Image

실제 사진처럼 보이는(photorealistic) 이미지 생성 프롬프트를 사용자와 단계별로 함께 만드는 스킬. 피사체 → 환경 → 동작과 끝점 → 카메라 → 조명과 스타일 순서로 AskUserQuestion 객관식 질문을 던지며 프롬프트를 쌓아 올리고, 최종적으로 영문 프롬프트와 한글 번역을 함께 낸다. 사용자가 "이미지 만들어줘", "사진 프롬프트", "실사 이미지", "현실적인 장면", "AI 이미지 프롬프트 다듬어줘", "Midjourney/GPT Image/Nano Banana/Imagen/Flux용 프롬프트" 등 이미지 생성이나 이미지 프롬프트를 언급하면, 스타일을 명시하지 않았더라도 이 스킬을 사용한다. 일러스트·로고·다이어그램처럼 명백히 사진이 아닌 결과물에는 쓰지 않는다.

用 agnes-img 调用 Agnes AI 生成或编辑图片（文生图、图生图、多图合成），默认走最新的 2.5 模型并在失败时自动降级 2.1，结果下载到工作区展示。当用户要求画图、生图、生成配图/海报/插画/素材，或改图、换风格、合成参考图时使用。

Generate images with Gemini from reference images - the same character across several pictures, a real product screenshot placed inside a generated scene, style matching - and debug Gemini image calls that return no image. Use when writing or fixing code that sends reference images to a Gemini image model, when a call returns an empty result or a content-filter finish reason, or when planning how many references to attach.

生成或编辑图片。适用于画图、生图、海报、封面、插画、头像、壁纸、产品图、参考图修改、换背景、改配色、局部重绘及连续修图；用户无需提及技术鸭或 jishuya。通过技术鸭图片 API 执行。仅分析图片、撰写提示词、搜索已有图片或修改网页/CSS/现有矢量代码时不触发；用户明确指定其他图片服务时遵循其选择。

Generate or edit images with the local `imagegenkit` command (ImageGenKit). Use it whenever the user asks to create, generate, render, mock up or edit an image, or a task needs a raster image asset (photos, product shots, thumbnails, mockups, icons, illustrations, backgrounds, app/store assets, transparent PNGs, restyles, combining reference images).

Generate and edit images with the gpt-image CLI using OpenAI GPT Image. Use for local GPT image generation, reference-based edits, masks, transparent cutouts, and saving generated assets to disk. This skill operates the CLI; it is not for implementing an image-generation application.

Generate or edit raster images (illustrations, photos, icons, logos, mockups, posters, UI art, textures, sprites, product shots, diagrams-as-art) by delegating to Codex's image generation tool via gpt-6.1-sol. Claude cannot generate images itself, so load this whenever the user asks to create, draw, render, generate, redesign, retouch or vary an image, or a project needs image assets. Teaches the detailed art-direction brief Codex follows literally, the edit/iteration loop, and verification. Uses the codex skill's script.

Le pasa fotos a Codex (OpenAI) para que genere imágenes nuevas o corrija/edite fotos con su generador integrado (gpt-image-2) y trae los resultados a una carpeta. Usar cuando el usuario diga 'pasale esto a Codex', 'generá con Codex', 'que Codex corrija esta foto', 'editá esta foto con Codex / con OpenAI / con gpt-image', 'sacale X a la foto con Codex', o pida imágenes 'con el generador de Codex'. Also triggers on 'generate/edit this image with Codex'. No hace video.

Generate or edit raster images with an OpenAI-compatible Images API (default model gpt-image-2). Use when the user asks for image generation, image edits, GPT Image, gpt-image-2, or image outputs that should be saved into the current project.

使用 OpenAI 官方 Images API 產生靜態圖片與原生動畫 GIF。當使用者要求繪圖、畫圖、生圖、圖片生成、動畫貼圖、GIF、animated GIF、text-to-image、GPT Image 2.6、Sunburst、Flare，或要在 Codex / GitHub Copilot 呼叫 OpenAI 繪圖模型時使用。以 uv 執行隨附的零第三方依賴 Python CLI；GIF 直接由模型輸出，不逐格合成。

编写、改写和检查 MiniMax-H3-Image 文生图及图像编辑提示词。用户说“H3提示词”“H3生图”“H3图像编辑”“按H3手册写”“人物一致性”“多图参考”“换装换姿势”且指向 H3 图像模型时使用；支持1至16张参考图分工、局部修改、换背景、删除补全、海报文字、网格多视图、写实摄影和概念设计。依据《MiniMax-H3-Image 使用手册》组织任务、布局、变化及保留项，默认给可直接粘贴的中文提示词；按要求给英文或结构化版本。不用于H3视频、Pruna P-H3-Image或其他模型的专有语法。

