菜单
Generate AI graphics (banners, illustrations, hero images, social cards, etc.) using Fal AI's Nano Banana 2. Use when the user asks to create, generate, or make an image, banner, illustration, graphic, or artwork. Always render at 1K resolution.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Batch-stamp copyright, artist, and rights information on all images in a directory. Use when you need to quickly add copyright metadata to a photo library before distribution or archival — a convenience wrapper around exiftool.
Decode JPEG XL images (.jxl) to PNG or JPEG. Use when you need to convert JXL files back to standard formats for compatibility, display, or further editing.
Extract embedded JPEG and PNG images from a PDF at their native resolution, without rasterizing. Use when a PDF is primarily a wrapper around photos or images and you want to recover them at full quality without the page rendering overhead.
Rasterize PDF pages to numbered image files (JPEG, PNG, or TIFF) at a specified resolution (DPI). Use when you need to convert PDF content to standalone images — one image per page — with optional page range selection.
Read and display EXIF, IPTC, XMP, and other metadata from images as structured JSON. Use when you need to inspect image metadata without modifying it — opposite of scrub-metadata, which removes metadata.
Write EXIF, IPTC, and XMP metadata tags to image files. Use when you need to add or update metadata like artist name, copyright, description, keywords, or GPS coordinates on individual images.
| name | ai-graphics |
| description | Generate AI graphics (banners, illustrations, hero images, social cards, etc.) using Fal AI's Nano Banana 2. Use when the user asks to create, generate, or make an image, banner, illustration, graphic, or artwork. Always render at 1K resolution. |
Generate AI graphics — banners, hero images, illustrations, social cards, README artwork — using Fal AI's Nano Banana 2 model via the nano-tech-diagrams MCP server.
Always use mcp__jungle-shared__nano-tech-diagrams__* (runs on ubuntuvm).
Do not use jungle-local — it claims to accept local filesystem paths but runs in an environment that cannot read the workstation's filesystem, so every local-path call fails with ENOENT. The shared instance is the working path; pair it with the staging endpoint below for any image input.
resolution: "1K". Do not use 0.5K, 2K, or 4K unless the user explicitly overrides. 1K is the standard for this workflow.aspect_ratio for the use case:
21:9 or 16:916:91:12:3 or 9:16download_to (e.g., assets/banner.png, images/hero.png) — don't just leave the fal.media URL floating.output_format to png unless the user asks otherwise.Nano Banana 2 rewards detailed, descriptive prompts. Include:
Keep prompts one dense paragraph rather than a bullet list.
mkdir -p via Bash).mcp__jungle-shared__nano-tech-diagrams__text_to_image with the prompt, aspect ratio, and output_format.fal.media URL. Do not rely on the MCP's download_to parameter — the ubuntuvm process cannot write to the workstation filesystem. Instead, capture the URL and download it yourself:
curl -sSL -o "/absolute/target/path.png" "<fal.media URL>"
Example:
mcp__jungle-shared__nano-tech-diagrams__text_to_image(
prompt="<detailed paragraph>",
aspect_ratio="21:9",
output_format="png"
)
# then: curl -sSL -o /abs/path/banner.png <returned URL>
Local input files must be staged to S3 first — the MCP cannot read local paths directly, but it accepts http(s) URLs.
mcp-staging S3 bucket:
URL=$(s3-stage /local/path/to/input.jpg)
s3-stage returns a presigned GET URL (default 1-hour TTL).mcp__jungle-shared__nano-tech-diagrams__image_to_image with image_path="$URL", plus your prompt, aspect ratio, and output_format. Use dictionary_words for any text that must be spelled correctly.fal.media URL. Download to the target folder:
curl -sSL -o "/absolute/target/path.jpg" "<fal.media URL>"
Staged objects expire after 1 day (bucket lifecycle), so you don't need to clean them up.