distinctive: line 97
Always provide one complete, copy-ready editable prompt rather than a separate list of variable names.
- Distinctive: line 118
Before responding, verify that:
- Distinctive: line 123
Every image has a complete editable prompt whose bracketed values are immediately usable defaults.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Image to Prompt
Reconstruct prompts from visible evidence. Describe what can be reproduced, distinguish observation from inference, and never pretend to know the original prompt, seed, model, lens, or generation settings.
Workflow
- Confirm that at least one usable image is available. If not, ask the user to upload it.
- Inspect the image at the highest available detail. For a local file, use the image-viewing tool before analyzing it.
- Identify the image category: photograph, illustration, anime, 3D render, product image, poster/editorial graphic, UI, architecture/interior, or mixed media.
- Decompose the image using the analysis order below.
- Resolve the user's target:
- Default: produce a model-neutral reconstruction prompt.
- If a model is named, adapt syntax and parameter suggestions to that model.
- If the user wants only the prompt, keep analysis brief but still perform it internally.
- For every image, also produce a complete editable version of the prompt. Keep the standard Chinese and English prompts concrete so they remain usable without editing.
- Produce the output in the prescribed format.
- State material uncertainty only when it affects reproducibility.
Analysis Order
Prioritize high-impact visual facts:
- Subject and action — identity category, count, pose, expression, interaction, clothing, props.
- Composition — orientation, crop, subject placement, symmetry, foreground/midground/background, negative space, viewpoint.
- Environment — location, background elements, weather, time-of-day cues, atmosphere.
- Lighting — direction, softness, contrast, key/fill/rim behavior, practical lights, shadow character.
- Color — dominant palette, accent colors, saturation, temperature, tonal range.
- Surface and detail — materials, texture, depth of field, grain, bloom, motion, reflections.
- Visual language — photographic, cinematic, editorial, commercial, painterly, vector, cel-shaded, clay, CGI, collage, and other observable traits.
- Capture/render cues — plausible shot scale, perspective, focal-length range, rendering method, or post-processing. Mark these as estimates when not directly verifiable.
- Text and layout — transcribe only legible text; describe uncertain text as layout blocks rather than inventing wording.
Use spatially explicit phrases such as “subject in the lower-right third” or “soft backlight from frame left.” Prefer concrete visual properties over vague quality words.
Fidelity Rules
- Preserve the reference's subject count, spatial relationships, camera angle, crop, lighting direction, palette, and major background geometry.
- Do not add narrative, symbolism, props, logos, text, anatomy, or scenery that is not visible.
- Do not identify a real person from appearance alone. Describe observable physical and styling traits.
- Do not assert an artist, brand, camera, lens, model, LoRA, checkpoint, seed, or parameter unless supplied by the user or visibly documented.
- If style attribution is uncertain, describe the visual characteristics instead of guessing a creator's name.
- Treat focal length, aperture, film stock, render engine, and generation parameters as reproduction suggestions, not recovered facts.
- For unreadable text, write
[unreadable text]or describe its placement and typographic character. - For multiple references, identify which traits come from each image and flag contradictions instead of silently blending them.
- Keep the prompt internally consistent. Remove terms that fight the observed image, even if they are fashionable prompt keywords.
Prompt Construction
Build the main prompt in this order:
subject → action/pose → environment → composition/viewpoint → lighting → color → materials/detail → medium/style → finish
Write natural, information-dense phrases. Avoid keyword spam, duplicated adjectives, unsupported “8K/masterpiece” language, and excessive camera jargon.
For model-specific output:
- Midjourney: place the natural-language prompt first; add only useful aspect ratio, stylize, chaos, or image-weight suggestions. Never claim exact original values.
- Stable Diffusion / SDXL / FLUX: separate positive and negative prompts; suggest dimensions or aspect ratio and optional guidance/steps only when useful. Treat values as starting points.
- DALL-E or model-neutral: use coherent natural language with explicit composition and constraints; do not add unsupported command syntax.
Default Output
Respond in the user's language. Unless the user asks for another format, return:
画面拆解
A compact factual summary covering subject, composition, lighting, color, medium/style, and key textures. Mark important estimates with 推测.
中文 Prompt
One polished, copy-ready reconstruction prompt.
English Prompt
One polished, copy-ready English prompt. Optimize it independently rather than translating word-for-word.
Negative Prompt
Include only defects and unwanted deviations relevant to this image. Omit this section when the target model does not benefit from negative prompts.
建议参数
Give aspect ratio first, then a small number of model-appropriate starting suggestions. Label all inferred settings as recommendations.
可编辑变量
Always provide one complete, copy-ready editable prompt rather than a separate list of variable names. Base it on the Chinese prompt and preserve all fixed visual anchors as normal prose.
Choose the number of editable spans according to the image and likely reuse needs; there is no fixed minimum or maximum. Expose meaningful, high-impact elements such as the subject, clothing, action, environment, lighting, palette, material, style, or aspect ratio, while leaving structural details fixed when changing them would destroy the reference's defining composition.
Wrap each editable value directly in full-width brackets 【】. Put a useful current value inside each bracket so the prompt works immediately without editing, and so the user can replace only the bracketed text. Do not use empty brackets, abstract variable labels, numbered placeholders, or a detached variable list.
Example:
一位【穿深红色长外套的年轻女性】站在【雨夜的城市街道】上,柔和的【蓝紫色霓虹侧光】从画面左侧照入,保持人物居中的中景构图与潮湿路面反光。
Special Cases
- Poster or graphic design: separate visual prompt from exact copy. Preserve hierarchy, grid, spacing, typography category, and text placement.
- UI screenshot: describe information architecture, layout, component styling, density, color tokens, and interaction states. Do not treat it as a photographic scene.
- Product image: prioritize product geometry, material, camera angle, grounding shadow, reflections, backdrop, and commercial lighting.
- Portrait: prioritize pose, gaze, crop, facial expression, hair, wardrobe, background separation, and skin-light interaction without inferring identity.
- Highly stylized image: describe line quality, shape language, shading method, texture, palette, and composition before using genre labels.
- Poor-quality or obstructed image: state the limiting issue and provide a best-effort prompt without inventing hidden details.
Quality Check
Before responding, verify that:
- The main prompt preserves the reference's dominant visual structure.
- Observations and reproduction suggestions are not presented as the same thing.
- The Chinese and English prompts are both directly usable.
- Every image has a complete editable prompt whose bracketed values are immediately usable defaults.
- The editable prompt uses
【】only around replaceable values and does not reduce the output to a variable list. - The negative prompt does not accidentally remove a defining feature.
- The output contains no invented text or falsely precise settings.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