Produce y adapta piezas gráficas publicitarias (banners de homepage, banners display, posts e historias de redes, fondos de carrusel) a partir de editables de agencia (PDF de Illustrator, PSD, presentaciones de Google Slides con key visuals) y entrega siempre una galería HTML de revisión para dar feedback. Úsala cada vez que el usuario pida crear, redimensionar, readaptar, corregir o reexportar banners, artes, piezas, key visuals o formatos de medios (970x250, 300x600, 320x50, 1080x1080, 1080x1920, 1920x1080, etc.), aunque no diga "skill" ni "galería": por ejemplo "hazme los banners de esta campaña", "pasa estos PSD a historia de Instagram", "corrige el precio en todas las piezas" o "el c...

Писать и чинить промпты для картиночных моделей GPT Image 2.5, Nano Banana 2.1 и Seedream 5. Использовать, когда нужен промпт для генерации или правки картинки одной из этих моделей, в том числе когда агент сам готовит его перед генерацией, и когда готовая картинка не слушается промпта.

Distill a supplied image into a standalone abstract visual mark using source-specific geometry, flat inks, and subtle print texture. Use for image-derived abstract logos, 从图片提炼抽象标识, 抽象视觉标识, or photo-to-symbol requests.

Transform a memory, sentence, emotion, story, object, place, or source image into one cohesive fictional vinyl release artifact containing a front sleeve, back sleeve, vinyl Side A, and vinyl Side B. Use when generating a believable 4:3 landscape product photograph of an independent-label record whose visual language, typography, printing history, packaging, and physical materials express the source concept. Deliver the record design image, a portable 3D turntable HTML page, and a 10-second vertical MP4 from that same scene.

Turn a written adventure location (PDF pages or text) plus rough mapping guidelines into a 3D Blender model and a print-ready isometric cutaway map poster in a screenprint style, with GM and player versions. Use when asked to map, model or make a poster or battle map of a dungeon, bunker, ship, station or building from an RPG adventure or any written description of a place.

Paint images (PNG/JPG/WebP) yourself with code instead of calling an image-generation model — photoreal 3D (path tracing), oil/watercolour painting, comics and engravings from 3D, people and animals, landscapes (terrain, sky, sea, trees), posters with typography, pixel art, generative abstracts. Use when asked to create, draw, paint or render an image.

Turns your coding agent into a launch checker. It reads your live site and the repo behind it, checks what the page promises against what the code does, scores launch readiness out of 100 and ranks the three fixes that matter. Use before a launch, a Show HN or a Product Hunt post, or for "is my site ready to launch", "will my signups be tracked", "check my SEO and social card".

Transcribe PDFs, scans, receipts, invoices, and images with PennyOCR. Use when a user wants page text, Markdown tables, page citations, or an OCR cost estimate.

Generate images from a text prompt, or edit/combine reference images, via the self-hosted OpenRouter image relay. Use when the user asks to create, draw, render, illustrate, or edit an image.

Draw anything as a code-generated, hand-drawn-looking line illustration with the doodle engine bundled in this skill (plain HTML + JS, no build). Line first: every doodle is a finished line drawing (?outline=1) before colour. Covers objects, products, people, animals, plants, flowers, buildings, interiors, landscapes, night and weather scenes, dioramas, storyboards and lettering, in marker or pen-and-ink with hatch, cross-hatch, ribs and stipple, in perspective or axonometric. Each form has a me

摄影保真调色。用于把「调色风格」变成可复用的 3D LUT 并批量应用，支持天空/人像/植被等语义分区、批次级自动沉淀与撤销、题材自动识别与选风格，并在没有参考图时用经典场景统计自查成片、识别素材是否为 log 未转色。只做逐像素色彩映射，不重绘、不改内容、不改变几何，适合对保真有硬要求的摄影交付。

Create platform-ready thumbnails from real video frames. Use when the user wants a thumbnail, cover image, YouTube cover, Shorts cover, Bilibili cover, Xiaohongshu cover, or other video poster image.

按原图尺寸和构图转绘白底 Y2K 低清像素涂鸦，五官与全图共用一致的粗糙线条，保留表情，配松散涂色和微型贴图，并按主体与风景分配结构线密度，也可编写提示词。用于鼠标手绘、笨拙粗糙的像素画法。