DEDUPED Hot AI repos. Real use cases. What people build. Get the Monday 5

How to ask typed questions about images and video frames with Start Lux Decision

Send an image as the state with choice, noul and score questions, as in the README's frame-by-frame demos.

Short answer: Start a decision server (vLLM with startlux-decide serve, or Prism); then open examples/ for the request with an image; send one frame with your typed questions, for example a choice for the stage of an event and a noul for whether to alert. All steps.

Who it is forA developer monitoring camera feeds or screenshots who needs structured answers, not captions.

The problemCaptioning models return free text that is hard to act on reliably.

The outcomeProbabilities for every option of every question about an image, ready to drive automation.

What you need first

  • Python environment that can install the package with vLLM support, or a Bamboo API key for the hosted model
  • A GPU for the local vLLM setup; the README numbers were measured on one H200
  • Disk space and network access to download a model from Hugging Face

Steps

  1. Start a decision server (vLLM with startlux-decide serve, or Prism)
  2. Open examples/ for the request with an image
  3. Send one frame with your typed questions, for example a choice for the stage of an event and a noul for whether to alert
  4. Read the probabilities and set your thresholds
  5. Send one frame a second for video, as the rescue and launch demos do

Prompt for your agent

Using StartLuxLabs/StartLux-Decision and the image request in examples/, send frames from <video or camera> once a second with a choice question for <stages> and a noul for <alert condition>, and log only frames where the alert probability is above 0.8.

Caveat. This card is written from the StartLuxLabs/StartLux-Decision README, not from a test run by Deduped. Before you start, make sure you have Python environment that can install the package with vLLM support, or a Bamboo API key for the hosted model. Check the repo for newer instructions before you rely on these steps.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for StartLuxLabs/StartLux-Decision and "Ask typed questions about images and video frames"

Add Deduped to Claude Code once

claude mcp add --transport http deduped https://dedupedai.com/mcp

Cursor, ChatGPT, others

Share: X LinkedIn Hacker News Email Copy link

Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer