Query local images with a Gemma 4 vision model via Ollama.
npx skills add https://github.com/gridaco/grida --skill vision-gridaco
Please help me install this Agent Skill. Skill: vision Source: https://github.com/gridaco/grida/tree/main/.agents/skills/vision Command: npx skills add https://github.com/gridaco/grida --skill vision-gridaco
Query images with a local Ollama vision model without loading the image into the main agent context. Use when you need to describe a screenshot, check whether rendered content is present, detect overlapping elements, or ask any visual question about a PNG/JPEG/WebP file.
Describe a target image by running the local vision script against a file to obtain a concise summary of its content.
High-intent search queries and answers about installing and using this skill.
Local image analysis via Gemma 4 through Ollama processes PNG, JPEG, and WebP files directly on your machine. This on-device approach describes screenshots and detects UI overlaps without exposing data to external servers.
Detecting overlapping UI elements in a PNG screenshot involves running a local vision script against the file. The Gemma 4 model evaluates the rendered canvas and produces textual results identifying layout and overlap issues.
Yes, Ollama is a required dependency to run local image description workflows. The process relies on a local Ollama server and the Gemma 4 vision model to process PNG, JPEG, and WebP assets and generate textual summaries.
Yes, you can verify if rendered content is present in a JPEG file by querying it with a local vision model. The Gemma 4 model via Ollama supports visual question answering and concise scene description for assets.
The primary limitation of using Gemma 4 for visual QA is that it requires a local Ollama server to be running. Additionally, it generates concise textual descriptions rather than modifying the underlying PNG, JPEG, or WebP files.
agongdaiCompress PNG, JPG/JPEG, and WebP images recursively under the project root.
CommunityToolkitCapture Playwright screenshots and embed them in GitHub PR descriptions.
MadAppGangCapture screenshots and inspect DOM state to diagnose UI layout issues.
liumeixinConfigure custom vision backends for Hermes Agent with base64 image handling.
