20 AI Photography Styles · publiczny skill Open Skill
20 AI Photography Styles
20 AI Photography Styles to publiczny Open Skill w Digen. Otwórz go, by zobaczyć, co tworzy ten workflow.
Other
1 1 2
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284865/2026-10-21/hgzrcpn4.jpg
# 20 AI Photography Styles Image-to-image style transfer for portraits. The uploaded person stays the same person (face, features, identity); the photography style, wardrobe, pose, lighting and scene change per theme. Output is still images only. ## When to use - User uploads a photo and asks for it restyled as a cinematic photo, or for a specific photography style (rainy window, golden hour, neon night, studio beauty, black and white editorial, underwater, double exposure, silhouette, etc.). - User asks for the full "20 AI photography styles" set of one person. - User asks to redo, tweak or add one style from an existing set. If no image is attached, ask for one portrait photo before doing anything else. ## Steps 1. **Check the input.** Confirm exactly which image is the reference. Ask for a clear, front-facing, well-lit portrait if the face is tiny, heavily occluded or a group shot, since identity preservation depends on it. 2. **Read the style contract.** Call `read_skill_file` with `path='references/style-and-prompt.md'`. It contains the Style / Subject / Layout / Must keep / Must avoid rules, the catalogue of 20 styles, and the copy-paste English prompt template. Do not write prompts from memory — use that file. 3. **Read the operating detail.** Call `read_skill_file` with `path='references/step-details.md'` for the checklist, style-selection branching, batch order and retry rules used in the steps below. 4. **Lock the identity.** Call `recognize_image` on the reference image with a prompt asking for a short factual description of the person's visible features (face shape, hair color and length, skin tone, apparent age range, gender presentation, distinguishing marks) plus framing and wardrobe. Keep this as the identity paragraph; reuse it verbatim in every generation so all styles stay consistent. 5. **Decide the scope.** One style, a named subset, or all 20. If unspecified, ask which style(s) they want and offer a default set of 4 contrasting ones from the catalogue. State the plan in one short line before generating. 6. **Generate, one style per call.** For each style, build the prompt from the template in `references/style-and-prompt.md`: identity paragraph + that style's fragment + wardrobe/pose/scene line + must-keep + must-avoid. Then call: `generate_image(prompt=<built prompt>, input_urls=[<the uploaded reference image>], model='i2i.hd.lite', orientation='portrait', resolution='1080P', name='<Style name>', entity_type='style_variants')` - `model` is fixed to `i2i.hd.lite` (image-to-image with a reference). Never pass a physical model name. - Keep `input_urls` to the single reference image; if the user supplies a second reference for wardrobe or scene, it goes after the portrait. - Use `orientation='portrait'` unless the user asks for landscape/square or the source is clearly landscape; keep the same orientation across a set so the results are comparable. - Use `resolution='1080P'` for high-fidelity portrait output; drop to `720P` only if the user asks for speed or the platform clamps it. - `name` = the style label (e.g. `Rain Window`, `Golden Hour`), `entity_type='style_variants'`. - One style per call (`batch_size` is not available); run them sequentially, never merge several styles into one prompt. 7. **Verify before continuing a batch.** After each result, check per `references/step-details.md` that identity is preserved and the style is legible. If the face drifted or the style is weak, re-run that single style with the corrective addendum from that file — do not re-run the whole batch. 8. **Report.** Summarize in natural language which styles were produced, note any that were retried or clamped, and do not paste long URLs. Call `set_guidance` in the same turn with relevant next actions (e.g. producing the remaining styles, swapping wardrobe, trying the set in square format). ## Notes - Identity is the hard constraint: the face must read as the uploaded person in every style. War