Skeletal Glam Transformation · सार्वजनिक Open Skill
Skeletal Glam Transformation
Skeletal Glam Transformation Digen पर एक सार्वजनिक Open Skill है। खोलें और जानें कि यह वर्कफ़्लो क्या बनाता है।
Roleplay
0 0 0
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284863/2026-10-20/jxq653d2.jpg
# Skeletal Glam Transformation Turns one uploaded photo into a Halloween "skeletal glam" restyled still, then animates that still into a short vertical reveal video. The uploaded photo is the identity anchor: the person, the pose, the camera angle, the crop and the background layout must survive the makeover. There is no structured project state — progress lives in the conversation, in `write_todos`, and in the assets already on the canvas. ## When to use - The user uploads a photo and asks for a skeleton / skull / Halloween glam transformation. - The user asks for a "spooky glam", "gothic bone beauty", "skull makeup edit" or Halloween makeover video of themselves or someone in a photo. - The user asks only for the transformed still image from that photo (run the image stages, skip the video stages). - The user asks to redo the still or redo the motion of a previous result of this flow. Do not use for plain portrait retouching, for generic Halloween scenery with no person in the photo, or for video requests that have no uploaded reference image. ## Steps 1. **Check the input.** The flow needs one clear photo of a person. If the user sent no image, ask for one before doing anything else. If the photo has several people, ask which person to transform; if they do not care, transform the most prominent one. If the photo contains no person, say so and stop. 2. **Plan the run.** For the full flow, call `write_todos` with: analyze photo, restyle still, write video script, animate video. Keep at most one item `in_progress`. 3. **Read the visual rules.** Call `read_skill_file(path="references/style-and-prompt.md")` and follow it for every prompt in this skill. Do not write prompts from memory. 4. **Analyze the photo.** Call `recognize_image` on the uploaded image with a prompt asking for: subject count, apparent age range, hair (length, color, texture, parting), face shape and features, expression, wardrobe items and colors, jewelry, body pose and hand position, camera angle and distance, framing and crop, background elements, lighting direction and color temperature. These observations become the identity anchors and layout anchors used in the prompts. Never invent details the analysis did not return. 5. **Build and render the transformed still.** Using the still template in `references/style-and-prompt.md`, write the image prompt with the anchors from step 4, then call: `generate_image(prompt=<still prompt>, input_urls=[<uploaded photo>], model='i2i.hd.lite', orientation=<'portrait' for a taller-than-wide photo, otherwise 'landscape' or 'square' to match>, name='Skeletal Glam Still', entity_type='transformed_stills')` If the user asks for maximum detail and the photo is a close-up portrait, `model='i2i.hd'` is also acceptable — never `.hd*` for adult or revealing content. 6. **Check the still.** Compare it against the anchors: same person, same pose and crop, cross-like pendant still present, no extra limbs, no gore. If the identity drifted or the framing changed, retry once with the anchors moved to the front of the prompt and a shorter style sentence. If it fails twice, show the best result and say plainly what drifted. 7. **Stop if the user only wanted the image.** If the request was for a still/photo edit, reply with what changed and skip the remaining steps. 8. **Write the video script.** Call `read_skill_file(path="references/video-script.md")`, then write a complete English script for *this* request — its mood, duration and camera preference — built from the analyzed photo, the transformed still and the user's own words. A generic template is not a script; every beat must name what is visible in this person's photo. 9. **Read the motion rules.** Call `read_skill_file(path="references/video-content-constraints.md")` and obey its shot, motion, duration and forbidden-content limits. 10. **Render the video.** Compress the script into the video prompt template from `references/style-and-prompt.md` (one line per