streamware

Describe Scene Example

Detailed scene descriptions using LLM (llava).

Quick Start

# Describe what's happening
sq watch "describe what's happening"
sq watch "opisz co się dzieje"

# Detailed analysis
sq watch "slow detailed analysis"
sq watch "szczegółowo opisz scenę"

Features

• LLM-powered: Uses llava:7b for rich descriptions
• Full scene analysis: Objects, activities, atmosphere
• Guarder filtering: Clean, relevant responses

Configuration

# .env settings for describe mode
SQ_STREAM_FPS=0.2          # 1 frame every 5 seconds
SQ_YOLO_SKIP_LLM_THRESHOLD=1.0  # Always use LLM
SQ_USE_GUARDER=true
SQ_MODEL=llava:7b

Output Examples

"A person is sitting at a desk working on a laptop. 
The room appears to be an office with natural lighting 
coming from a window on the left."

"Two people are having a conversation near the entrance. 
One is holding a coffee cup. The area looks like a 
building lobby with modern furniture."

Python API

from streamware.intent import parse_intent, apply_intent
from streamware.core import flow

# Parse intent
intent = parse_intent("describe scene")
print(f"LLM: {intent.llm}")        # True
print(f"Model: {intent.llm_model}") # llava:7b

# Build URI
uri = f"live://narrator?source={url}"
uri += "&mode=full"
uri += "&tts=true"
uri += "&duration=60"

result = flow(uri).run()

LLM Models

Model	Speed	Quality
moondream	Fast	Basic
llava:7b	Medium	Good
llava:13b	Slow	Best

This site is open source. Improve this page.