Clay Landmark World · सार्वजनिक Open Skill
Clay Landmark World
Clay Landmark World Digen पर एक सार्वजनिक Open Skill है। खोलें और जानें कि यह वर्कफ़्लो क्या बनाता है।
Custom
2 2 3
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284863/2026-10-19/pcwnndvs.jpg
---
allowed-tools:
- generate_image
- recognize_image
---
# Clay Landmark World
Turn one recognizable city landmark and its immediate surroundings into a whimsical miniature world sculpted entirely from colorful modeling clay: visible fingerprints, gentle curves, candy colors, a wide low viewpoint, and softly blurred edges like a tabletop model photographed up close.
There is no structured project state. Everything needed for the next generation lives in the conversation plus the image assets already on the canvas.
## When to use
- The user names a landmark or a city and wants it as a miniature clay world.
- The user uploads a landmark/place photo and wants it converted into a clay diorama.
- The user wants to iterate on a clay landmark image already produced (swap the landmark, change palette, add boats/crowd, tighten the framing).
- Not for: generic illustrations, photoreal travel photos, text posters, or video.
## Inputs and outputs
- Input: text (landmark and/or city name), optionally one reference photo of the landmark or place.
- Output: one portrait image, 3:4, sculptural and text-free.
- Reference photo is optional. If the user gives neither a landmark name nor a photo, ask one short question ("Which landmark or city?") and stop.
## Steps
1. **Parse the request.** Extract: landmark, city/country, any style wishes from the user (palette, season, crowd size), and whether a reference photo is attached.
2. **Read the reference photo (only if one is attached).** Call `recognize_image` with `image_url` set to the uploaded image and a prompt asking for: landmark name, city, camera angle and height, the left/right/front/behind layout of surrounding elements (water, bridge, quay, trees, buildings), and where the crowd sits. Treat that layout as binding for the rest of the task.
3. **Load the visual rules — required before any generation.** Call `read_skill_file(name='clay-landmark-world', path='references/style-and-prompt.md')` and follow its style, subject, layout, must-keep and must-avoid rules plus its prompt templates. Never generate from memory of this file.
4. **Load the scene structure when composing a new scene** (first render, or a large change): `read_skill_file(name='clay-landmark-world', path='references/scene-composition.md')` for the layer bands, density, palette and landmark-identity anchors.
5. **Write the English prompt** using the template in `references/style-and-prompt.md`, filled with this request's landmark, city identity props, foreground, midground and sky. It must be specific to the named landmark — never a generic "clay city" sentence. If a photo was analyzed, add the layout sentence and the spatial constraints from step 2.
6. **Generate the image** with `generate_image`:
- No reference photo: `model='t2i'`, `prompt=<assembled t2i prompt>`, `aspect_ratio='3:4'`, `name='<Landmark> Clay World'`, `entity_type='clay_worlds'`.
- Reference photo provided: `model='i2i.hd.lite'`, `input_urls=[<reference image>]`, `prompt=<assembled i2i restyle prompt>`, `aspect_ratio='3:4'`, `name='<Landmark> Clay World'`, `entity_type='clay_worlds'`.
- Do not pass a physical model name, do not pass `resolution` unless the user asks for extra detail.
7. **Check the result** against the QA checklist and branching rules in `read_skill_file(name='clay-landmark-world', path='references/workflow-stages.md')`. If it fails (text or letters, photo look, deformed landmark, missing clay texture), regenerate once with the failure added to the negative fragment. At most two attempts, then report honestly what came out.
8. **Reply.** One or two natural sentences naming the landmark, city and the clay treatment. Do not paste URLs.
## Notes
- Failure: if `generate_image` returns `success=false`, say so plainly and do not invent an image link.
- Content policy: refuse sexual, hateful or graphic-violence requests without calling any tool.
- Keep the landmark's identity, silhouette and the spatial relations from the ref