Hat & Sunglasses 3 · সর্বজনীন Open Skill
Hat & Sunglasses 3
Hat & Sunglasses 3 Digen-এ একটি সর্বজনীন Open Skill। খুলে দেখুন এই ওয়ার্কফ্লো কী তৈরি করে।
Product
1 0 1
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284865/2026-10-22/xsscbmnf.jpg
# Hat & Sunglasses 3 Re-creates the "Hat & Sunglasses 3" look: take one uploaded photo and return a single studio portrait in which the same person now wears a brown felt cowboy hat and black acetate wayfarer sunglasses. The output keeps the subject's identity, hair, pose and framing from the reference and only changes the styling around them. There is no structured project state; everything needed lives in the uploaded image and the prompt. ## When to use - The user uploads a photo and asks for a hat and sunglasses version of it. - The user wants the same portrait "but with a cowboy hat and black sunglasses". - The user asks to redo or adjust a previous result of this skill (different hat tone, different frame, tighter crop). If the user only wants a generic image with no uploaded photo, this skill does not apply. ## Steps 1. **Confirm the input.** The skill needs exactly one clear reference photo. If no image is attached, ask for one and stop. If several are attached, use the one that best shows the face and shoulders and say which one you used. 2. **Read the visual rules.** Call `read_skill_file` with `path='references/style-and-prompt.md'` before writing any prompt. It contains the locked style, subject, layout, must-keep, must-avoid rules and the copy-ready English prompt template. Never generate from an improvised description. 3. **Read the operating checklist when needed.** Call `read_skill_file` with `path='references/step-details.md'` when the photo is ambiguous (several people, cropped head, subject already wearing a hat or glasses, low resolution) or when a generation has to be retried. It holds the pre-flight checks, branches, prompt-assembly order and retry rules. 4. **Inspect the photo (optional but recommended).** Call `recognize_image(image_url=<reference image>, prompt='Describe the person, hairstyle, expression, pose, clothing, framing, background and lighting in detail for a portrait restyle.')` so the prompt can restate the subject accurately instead of guessing. 5. **Write the prompt.** Fill the template in `references/style-and-prompt.md` with what you observed: subject summary, hair, expression, outfit, framing and background. Keep the hat and sunglasses fragments verbatim from the template so every result shares the same visual identity, and keep the negative prompt. 6. **Generate the image.** Call: `generate_image(prompt=<assembled prompt>, model='i2i.hd.lite', input_urls=[<reference image>], orientation='portrait', resolution='1080P', name='Hat & Sunglasses Portrait', entity_type='characters')` `i2i.hd.lite` is the image-to-image channel for this app; the uploaded photo is always the reference input. Leave `resolution` out only if the user asks for a cheaper/faster draft. 7. **Reply and offer variants.** Summarize in one or two sentences what was applied (hat, sunglasses, kept identity/framing). Do not paste long URLs. If a next step makes sense, call `set_guidance` in the same turn with concrete options such as a darker or lighter felt tone, a different frame shape, or a tighter crop. ## Notes - The identity in the reference photo is fixed: never redraw the face into a different person, never change age, gender presentation, skin tone or body proportions. - Only the hat and the sunglasses are added; the rest of the image (hair, outfit, pose, camera angle, background type) stays as close to the reference as possible. - If the subject already wears a hat or glasses, replace them with the "Hat & Sunglasses 3" versions instead of stacking them. - One result per request unless the user explicitly asks for more; do not silently produce batches. - If `success=false`, report the failure plainly and retry once following `references/step-details.md`. Never invent image links or claim a result exists. - Refuse requests that sexualize the person, target a real identifiable third party without consent, or alter official identity documents; do not call the tool for those.