Gothic Portrait · skill pubblica Open Skill
Gothic Portrait
Gothic Portrait è una Open Skill pubblica su Digen. Aprila per vedere cosa crea questo workflow.
Roleplay
1 0 2
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284865/2026-10-20/evce2tcx.jpg
# Gothic Portrait
Turn an uploaded portrait into a Gothic portrait: the person keeps their own face and identity, while the costume, pose and setting are rebuilt as a Halloween / Gothic scene rendered in deep black and white with exactly one glowing red detail. Everything is a single image-to-image generation — there is no structured project state and no multi-stage pipeline.
## When to use
- The user uploads a photo (portrait, selfie, full-body, cosplay shot) and asks for a "gothic portrait", "Halloween version", "Victorian vampire look", "dark monochrome portrait", etc.
- The user already received a Gothic portrait in this conversation and now wants another scene (manor hall / ruined chapel / graveyard), another costume, another pose, or a different red accent element.
- The user uploads an image plus descriptive text ("as a widow in a chapel", "with a raven") and wants that text folded into the Gothic treatment.
Do not use for plain retouching, background removal, or photorealistic color restorations — this skill always applies the Gothic monochrome treatment described in `references/style-and-prompt.md`.
## Inputs and outputs
- **Reference image (required):** the person to transform. Passed to `generate_image` as `input_urls=[<reference>]`, which makes the call image-to-image.
- **Text (optional):** the user's wishes about scene, costume, pose, props or the red accent. Fold it into the prompt; where the text is silent, use the defaults from the references.
- **Output:** one image asset, vertical 3:4, 2K.
## Steps
1. **Collect inputs.** Confirm which image is the reference. If the user gave text, note the specifics (scene, costume, pose, prop, red accent). If no image was uploaded, ask for one and stop — this skill never invents a stranger's portrait from scratch.
2. **Read the standing style rules.** Call `read_skill_file(name="gothic-portrait", path="references/style-and-prompt.md")`. These are hard constraints (identity preservation, monochrome rules, single red accent, 3:4 layout, must-keep / must-avoid). Do not generate before reading it.
3. **Pick the scene template.** Call `read_skill_file(name="gothic-portrait", path="references/prompt-templates.md")` and choose the template that matches the user's text (candlelit manor hall / ruined chapel / misty graveyard / custom). If the user named no scene, use the default template marked there.
4. **Read the assembly and check procedure.** Call `read_skill_file(name="gothic-portrait", path="references/step-details.md")` and follow its checklist for building the final English prompt, handling edge cases (group photo, ambiguous text, conflicting requests) and the retry rules.
5. **Inspect the reference when identity is at stake.** Optionally call `recognize_image(image_url=<reference>, prompt="Describe the person's face, apparent age, hair, distinguishing features, current pose and framing.")` so the prompt can restate what must survive. Skip it when the reference is obviously clear.
6. **Generate.** Call `generate_image` with:
- `prompt`: the fully assembled English prompt (standing rules + chosen template + the user's text specifics, translated into English visual instructions).
- `input_urls=[<reference>]` — never omit it; this is an image-to-image task.
- `model='i2i.hd.lite'` — the image-to-image channel for this app.
- `orientation='portrait'` and `aspect_ratio='3:4'` — the app's documented vertical format.
- `resolution='2K'`.
- `name`: a short descriptive title, e.g. `Gothic Portrait - Manor Hall`.
- `entity_type='characters'`.
7. **Verify against the checklist.** Apply the quality check in `references/step-details.md` (identity still recognizable, monochrome except one red element, 3:4 vertical, no text overlays). If it fails, regenerate at most twice with a strengthened identity clause; then report honestly what is off.
8. **Reply briefly.** One or two sentences describing the result (scene, costume, the red accent) in natural language. Do