1950s Poster Filter · skill pública Open Skill
1950s Poster Filter
1950s Poster Filter é uma Open Skill pública no Digen. Abra para ver o que este fluxo cria.
Auto
0 0 0
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4150474/2026-10-15/fn9zeh4m.jpg
# Vintage Poster Portrait
Restyle one uploaded image as a 1950s American advertising / pin-up poster illustration: idealized hand-painted figure, playful wink, airbrushed rosy blush, halftone dots and offset-print grain on aged paper.
Image in → image out. The whole job is **one** `generate_image` call with `model='i2i'` and the user's image in `input_urls`. Text-to-image is out of scope.
## When to use
- The user uploads (or points at) an image and asks for a 1950s poster / vintage pin-up / retro advertising art version of it.
- The user asks for a "vintage poster filter", "50s poster look", "aged print grain", "retro illustration of my photo".
- The user wants to redo only part of that look on a previous result: another colorway, less blush, no wink, stronger or weaker grain, add a title.
- The user asks for one more version / variation of a result this skill produced earlier.
## Steps
1. **Read the style spec first (required).**
`read_skill_file(name='vintage-poster-portrait', path='references/vintage-poster-style.md')`
It holds the locked palette, materials, lighting rules, the face/hair/wardrobe recipe, the print-artifact vocabulary and the copy-paste prompt template. Do not improvise style wording: a consistent, recognizable look is the entire point of this skill. Once per conversation is enough; re-read only the section you need when the user asks for a variation.
2. **Get the input image.** It must be an image asset from this conversation — something the user uploaded, or an image produced earlier in the conversation. Use its asset id / s3 uri. **Never** invent or guess a URL. If the user gave no image, ask them to upload one; do not run text-to-image.
3. **Build the prompt.** Fill the template from the reference file. Keep the two halves intact:
- *Subject lock* — what must survive from the input: number of people, identity markers (hair colour and length, glasses, skin tone, apparent age), pose, framing and crop.
- *Style lock* — the fixed tokens: limited cream / blush-pink / poster-red / warm-caramel palette, gouache + airbrush pin-up rendering, arched brows, cat-eye lashes, **one playful winking eye**, airbrushed cheek blush, glossy matte poster-red lips, sculpted 50s waves, cream circle halo on a flat blush-pink ground, halftone dots, offset misregistration, aged paper grain, soft frontal studio key light.
Prompt language is **English**. Any text that appears *in* the image stays in the user's language.
4. **Generate once.**
```
generate_image(
prompt=<subject lock + style lock from the reference template>,
input_urls=[<asset id / s3 uri of the user's image>],
model='i2i',
orientation=<match the input: 'portrait' for vertical, 'square' for square, 'landscape' for wide>,
name='<short canvas title, e.g. "Vintage Poster – Anna">',
entity_type='vintage_posters'
)
```
- `orientation` follows the input image; carry it over instead of re-framing the subject.
- Only `orientation` + `resolution` control size. Never pass pixel width/height. Leave `resolution` empty (default `720P`) unless the user asks for higher quality; then pass `1080P`.
- When the user asks for maximum fidelity / finer detail, use `model='i2i.hd'` — same look, higher quality, and it does not support NSFW content.
- Do not pass `group_id` / `index`. Do not add a second model call for the same request.
5. **If it fails** (`success=false`), say so plainly, retry once with a shortened prompt, and never claim success or fabricate a result link.
6. **Reply.** One or two natural sentences about what changed (style, palette, grain, wink), then stop — do not paste long URLs; assets land on the canvas automatically. When the turn ends, you may call `set_guidance` in the same turn with 2–3 concrete next steps, e.g. "One more version, with heavier grain", "Swap the colorway: mint + cream white", "Add the headline text Vintage Poster Portrait". Nothing else is worth a multi-step todo lis