# Stylescape > Visual style exploration for independent musicians and artist projects. Use when a musician wants to dial in visuals, generate style directions, create stylescapes, explore cover art / social / merch / video aesthetics, compare visual lanes, or turn generated image variants into brand guidelines. This skill reads ARTIST.md and DESIGN.md, proposes 3-5 distinct visual lanes, uses available image generation tools when present, then distills the selected direction back into DESIGN.md so later promo and video skills inherit a coherent visual world. - Skill: `manzanita-research/stylescape` (Agent Skill) - Install (CLI): `npx skillmds@latest add manzanita-research/stylescape` - Raw SKILL.md: https://api.skillmd.com/api/skills/manzanita-research/stylescape/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: Manzanita-Research (https://skillmd.com/u/manzanita-research) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/manzanita-research/stylescape --- # Artist Stylescape Generate visual lanes, help the artist choose, then update `DESIGN.md`. This is not a logo generator and not a final cover-art pipeline. It is a taste-calibration instrument. The user may invoke this as: - `/artist:stylescape` - `/artist:stylescape quick` - `/artist:stylescape deep` - `/artist:stylescape refresh` If no mode is clear, default to `quick`. ## Ground Rules - Read `ARTIST.md` first. If it is missing or placeholder, run `/artist:onboard` before continuing. - Read `DESIGN.md` if present. Treat it as a working hypothesis, not law. - Ask one clarifying question if the intended asset surface is unclear. - Generate visual options as distinct lanes, not minor variations of one idea. - Do not generate images immediately. First present lane briefs and wait for explicit user approval. - Never collapse the artist into genre defaults. Preserve anti-references. - Do not update `DESIGN.md` with a lane until the user chooses or clearly approves it. - If image generation is unavailable, write image prompts and a comparison matrix instead of pretending images were generated. - Keep generated images and notes in a project-local `stylescapes/` folder when the project filesystem is available. - A stylescape is not one impressive hero image. It should expose reusable visual grammar: palette, texture, typography feel, motif, composition, motion implication, and asset applications. - "Looks like AI" is a hard failure. Reject glossy CGI, generic fantasy concept art, over-rendered cosmic digital painting, fake game-cinematic environments, and prompt-candy visuals unless the artist's actual references require that exact medium. ## Preflight Before making images or prompts: 1. Locate the artist project directory. 2. Read `ARTIST.md` and `DESIGN.md`. 3. Identify: - current era and primary goal - visual thesis - taste anchors - anti-references - palette and motif hypotheses - the output surface: social, lyric video, visualizer, cover art, merch, show poster, or general brand system 4. Name three reflex visual defaults you might accidentally reach for, then reject them. 5. Say: ```text PEDALBOARD_STYLESCAPE_PREFLIGHT: artist=pass design=pass|missing surface=known|asked imagegen=available|unavailable mutation=closed ``` Open mutation only after the user has chosen a lane or explicitly asks for drafts to be saved. ## Modes ### Quick Use for a first stylescape pass. 1. Propose 3 visual lanes with names and 2-3 bullets each. 2. Score each lane against the artist context and anti-references. 3. Ask the user to pick lanes to generate. Only proceed after explicit approval. 4. For each approved lane, create one stylescape board or concept image suitable for a moodboard. 5. Compare the lanes against `ARTIST.md`. 6. Ask which lane to keep, merge, or reject. 7. Update `DESIGN.md` only after selection. ### Deep Use for more serious brand-world work. 1. Propose 4-5 lanes. 2. Score each lane and ask the user which to generate. 3. For each approved lane, generate 2-3 assets: - key art / world image - social still or vertical crop concept - merch or poster object concept 4. Create a `STYLESCAPE.md` with image references, prompt text, strengths, risks, and recommended usage. 5. Distill the selected direction into `DESIGN.md`. ### Refresh Use when visual work already exists. 1. Read `DESIGN.md` and any recent `stylescapes/**/STYLESCAPE.md`. 2. Ask what feels wrong, stale, too generic, or newly exciting. 3. Generate replacement lanes only for the parts that need movement. 