Find objects across batch images directly via Ollama vision model (huihui_ai/qwen2.5-vl-abliterated:7b) and visualize coordinates automatically.
Inference connection
Use /ollama to follow the Pinpoint server’s OLLAMA environment or .env setting. Enter an absolute HTTP(S) origin to connect this browser directly. The override is stored only in this browser.
Exposedness applies only to anatomical searches. Partially Exposed includes bare skin and shape-revealing tight clothing; General also includes loose clothing, while Fully Exposed requires visible skin. Images use a model-aligned pixel budget with no artificial white padding. Precise Grounding is optional; when enabled, suspicious boxes get one enlarged crop pass, with multiple boxes refined in parallel. View Check normally arrives in the initial detection response.
A preset saves the search prompt and a folder path. Loading it replaces the current images with every supported image in that folder.
Tip: You can search for general objects (person running) or specific features (runner with headband, purple shirt, bib number).
Download a request with your selected images, run this PowerShell command in the download folder, then paste the returned response text into Model Response.
Prompt text: