Image

Accept images or PDFs and output a typed, searchable PDF while preserving the original page layout, short lines, whitespace, side notes, colors, tables, diagrams, and mathematical notation. Use for digitizing handwritten notes, photos, scanned PDFs, or mixed printed/handwritten pages rather than summaries or redesigned notes.

Reverse-engineer a reference image into a visible extraction report and reusable prompts. Extracts style, lighting, lens and optics, color palette and grading, composition, materials and post-processing, each value tagged as observed, inferred or unknown. 根据图片提取风格、光线、镜头、调色色调与构图材质，出结构化反推报告与可复用提示词。 Four modes: extraction report, high-fidelity recreation prompt, style-only extraction with the subject stripped for reuse on a new subject, and person-swap recreation that rebuilds an image with a...

Gives Blender objects materials that actually render (Principled BSDF, glass, metal, emission, procedural textures), sets up lighting, world, camera and render settings, and produces a final image through a Blender connection (the official BlenderLab server). Use when the user asks for materials, colors, textures, lighting, a camera or a render of a Blender scene, or when a render comes out black, grey, flat or noisy.

Builds and edits 3D models in the user's Blender through a Blender connection (the official BlenderLab server) - inspecting the scene, writing bpy code in small verified steps, real-world scale, clean naming, modifiers and checking the result with viewport screenshots. Use when the user asks to model, build, create, fix or change an object or a scene in Blender, or to make a 3D model from a description or a reference image.

Builds stylized low-poly models and scenes in Blender through a Blender connection (the official BlenderLab server) - chunky faceted shapes with few polygons, flat shading, a small color palette of plain materials, exaggerated proportions, and the soft lighting the style needs. Use when the user asks for low poly, faceted, stylized, cartoon or game-style models (a house, tree, rock, terrain, vehicle, prop, diorama) or wants a model to look like a low-poly reference image.

Turns an image into a 3D model (GLB) with the user's local ComfyUI - choosing the right saved workflow for the image, running it with the values the user saved, picking the generator when a workflow offers two, lowering the face count only when the user asks for a lighter model, and reporting the result honestly. Use when the user attaches or names an image and asks for a 3D model, mesh or GLB of what it shows, or asks to reconstruct an object in 3D from a picture. Needs the comfy_workflows...

> 将 Markdown/纯文本内容智能分析、排版并导出为高分辨率精美长图或分享海报（内置 18 种视觉风格）。 当用户说「生成长图」「做海报」「文字转图片」「高颜值排版」「做大图」「make a poster」 「export as image」「generate image from text/markdown」等意图时触发本技能。

Search oip-visual-v2 for traceable prompt-image pairs and open local result galleries. Use when finding, inspecting, comparing, browsing, or copying real image-generation references or exact source prompts. After references are selected, adapt them into a derived bilingual prompt when requested; use img-gen-taste first when the visual direction is still unclear.

Develop art direction for image generation with curated style cards distilled from Open Image Prompts. Use when choosing or comparing visual directions, improving an existing generation prompt, or translating a subject, use case, or supplied image into a compact creative spec. Use img-gen-prompts for real archive examples, exact source prompts, or gallery browsing.

Generate and revise art-directed website design references with GPT Image 2.6 or the host image tool. Use for visual hero, landing page, section, desktop or mobile mockups, same-site continuations, responsive recompositions, and precise edits to approved frontend comps.

Turn a technical input (an experiment and its results, a paper, a pipeline, an algorithm, a system change, a benchmark) into ONE story-style explainer slide — a 16:9 image of 3–6 numbered cards (discovery → problem → method → where the numbers come from → result + lesson) that a junior engineer understands at first read, with traceable numbers, a visible WHY per method step, and an honest result. Use this whenever the user asks for "one slide", "a slide that explains", "a story image", "an...

Design, check and export 3D parts in code with CadQuery and agentcad, measuring the built geometry instead of trusting parameters. Use on any task that designs, modifies, checks or exports a 3D part, assembly or print in code (CadQuery, STEP, STL, 3MF), including "make a bracket/enclosure/mount/case", fitting parts together, checking printability, or reviewing CAD another agent produced.

将源PPT按先内容规划、3–5张统一生图母板、逐页整页视觉稿、Canva拆解重建、可编辑PPT整合的流程升级。适用于竞赛路演PPT及前5页试制；页数可配，动画与GIF在全套页面重建后统一处理。

Use when a stack or architecture diagram is requested.

根据两张照片生成“为你拍照”上下拼贴：图2仅裁切和等比缩放为横图，只将图1人物生成在图2场景中为对方拍照的第三视角。用于为你拍照、拍摄者与被拍摄者、情侣摄影上下拼图等请求。

Use when refining or polishing existing pixel art in Aseprite with the aseprite MCP tools - dithering (ordered/Bayer, patterns, gradients), anti-aliasing curves, hue-shifted shading, colour ramps, limited or retro palettes, contrast and readability, or material textures like metal, stone, wood, fabric and glass.

Structure and space in product UI on the web. It sets the spacing scale, grouping, alignment, density (compact or comfortable), hit areas, app shells (sidebar, toolbar, content, inspector), max widths, container-query responsiveness, dropdown and popover sizing, room for long and translated text, layout stability (no shift on load) and sticky elements that must not cover focus.

Turn a photo of a phone into a reusable template and put any image on its screen, by code, with no image generator. Use for a new phone photo, a new template, or checking one.

Use when drawing a still pixel-art sprite, scene, icon or tile in Aseprite with the aseprite MCP tools: planning size and palette, blocking shapes, shading, outlining and checking the result.

Locate a thing in an image. Ask h2oai/h2o-lightning-4b for a bounding box around a referring phrase ("the total amount", "the red square", "the coffee mug") and get fractions of the image, pixel coordinates, a calibrated confidence (P(IoU >= 0.5)) and a `present` probability that gates the answer.

Turn a theme, a brief, or a passage of text into a finished collage artwork. Use when the user asks for collage art, a collage poster, a cut-paper composition, mixed-media / papier collé / photomontage / décollage styling, or wants visual imagery derived from a piece of writing.

Draws flowcharts and state diagrams as paper-style posters (cream paper, typewriter type, hatched shadows, right-angle connectors, problems ringed in red) and exports HTML, PNG and SVG. Use when the user asks for a flowchart, process or workflow diagram, state diagram, pipeline or experiment-plan figure, or wants Mermaid turned into a polished image.

Index a folder of images, audio, or text with a local multimodal embedding model and search it in natural language. Builds a resumable vector index through any OpenAI-compatible /v1/embeddings server that accepts multimodal content parts, such as llama.cpp's llama-server running EmbeddingGemma 2, then ranks files by semantic similarity with color filtering, multi-facet AND queries, JSON output, and HTML contact sheet…

Use when the user wants Claude to draw or edit in the Aseprite window they have open, asks why Aseprite shows no change, or mentions live mode, the open file, an unsaved tab, or undoing Claude's steps in Aseprite.