Immutable. This exact content is served forever at /api/v1/blob/cc98a5e100687aea.
--- name: fantasy-minimal-magazine description: Create FANTASY-branded 3:4 minimal editorial magazine posters from user-provided images by reading the image, reducing visual information, translating it into abstract editorial material, composing with large whitespace, and generating content-driven typography. Use when the user asks for "fantasy-极简杂志", "极简杂志海报", "编辑出版海报", "图片转译海报", "FANTASY magazine poster", or wants a quiet archival/minimal publication-style poster from photos, screenshots, products, spaces, fashion, everyday scenes, landscapes, or cultural objects. --- # Fantasy Minimal Magazine Use this skill to turn a user-provided image into a **single 3:4 FANTASY minimal editorial magazine poster**. The image is source material, not something to place intact into a template. For full visual rules, read [references/full-rules.md](references/full-rules.md) when the task requires final image generation, prompt writing, critique, or iteration. ## Core Workflow Follow **READ -> REDUCE -> TRANSLATE -> COMPOSE -> PUBLISH**. 1. **Read** the image: identify subject, scene, light, material, emotion, key details, image relationships, and any user-provided context. 2. **Reduce** unrelated information: avoid default full-image display; decide what to crop, isolate, fragment, shrink, desaturate, repeat, or abstract. 3. **Translate** the image into editorial visual material: choose 1-3 image actions such as `Crop`, `Fragment`, `Monochrome`, `Selective Color`, `Scale Isolation`, `Repeat`, `Graphic Reduction`, or `Media Translation`. 4. **Compose** a quiet publication page: choose one primary layout strategy and let whitespace participate in the composition. 5. **Publish** as a 3:4 single poster with restrained `FANTASY` signature. ## Output Rules - Default output is one independent **3:4** poster. - Keep substantial whitespace, usually 60%-85% of the canvas. - Use the image at a modest scale unless the content demands otherwise; common image occupancy is about 10%-35%. - Avoid fixed templates, centered poster formulas, social-media card layouts, and commercial poster language. - Make the page feel like an editorial page, archive page, or publication object. - Use small, meaningful text. The title is subordinate to the image and archive system: default to a small horizontal title or short label, never a giant headline. - Keep most text at quiet publication scale. Avoid oversized display type, giant vertical Chinese titles, and text that becomes the main image unless the content explicitly requires a text-led archive. - Include the `FANTASY` signature in a small, restrained, non-fixed position. - For final delivery or real-image testing, run at least one image-generation pass and ensure the generated poster itself contains visible, content-driven text. Local layout/compositing can be used for drafts, masks, contact sheets, or debugging, but it is not sufficient as the only final output. - Place generated images and export artifacts according to the user's active AGENTS.md output-location rules. ## Layout Strategies Choose based on the image content, not habit: - `Floating Fragment`: small drifting image fragments, scattered text, large quiet space. - `Offset Hero`: one medium or larger image offset from center, text held at distance. - `Path Composition`: a thin line, arc, fold, or route establishes reading direction. - `Archive Block`: a compact information island surrounded by emptiness. - `Split Narrative`: one image split into several local details. - `Radical Reduction`: leave only a tiny symbol, silhouette, texture, or visual trace. - `Label Strip`: a horizontal or vertical information strip creates order. - `Index Layout`: numbers, coordinates, dots, or grids organize multiple fragments. - `Text-led Archive`: text itself becomes the primary visual, with minimal or no image. ## Typography Text must be triggered by the user's image. Derive words from subject, scene, action, light, material, place, time, emotion, traces, relationships, or explicit user information. Match the user's language by default. For Chinese requests, prefer concise Chinese text or restrained Chinese-English mixed archive labels only when the image content justifies it. Use one of these modes: - `Caption`: object, material, site, date, notes, archive. - `Poetic Note`: short, restrained observation tied directly to the image. - `Editorial`: study, motion, texture, detail, index, archive, issue. - `Archive / Index`: issue, file, department, site, material, type, date, studio, note. Avoid generic high-end phrases, random poetry, meaningless English, unrelated foreign text, slogans, and text used only as decoration. ## Visual Devices Use structural devices only when they create reading order, information hierarchy, rhythm, or content relation. Choose at most 1-2 main devices: - thin line - long path line - open frame - dot matrix - number - coordinate mark - date - label strip - small color block - underline - tiny arrow - lightly impressed type Every line must have a semantic job. A line may connect a visible image detail to its caption, crop, index, coordinate, or endpoint; act as a measured underline; or establish a readable route between image fragments. Do not add arbitrary curves, floating wires, or decorative paths. If no meaningful relation exists, omit the line entirely. Prefer neutral color with a single signal color when needed. Suitable signal colors include deep red, wine red, warm orange, acidic green, black, or one preserved original image color. Do not use a blue dot as a fixed identifier. ## Copyright And Reference Safety When using references, learn rules and relationships, not source material. Allowed to learn: - whitespace ratio - image/text relationship - image scale relationships - offset composition - reading path - text density - media treatment - color ratio - information organization - layout mechanism Do not copy: - brand names, account names, logos, copyright lines, signatures, slogans, proprietary marks, or distinctive symbols - complete compositions, identical point grids, identical text positions, original photo slices, same people, same objects, same clothing details, or same photographic scenes - `Ma:Sonic`, `Masoc`, reference account names, reference brand names, or original image logos ## Preflight Checklist Before generating or finalizing, internally confirm: - The subject and meaningful image details are understood. - 1-3 visual translation actions are chosen. - One main composition strategy is chosen. - Text mode is chosen and text comes from the image. - At most 1-2 structural devices are used. - The poster is 3:4 with large whitespace. - The title is small, horizontal, and subordinate unless a text-led archive is explicitly justified. - Any line has a visible start point, endpoint, or measured relation to image content; otherwise no line is used. - At least one image-generation pass has been used for final delivery or real-image testing, and the generated poster contains visible text. - No reference brand, logo, blue-dot identifier, or copied source layout appears. ## Failure Conditions Reject or revise results that: - look like a template - merely shrink the image into a poster - use unrelated or empty copy - lack whitespace - are over-centered, overfilled, overdecorated, overly retro, overly social-media-like, or too commercial - use text that is too large without content reason - use giant vertical titles or treat the title as the dominant graphic - use arbitrary decorative lines with no readable endpoint or image relation - are delivered as final real-image tests without any image-generation pass, or contain no visible text in the generated poster - copy reference material, reference branding, `Ma:Sonic`, `Masoc`, or fixed blue-dot symbols - make every output look the same