brainstorming
brainstorming
You MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation).
photo-small-world-revivalReinterpret a user-supplied everyday photo as a small, source-specific hand-drawn world on spacious white paper. Unlike a general photo-revival treatment, this skill must preserve one primary subject, one companion element, and their clearly visible relation, then perform a restrained reconstruction using exactly one fixed scene recipe (Held Moment, Table I…
Independent GitHub snapshot · Automated classification · Original documentation governs
Generate a finished raster illustration by default. Keep the relationship that made the snapshot worth taking and let most of the page remain unoccupied.
Return the image, selected scene recipe, and a short Chinese rationale. Reveal the final prompt only when requested.
Every result must contain:
The page must read as:
one remembered relation held sharply -> most descriptive reality released
Do not preserve photographic pixels anywhere in the final image.
Build a Small World Card:
Write one internal sentence:
Keep [primary] in relation to [companion] through [visible gesture]; let [color witness] carry the remembered emphasis.
If no primary-companion relation can be identified, use a different skill instead of drawing a generic centered object.
Choose exactly one recipe. Do not combine recipes.
Use for a person with another person, pet, toy, tool, flower, meal, or carried object.
Use for food, tableware, flowers, tools, books, collections, and domestic still life.
Use for a subject looking through a window, gate, door, train frame, balcony, or architectural opening.
Use for walking, cycling, vehicles, wind, shoreline movement, or a fleeting street moment.
Choose one primary medium and at most one supporting medium:
Keep handmade pressure, imperfect registration, absorbed edges, and limited local texture. Use natural irregular contours that meet the paper directly. Avoid sticker outlines, fuzzy cutout halos, glossy digital rendering, and uniform full-page grain.
Preserve the primary-companion relation, gesture, and identity cues. Omit roughly 70-90% of descriptive detail. Do not copy the full photographic framing, realistic shadows, depth of field, or lens effects.
Choose one exact vivid hue already present in the source or a close intensified version of it. Assign it to the source element that carried the hue; do not move it to an unrelated object.
Let the hue occupy roughly 18-40% of the active cluster while remaining below roughly 5% of the whole page. Use one opaque or richly pigmented area and at most one much smaller echo inside the same relation.
Keep other marks charcoal, graphite, warm gray, faded ink, or one low-chroma source-compatible wash. Do not make the entire illustration pastel or low saturation; the witness hue must remain clear at thumbnail size.
Unless the user requests no text, include one exact handwritten caption:
If the user supplies text, reproduce it verbatim. Otherwise write a concrete phrase about the retained relation, not a generic mood. Do not invent dates, locations, signatures, or diary facts.
Use small graphite, colored pencil, or dry ink handwriting. Keep it adjacent to the cluster with generous space; do not use poster headlines or pseudo-writing.
Compile four compact paragraphs, normally 130-210 words:
Always include:
Use the supplied photograph as a semantic and structural reference only. Do not reproduce, embed, crop, collage, trace, or retain photographic pixels or photorealistic regions from it.
The final image must contain original hand-rendered illustration, paper, and the exact caption only. Preserve the primary-companion relation and the listed identity cues, but discard most descriptive scenery.
Write only visible-pixel instructions. Do not mention analysis, skill names, API details, file paths, or checklist wording.
Use the supplied image as the required semantic reference. If the user specifies an image provider, follow that provider's instructions for generation, credentials, output paths, and validation. Otherwise use the image-generation workflow available in the current environment.
Inspect normal and thumbnail views. Regenerate at most once for one observed defect:
Avoid photographic pixels, photo filters, photorealistic cutouts, full-scene redraws, realistic portraits without request, dense backgrounds, centered mascot poses, sticker outlines, fuzzy halos, cute cartoon, kawaii, anime, children's-book sweetness, product flat lays, many colorful objects, pastel wash across the page, arbitrary flowers or dots, scrapbook tape, floating paper, cast shadows, curled corners, 3D mockups, cinematic lighting, depth of field, neon, logos, watermarks, long text, and illegible pseudo-writing.
Verify:

**创作说明**
[用简短中文说明保留的主体关系、删去的现实、手绘材料和色彩见证。]
**场景配方**
[Held Moment / Table Island / Window Distance / Passing Trace]