gemini-upscale
Upscales photos and images to a target size from 1K up to 8K (the default) and beyond with the gemini-media MCP server. It plans the cheapest set of crops that covers the target (one tile up to about 4K, a grid of 4K tiles beyond), re-renders each with Nano Banana 2.1 using the whole photo as context, aligns and blends them into one large PNG, then optionally re-renders a few regions to fix defects such as mismatched eyes. Use it when the user wants to upscale, enlarge or add detail to a photo for print, wallpapers or large screens, even if they never mention Gemini. Limits - it invents plausible detail rather than recovering the original, so faces, eyes, small text and textures can change; it is unsuitable for forensic, archival, evidence or product-label work; a run costs about $0.04 at 1K to $0.50-1.20 at 8K on nb2.
- Version
- 1.1.0
- License
- Apache-2.0
- Compatibility
- Requires the gemini-media MCP server v1.1 or later (tile_image and stitch_tiles tools) with a Gemini API key or Vertex AI project.
Pinned to revision 405d92cf84d6, so it is the text this page describes rather than whatever the author pushed since.
Files
- skills/gemini-upscale/SKILL.md
- skills/gemini-upscale/evals/evals.json
- skills/gemini-upscale/evals/trigger-queries.json
- skills/gemini-upscale/references/tiling-guide.md
Every link opens the file at its source, pinned to the revision this page describes.