4. Patch `DESIGN.md` narrowly. ## Image Generation Workflow Use the available image generation tool when the environment supports it. In Codex, use the `imagegen` skill / built-in `image_gen` tool by default. Before using image generation, run the anti-slop gate: 1. Name the lane's **medium**: documentary photo, scanned print artifact, xerox collage, risograph poster, stage-light photograph, merch/product photo, hand-painted illustration, 3D render, etc. 2. Name the lane's **physical object**: record sleeve, ritual scarf, burned show flyer, altar cloth, tour poster, VHS still, black candle label, lyric booklet, venue projection, merch table photo, etc. 3. Name the lane's **non-AI reference logic**: what would this look like if a human designer, photographer, printmaker, art director, or stage designer made it? 4. Reject the lane if its only visual logic is "beautiful generated image." For each generated lane: - Use case: `stylized-concept`, `ads-marketing`, or `product-mockup`, depending on surface. - Asset type: stylescape board, not a final cover unless explicitly requested. - Prompt from `ARTIST.md` + `DESIGN.md`, with the lane's unique thesis. - Include constraints and anti-references explicitly. - Ask for a composed board when exploring brand direction: one large atmosphere panel plus 4-6 smaller studies for palette, texture, motif, type feel, social crop, and object/merch application. - Avoid text in images unless the user asks for a poster or merch mockup with exact copy. - Prefer 16:9 for visualizer/brand-world lanes, 9:16 for reels, 1:1 for grid/moodboard, and product framing for merch. - Reject generated outputs that are merely beautiful but generic. Say why, then revise the prompt before saving them as guidance. Prompt against AI/CGI slop explicitly: - Prefer physical, camera, print, scan, textile, object, stage, or editorial art-direction language. - Use lens/film/print/process words when relevant: 35mm flash photo, xerox, risograph, offset ink, screenprint, tour-poster paper, VHS still, projected stage light, embroidered thread, foil-stamped card, candle-label photo. - Avoid or ban: CGI, 3D render, Unreal Engine, Octane render, glossy digital painting, fantasy concept art, videogame environment, hyper-detailed cosmic goddess, generic tarot, ornate AI glyph overload, "epic cinematic" as the only idea. - If the lane truly needs illustration, specify a human-feeling process: gouache poster, ink drawing, screenprinted gig poster, illuminated manuscript scan, airbrushed van-art panel, handmade collage, photocopied zine spread. Save final selected images into: ```text stylescapes/YYYY-MM-DD-/ ``` Suggested files: ```text stylescapes/YYYY-MM-DD-/ ├── STYLESCAPE.md ├── lane-1-.png ├── lane-2-.png └── lane-3-.png ``` If images cannot be generated, still write `STYLESCAPE.md` with complete prompts. ## Lane Template Use this structure before generation: ```md ## Lane: Thesis: - Best for: - Visual ingredients: - What it protects from ARTIST.md: - Risk: - Anti-reference risk: - Medium: - Physical object: - Slop risks: - Score: - Prompt: ``` Lane names should be memorable and specific. Avoid generic names like "Dark", "Light", "Modern", or "Vintage". ## Scoring Score each proposed lane before image generation: ```md ### Score - Artist fit: /5 - Distinctiveness: /5 - Anti-reference safety: /5 - Anti-AI-slop safety: /5 - Reusable system value: /5 - Promo/video usefulness: /5 ``` If a lane scores below 4 on anti-reference safety, anti-AI-slop safety, or reusable system value, rewrite it before generating. Common failure modes: - **Poster trap:** one dramatic scene, no reusable system. - **Genre-default trap:** generic metal, witchcore, fantasy goddess, wellness, or cinematic goth defaults. - **Image-model candy:** beautiful saturated cosmic art that ignores the artist's actual thesis. - **CGI-slop trap:** glossy 3D/concept-art surfaces, impossible polished temples, over-rendered cosmic figures, and fake game-cinematic lighting. - **Prompt ontology trap:** the prompt names the subject but not the medium, object, or human production method. - **Portrait drift:** attractive avatar image with weak ritual/cosmological grammar. - **Moodboard blur:** pretty references with no decisions about palette, type, composition, or usage. ## STYLESCAPE.md Template ```md # Stylescape: ## Source Context - ARTIST.md: - DESIGN.md: - Goal: - Surface: ## Lanes ### - Image: - Thesis: - Strengths: - Risks: - Medium: - Physical object: - Score: - Prompt: ## Recommendation ## User Decision ## DESIGN.md Patch Notes - ``` ## DESIGN.md Update Rules After the user chooses a lane, update only the relevant sections: - `Visual Thesis` - `References` - `Palette` - `Type And Layout` - `Image Treatment` - `Motion` - `Symbols And Motifs` - `Merch And Objects` - `Never Do This` - `Still Unknown` Preserve useful existing language. Add a short note if the chosen direction merges multiple lanes. ## Quality Bar A good stylescape makes future visual work easier. The user should be able to say: - "Yes, this is the world." - "This is close, but too soft / too literal / too trendy." - "Make the video / poster / merch feel like lane 2 with lane 3's palette." - "I know what to make next." If the generated images are beautiful but strategically wrong, say so and regenerate with a narrower prompt.