Skip to main content

mono-color

Generate one- or two-ink editorial print poster images.

Zur Installation springen

Quellinformationen

Repository
NousResearch/hermes-agent
Letzte Quellaktivität
13. September 2026 um 03:58
Erkannte Sprache von SKILL.md
Englisch
Sterne
247.798
Forks
52.199

Installationsoptionen

Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.

Quelldateien prüfen

Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.

Datei-Explorer
11 Dateien

SKILL.md wird angezeigt

SKILL.md
Quellanweisungen · Schreibgeschützte Vorschau
name
mono-color
description
Generate one- or two-ink editorial print poster images.
version
1.0.0
author
Yan Liu (adapted by Nous Research)
license
MIT
dependencies
[]
platforms
["linux","macos","windows"]
metadata
{"hermes":{"tags":["design","poster","print","duotone","risograph","editorial","image-generation"],"category":"creative","related_skills":["baoyu-infographic","meme-generation","pixel-art"],"upstream":"https://github.com/yanliudesign/mono-color-skill"}}
# Mono-Color Editorial Print Skill Turn any user theme, sentence, or reference photo into an original printed editorial artifact with one stable visual language: adaptive neutral substrate + one or two inks + mechanically reproduced image + typographic tension + concise human voice. This skill designs and generates the image; it does not imitate any one reference, copy a source composition, wording, logo, or artwork, and it never uses more than two printing inks. ## When to Use The user asks for a monochrome editorial poster, duotone print, risograph/zine poster, halftone photo treatment, one-ink or two-ink cover, or names the mono-color style. Chinese trigger vocabulary includes 单色海报、双色印刷、单色调视觉、蓝色/绿色孔版印刷、网点照片、复古或当代编辑排版. Do not trigger merely because a request mentions a color. ## Prerequisites - The Hermes `image_generate` tool (search/describe it via the deferred-tool catalog if not loaded). If image generation is unavailable, deliver prompt-only and say so. - The `design-system/` catalogs bundled with this skill (see Quick Reference). ## Quick Reference Print modes: | Mode | When | |---|---| | Pure one-ink | User explicitly requests one ink, monochrome, or one named ink without a second color | | Chromatic ink + black | Quiet, observational, natural, architectural, long-form subjects; chromatic plate carries the image, carbon/charcoal carries text | | Complementary duotone | General default; dominant plate 70–85%, accent 15–30% with a specific role; fallback pair Cobalt + Terracotta `#2148B8` + `#C65F38` | | Overprint duotone | Two plates deliberately overlap; the darker mixed zone is not a third ink | Catalogs (source of truth — **when an exact value differs, the catalog wins over any prose**; read only the catalog relevant to the current decision): | File | Provides | |---|---| | `design-system/colors.json` | Substrate IDs + exact hex, one-ink palette, approved two-ink pairs | | `design-system/compositions.json` | Layout-family IDs and geometry | | `design-system/typography.json` | Type-hierarchy role IDs | | `design-system/rhythm.json` | Visual tension profiles, focal events, unresolved edges | | `design-system/imperfections.json` | Controlled print-imperfection effect IDs and ranges | | `design-system/carriers.json` | Carrier signals (poster, journal page, cover, etc.) | References: - `references/visual-language.md` — full color/space/image-treatment/typography/tone rules - `references/composition.md` — layout decision flow, layout families, composition grammar, rhythm - `references/quality-gate.md` — originality firewall, hard avoids, inspection checklist ## Procedure 1. **Read the input.** Extract five things: - **Subject:** the one person, object, scene, or idea that must remain recognizable. - **Intent:** poetic observation, announcement, field note, personal statement, cultural poster, or specimen page. - **Words:** preserve exact supplied text verbatim in its original language — never translate or rewrite it. If no text is supplied, invent one English display phrase of 2–8 words and keep it stable across retries. Omit text only on explicit request. - **Image role:** hero photograph, isolated specimen, cropped fragment, texture source, or none. - **Representation:** faithful reproduction (default) or abstract symbol extraction (when the user asks for abstract, artistic, loose, experimental, less realistic, or less photographic treatment). For a complex topic, pick one concrete visual metaphor; do not illustrate every point. If the user supplies an image, preserve its identity and factual content — crop/isolate/halftone it, never replace the subject or invent branded details. 2. **Resolve the recipe manifest.** Fill every field; do not skip any and do not expose the manifest unless the user asks for process details. Look up IDs and exact values in `design-system/`. ````yaml subject: <one recognizable subject> intent: <one intent from Input Reading> exact_text: <user text, generated 2-8 word phrase, or none> text_language: <language of supplied text, otherwise English> representation: <faithful reproduction or abstract symbol extraction> ratio: <explicit ratio or 3:4> carrier: <one carrier ID from design-system/carriers.json or none> substrate: <one substrate ID and exact hex from design-system/colors.json> mode: <pure one-ink, chromatic + black, complementary duotone, or overprint duotone> palette: <one palette ID from design-system/colors.json> inks: <the palette's named ink or approved pair with exact hex values> plate_roles: <one explicit role per ink plate> layout: <one composition ID from design-system/compositions.json> empty_paper: <explicit percentage> visual_tension: <relaxed, balanced, or assertive from design-system/rhythm.json> focal_event: <one strong visual event from design-system/rhythm.json> release_zone: <one deliberately quiet region that gives the focal event room> unresolved_edge: <one optional edge behavior from design-system/rhythm.json or none> image_treatment: <one mechanical reproduction