Street Artist
SkillDev toolsAnalyze supplied portrait references into a stable impression map, then create a kind Street Caricature or Romance Watercolor character portrait with coherent feature relationships and targeted revision. Use when a real person's recognizable visual identity should be interpreted rather than traced. Do not use for graffiti or murals, restoration, face swaps, photorealistic retouching, personality inference, or imitation of a named living artist.
Available today. Use it from your connected AI after setup.
No other account needed.
Connect ahel once, and every AI you use reads what you have installed.
Then ask your AI: use the Street Artist skill
What this skill tells your AI
The instructions your AI receives, as published by kyungseo/skillstead in skills/street-portrait-artist/SKILL.md and read by ahel’s review.
Create a portrait that feels observed and drawn by an artist, not passed through a style filter. Extract the subject's recognizable visual grammar once, then interpret it through one of two modes:
Street Caricature/Exaggerate: a kind, compact caricature with one relational idea, open-paper facial planes, near-monochrome ink or graphite, and at most a tiny muted spot color.Romance Watercolor/Illuminate: a delicate pen-and-watercolor character portrait with restrained idealization.
Both modes must preserve the same identity relationships. They differ in what they amplify, not in who the subject is.
Required References
Before generating or editing, read these files completely:
references/impression-map.md— reference roles, identity grammar, primary anchor, and action-reaction redesign.references/production-workflow.md— permission, input profiles, capability fallbacks, output sizes, inspection, and revision.
Then read only the selected mode reference:
references/street-caricature.mdforStreet Caricature.references/romance-watercolor.mdforRomance Watercolor.
If a required reference cannot be read, stop and report that the package is incomplete.
Use When
- The user supplies one or more clear portraits and wants a recognizably similar, visibly authored character portrait.
- The request calls for a street-fair caricature, cute caricature, hand-drawn ink portrait, delicate pen-and-watercolor portrait, or lyrical character portrait.
- A paired
Twin Portraitshould show two interpretations of one person from one shared identity analysis.
Do Not Use When
- The task is graffiti, a mural, public-space street art, restoration, colorization, face swapping, beautification, age transformation, photorealistic retouching, or a fictional character without a likeness reference.
- The primary deliverable is a text-heavy poster, infographic, or social card. This skill may create the portrait layer; the host artifact workflow owns typography, layout, file placement, and publication.
- The user asks to infer personality, ethnicity, health, attractiveness, or other sensitive or unverifiable traits from appearance.
- The request asks to copy a named living artist, studio, brand, or existing artwork. Translate it into generic visual qualities or ask for a non-proprietary direction.
Procedure
- Inspect only the references needed for the task. Confirm likeness permission and choose
Quick Sketchfor one usable photo orStudio Portraitfor two or three complementary photos. Never average faces or combine poses. - Write a compact
Impression Map: head frame, T-axis, mouth-chin rhythm, outer anchors, expression, and one primary anchor. Separate observations from artistic choices and never infer personality. - Choose the mode from the request. If no mode is given, disclose
Street Caricatureas the default; ask when the choice would materially change the intended result. - Redesign relationships coherently. Any amplified feature must create a compensating action elsewhere rather than an
isolated enlargement. For
Street Caricature, lock the relational construction before the ink finish; forRomance Watercolor, reduce the scene to a few connected color masses and environmental anchors before adding selective linework. Keep non-anchor features supportive and preserve the subject's expression and reference roles. - Generate the artwork itself with an available reference-image capability. Default to
social-feed-portraitat4:5, targeting an exact1080 x 1350 pxexport when the current surface can create and verify it. Do not add text, signatures, watermarks, logos, frames, phones, or feed mockups unless requested. - Inspect identity, mode adherence, analog physicality, composition, unintended text, actual dimensions, and edge
safety separately. Deliver a concise
Artist's Notenaming the primary anchor and the structural choice it drove. - For revision, change the Impression Map or one named rendering defect, restate accepted invariants, and preserve the best accepted result. Do not blindly regenerate after repeated identity drift.
Failure Boundary
- If no accessible portrait is available, request an upload or attachment; never invent the subject or claim a local path was read.
- If image generation or editing is unavailable, provide a structured production prompt and state that no portrait was generated. A prompt is not a finished artwork.
- If exact raster export is unavailable, preserve the requested composition, report actual dimensions, and mark exact export unavailable. Never stretch the image or fabricate dimensions.
- Do not promise guaranteed likeness, deterministic regeneration, cross-product pixel parity, or human authorship.
- Do not reuse a supplied portrait, Impression Map, or output as a public example, persistent character bible, or unrelated fixture without separate permission.
Signals
- GitHub stars
- 78
- Forks
- 15
- Last commit
- Sep 2026
Advanced
- Catalog kind
- skill
- Gateway key
street-portrait-artist- Source
- github.com/kyungseo/skillstead