Style Effect Glasses · skill pubblica Open Skill

Style Effect Glasses

Style Effect Glasses è una Open Skill pubblica su Digen. Aprila per vedere cosa crea questo workflow.

@wwws

Custom

2 0 2

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284863/2026-10-19/kyhwxgsy.jpg

# Style Effect Glasses

Turn one uploaded photo into a single stylized image where the person from the photo wears translucent, glowing see-through smart glasses, lit by neon magenta/cyan light. The uploaded subject is preserved; only the styling, the visor and the light treatment are added. This skill has no structured project state — work from the conversation plus the canvas assets.

## When to use

- The user uploads a portrait/photo and asks for a "style effect", "see-through glasses", "X-ray glasses", "HUD visor", "cyber portrait" or similar stylized single image.
- The user wants an existing result re-generated with a different crop, caption, or light color.
- The user wants a poster-like version with a short headline text.

Do not use for: video requests, pure text-to-image art with no reference photo, or photo retouching that must not add glasses (ask which they want if ambiguous).

## Inputs and outputs

| Input | Required | How it is used |
|---|---|---|
| Reference image | Yes (for `i2i.hd.lite`) | Passed as `input_urls`; it defines identity, pose, clothing and framing. Its content is described in the prompt. |
| Text | Optional | Free description of the desired look (crop, caption, mood, background color). Caption text is quoted verbatim into the prompt as lettering, never as the subject description. |
| Output | — | One image on the canvas, `entity_type='style_effects'`. |

## Steps

1. **Read the reference.** If the photo's contents are not obvious (low light, group shot, heavy crop), call `recognize_image(image_url=<reference>)` to get subject count, pose, hands position, clothing and framing. Do not guess.
2. **Load the visual rules.** Call `read_skill_file(name='style-effect-glasses', path='references/style-and-prompt.md')` and follow its Style / Subject / Layout / Must keep / Must avoid rules and its prompt template. Every generation in this skill must be composed from that file.
3. **Load the sub-steps** when the case is not a plain single-portrait: call `read_skill_file(name='style-effect-glasses', path='references/step-details.md')` for input triage, the orientation/resolution decision table, the identity-preservation checklist, the post-generation quality check and the retry branches.
4. **Pick a template.** For poster versions, close-ups, hand-adjusting-the-frame poses or group photos, also call `read_skill_file(name='style-effect-glasses', path='references/prompt-templates.md')` and fill the matching template instead of writing the prompt from scratch.
5. **Write the prompt in English.** Fill the template slots (subject, pose, clothing, caption) from the user's photo and their text request. Keep the see-through visor, the neon magenta→violet→cyan light, the bokeh dots and the thin holographic line accents. Never invent content the user did not show or ask for.
6. **Choose size.** Follow the orientation/resolution table in `references/step-details.md`: keep the source aspect (`portrait` for vertical photos, `landscape` for wide ones, `square` for square posters), and use `resolution='1080P'` by default, `2K` only when the user asks for print/large output.
7. **Generate.**
   - With a reference photo (normal path): `generate_image(prompt=<filled template>, input_urls=[<reference asset id or uri>], model='i2i.hd.lite', orientation=<from step 6>, resolution=<from step 6>, name='<short title, e.g. Style Effect Portrait>', entity_type='style_effects')`.
   - Only if the user has no photo at all and explicitly wants a lookalike: use `model='t2i'` and build the subject from their description; tell them the result will not preserve a real identity.
8. **Quality check.** Compare the result against the checklist in `references/step-details.md`. If identity, the eyes-through-glass read, or the hand/finger count is wrong, regenerate once with a tightened prompt (cap: two retries). Never present a broken result as final.
9. **Reply briefly.** One or two sentences on what was added (translucent visor, neon rim light,