Vision setup for streamers (Pro+)

Give your AI co-host real eyesight. This Pro+ checklist wires OBS captures into multimodal models so reactions match exactly what’s on screen.

Total time · 12Difficulty · IntermediateHostFabric Pro or Platform ProOBS 28+ with WebSocket enabledGame/scene source
01 · Confirm you’re on Pro+
1 min

Vision is gated to Pro/Platform Pro (or Vision add-on). If you see a Vision upsell in Control → Billing, upgrade first so captures aren’t blocked.

  • Control → Billing → Upgrade to Pro or Platform Pro
  • Refresh Control to clear cached tier
02 · Enable OBS WebSocket
2 min

In OBS, open Tools → WebSocket Server Settings. Turn it on, set a strong password, and note the port (default 4455).

  • Toggle “Enable WebSocket server”
  • Set password + confirm port 4455
  • Apply & restart OBS if prompted
03 · Drop in vision env settings
2 min

Populate OBS connection in your environment or project settings: URL (e.g., ws://127.0.0.1:4455) and the WebSocket password. Optional: set OBS_CAPTURE_SOURCE to the scene/source name if you want to force a specific view.

  • Set OBS_WEBSOCKET_URL + OBS_WEBSOCKET_PASSWORD
  • Optional: OBS_CAPTURE_SOURCE for a dedicated scene
  • Redeploy/restart if needed so env vars take effect
04 · Place the overlay (unchanged)
2 min

Use the existing overlay URL as a Browser Source at 1920×1080. Vision rides on the same overlay connection—no new browser source needed.

  • Control → Overlay → Copy URL
  • OBS: Add Browser Source (1920×1080, 60 FPS)
  • Lock layer above gameplay
05 · Test a capture
3 min

In Control, send a vision-friendly prompt: “What do you see?” or use a /capture trigger. Watch logs for a successful capture and verify the response references on-screen visuals.

  • Send prompt containing “look/see/check out”
  • Watch for capture success and latency under 3s
  • Confirm AI references the actual scene (not generic text)
06 · Tune safety & rate limits
2 min

Keep captures clean: Vision auto-limits to 1 frame per 2s and 10/min per room. Avoid PII on screen; keep overlays tidy to reduce hallucinations.

  • Close unnecessary windows
  • Keep scene text readable (avoid tiny HUDs)
  • Leave room for overlay speech bubbles
Fast path: run a rehearsal with a puzzle game or lobby screen—easy to verify that the AI calls out what it actually sees before showtime.