process> type_hierarchy: <one role ID from design-system/typography.json> disruption: <one deliberate disruption> imperfection_seed: <stable hash derived from the resolved recipe> imperfections: <0-2 restrained effect IDs for contemporary work, or 2-3 for tactile/vintage work> ```` Defaults when the user hasn't chosen: ratio `3:4`; substrate Neutral White `#FAFAF7` (Cool Gray `#E9E9E5` for architecture/tech/restrained; Pale Beige `#F5F1E8` only for tactile/archival/nostalgic subjects — never assume beige merely because the work uses halftone or risograph language); mode complementary duotone with Cobalt + Terracotta; empty paper `35%`; tension `relaxed` for reflective/leisure/unspecified cultural subjects, `balanced` for editorial information, `assertive` only for forceful declarations; disruption = one off-center image crop, or one oversized word when there is no image. Explicit user choices override defaults unless they violate the two-ink limit or the originality firewall. Identical inputs must resolve to the identical manifest — never vary palette, layout, percentages, or process for novelty. Resolve generic color words consistently: blue→Cobalt, green→Botanical Green, orange→Terracotta Orange, red→Signal Red, purple→Aubergine, black→Charcoal; green+black→Mint Green + Charcoal; blue+orange→Cobalt + Terracotta. Exact named inks always take precedence. 3. **Choose the layout.** Walk the decision flow in `references/composition.md` top to bottom and take the first match (events→ruled information poster; botanical→archival plate; repeated object→object field; crossing layers→overprint collage; supplied photo→image field or editorial cover; isolated objects→specimen annotation; phrase-as-subject→type-led declaration; essay-like→editorial journal; otherwise editorial cover). 4. **Compile the prompt** in five compact paragraphs, in order: 1. **Canvas and ink:** ratio, exact substrate hex and reason, exact one/two-ink palette hexes, print mode, plate roles, flat front-facing page (no mockup, frame, desk, or shadow). 2. **Original composition:** layout family, tension profile, one focal event, one release zone, margins (5–9%), empty-paper percentage (25–55%), grid, dominant object scale (45–80% of page) and edge crop, optional unresolved edge, one manual gesture. 3. **Subject:** what appears; for faithful reproduction, preservation/crop/halftone/paper exposure; for abstract extraction, the 2–4 identity anchors, dominant mass, structural contour, repeated rhythm, and where exposed paper cuts through. 4. **Typography and words:** hierarchy, type voices, exact short display text, and the explicit overlap/crossing/split/tight alignment between headline and dominant object. 5. **Material and avoids:** dots, fibers, bleed, misregistration, plus the hard negative constraints from `references/quality-gate.md`. Describe only visible outcomes. Never mention reference artists, studios, sample posters, or "in the style of." 5. **Generate and inspect.** Call `image_generate` with the compiled prompt. Inspect at full and thumbnail size against the checklist in `references/quality-gate.md`; regenerate once on failure (extra ink, missing plate roles, empty paper outside 25–55%, unrecognizable subject, no ≥5x type scale jump, garbled text, composition copying a reference, no identifiable focal event). If exact text still renders wrong after one retry, generate a text-light base image and state that typography should be overlaid in a layout tool — never pretend distorted text is correct. 6. **Deliver.** Save outputs under `./mono-color-output/` in the user's working directory (create it if needed), or another location the user names. Present: 1. the generated image (path or rendered); 2. the final prompt in a fenced `text` block; 3. a short recipe note: Mode, Ink (exact hexes), Layout, Type (editorial + utility voice), Process, and one Originality sentence naming the structural departures from any supplied reference. Stop at prompt-only only when the user explicitly asks or image generation is unavailable. ## Pitfalls - **Never more than two printing inks.** The substrate is not an ink; overprint mixing and density variation are not extra inks. Gradients, rainbow accents, and full-color photography are always out. - **Catalog wins over prose.** When a hex, ID, range, or geometry in `design-system/` differs from any prose description, use the catalog value. - **Preserve supplied text verbatim** — original language, exact wording, no translation unless asked. Never distort microcopy or factual text with imperfection effects. - **Never copy a source composition, wording, logo, or artwork.** Change at least four structural features from any supplied reference (see the originality firewall). No fake signatures, mastheads, sponsors, URLs, or invented branding. - **Contemporary by default.** Do not add yellowed paper, sepia, distressed borders, or retro props merely because the work uses halftone or limited inks — only when the user asks for vintage/archival mood. - **One focal event, one release zone.** Never center everything, never distribute elements evenly like a template, never fill the quiet zone with decoration. ## Verification - Manifest fully resolved, all IDs present in the `design-system/` catalogs. - Result uses one intentional white/gray/pale-beige substrate and ≤2 inks with clear plate roles. - 25–55% visibly empty paper; one dominant object at 45–80%; headline visibly crosses or locks to it. - Type hierarchy shows a 5–12x scale jump with ≤3 type voices. - Supplied subject and text preserved exactly; ≥4 structural features differ from every supplied reference. - An image was generated (unless prompt-only was requested) and saved under the output directory, and the recipe note was delivered. ## Notice Upstream example artwork is not included: `examples/` in the source repo is all-rights-reserved (see upstream ASSET-LICENSE.md); only MIT-licensed text and design-system catalogs are vendored here. Code and text are MIT (see `LICENSE.txt`).
Auf GitHub ansehen