Snowline Summit Celebration Portrait · publiczny skill Open Skill

Snowline Summit Celebration Portrait

Snowline Summit Celebration Portrait to publiczny Open Skill w Digen. Otwórz go, by zobaczyć, co tworzy ten workflow.

@eggs

Sports

0 0 0

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284862/2026-10-16/yx44u27c.jpg

# Snowline Summit Celebration Portrait

Converts one user-supplied photo into one edited image: the same recognizable person, now celebrating at a snowline mountain pass with both arms raised, in a hardshell jacket with a mountaineering backpack and an ice axe, under a deep blue sky.

Input: one reference photo. Output: one image. No structured project state is kept — everything needed lives in the conversation and in the generated canvas asset.

## When to use

- The user uploads a photo and asks for a snowy-summit / mountain-pass celebration version of it.
- The user asks to swap the background of a photo to an alpine snowline pass while keeping the outfit, or to swap the outfit to an expedition jacket + backpack + ice axe while keeping the background.
- The user wants the previous result re-tuned (stronger celebration pose, more distant climbers, colder grade, tighter selfie framing).

## Steps

1. **Confirm the input.** An input photo is required. If no image is attached, ask for one and stop. If the user attached several, ask which one to use.

2. **Read the visual conventions before generating.** Call `read_skill_file(name='snowline-summit-celebration-portrait', path='references/style-and-prompt.md')` and follow it — it holds the Style / Content / Layout / Must keep / Must avoid clauses and the copy-paste prompt template. Do not improvise the look from memory.

3. **Inspect the reference when anything is unclear** (multiple people, face too small, unclear clothing, unclear whether the user wants a full transformation). Call `recognize_image(image_url=<reference>, prompt='Describe each person: identity-facing features (face, hair, glasses, facial hair, skin tone), current clothing, current background, camera angle and framing.')`. Use the result only to fill the `<...>` slots in the prompt template — never to invent details the user did not ask for.

4. **Fill the prompt template** from `references/style-and-prompt.md`:
   - The **reference image** goes into `input_urls` (pass the uploaded asset_id, or its `s3://` uri / https URL as given). One reference image only.
   - In the prompt text, bind the reference explicitly with the identity clause from the reference file, e.g. *"the person from the reference image — keep identity, facial features, hairstyle and skin tone exactly"*.
   - Fill the person slot (gender presentation, approximate age, hair, glasses/facial hair, skin tone) and the request slot (full transformation / background-only / outfit-only / re-tune).
   - Keep the fixed clauses: snowline pass, both arms raised in celebration, hardshell jacket with hood down and collar high, loaded mountaineering backpack with chest and shoulder straps visible, gloves, one hand gripping an ice axe with the shaft up and the pick visible, deep saturated cobalt-blue sky, distant tiny line of climbers as dark silhouettes along a diagonal ridge.

5. **Generate the image** with `generate_image`:
   - `prompt` = the assembled prompt (English).
   - `input_urls=[<reference image>]`.
   - `model='i2i.hd.lite'` (image-to-image channel; write the channel, never a physical model name).
   - `orientation='portrait'` by default — it matches the arm's-length selfie framing of this theme. Use `landscape` only when the source photo is clearly landscape or the user asks for it; `square` only if the user explicitly asks.
   - Leave `resolution` unset (platform default). Pass `resolution='1080P'` only when the user asks for a higher-resolution output.
   - `name` = a short canvas title, e.g. `Snowline summit celebration`.
   - `entity_type='portraits'`.
   - Do not pass `width`/`height`, batch counts, or `group_id`/`index`.

6. **Read `references/step-details.md`** when the request is partial (background-only or outfit-only), the reference has several people, or a first result needs a retry. It holds the branch rules, retry conditions and the reply checklist.

7. **Reply in natural language** — say what was produced (person, location, pose, gear