Skip to main content
Decart is a real-time AI video transformation service that applies artistic styles and effects to video streams. The Decart plugin enables you to transform your agent’s video feed with AI-powered restyling, creating immersive visual experiences like animated styles, cyberpunk aesthetics, or any custom prompt-based transformation. The Decart plugin for the Vision Agents SDK provides real-time video processing that transforms incoming video streams and publishes the styled output back to the call.

​
Features

  • 🎨 Real-time Video Restyling: Transform video with AI-powered style transfer
  • 🚀 WebRTC Streaming: Low-latency real-time video processing
  • 🔄 Dynamic Prompt Updates: Change styles on-the-fly during calls
  • 🪞 Mirror Mode: Optional mirroring for front-facing cameras

​
Installation

Install the Decart plugin with 
uv add vision-agents-plugins-decart

​
Example

Check out our Decart example to see working code samples using the plugin, or read on for some key details.

​
Initialization

The Decart plugin exists in the form of the RestylingProcessor class: 
from vision_agents.plugins import decart

processor = decart.RestylingProcessor(
    initial_prompt="Studio Ghibli animation style",
    model="mirage_v2"
)
To initialize without passing in the API key, make sure the DECART_API_KEY is available as an environment variable. You can do this either by defining it in a .env file or exporting it directly in your terminal.

​
Parameters

These are the parameters available in the Decart RestylingProcessor plugin for you to customize:
NameTypeDefaultDescription
api_keystr or NoneNoneYour Decart API key. If not provided, the plugin will look for the DECART_API_KEY environment variable.
modelstr"mirage_v2"Decart model name to use for video transformation.
initial_promptstr"Cyberpunk city"Initial style prompt describing the desired visual transformation.
enrichboolTrueWhether to enrich the prompt with additional details for better results.
mirrorboolTrueEnable mirror mode for front-facing cameras.
widthint1280Output video width in pixels.
heightint720Output video height in pixels.

​
How It Works

The Decart integration processes video in real-time:
  1. Video Input: Receives the user’s video stream from the call
  2. AI Processing: Sends frames to Decart’s Realtime API via WebSocket
  3. Style Transfer: Decart applies the prompt-based transformation to each frame
  4. Video Output: Publishes the transformed video back to the call
The processor maintains a persistent WebSocket connection to Decart’s API for low-latency processing.

​
Usage in Agent

Add the RestylingProcessor to your agent’s processors list: 
from vision_agents.core import Agent, User
from vision_agents.plugins import getstream, openai, elevenlabs, deepgram, decart

async def create_agent():
    processor = decart.RestylingProcessor(
        initial_prompt="A cute animated movie with vibrant colours",
        model="mirage_v2"
    )
    
    agent = Agent(
        edge=getstream.Edge(),
        agent_user=User(name="Story Teller", id="agent"),
        instructions="You are a story teller with magical video effects.",
        llm=openai.LLM(model="gpt-4o-mini"),
        tts=elevenlabs.TTS(),
        stt=deepgram.STT(),
        processors=[processor]
    )
    
    return agent

​
Dynamic Prompt Updates

You can change the video style during a call by updating the prompt. This is useful for creating interactive experiences where the visual style responds to the conversation: 
from vision_agents.core import Agent, User
from vision_agents.plugins import getstream, openai, decart

async def create_agent():
    processor = decart.RestylingProcessor(
        initial_prompt="Peaceful forest scene",
        model="mirage_v2"
    )
    
    llm = openai.LLM(model="gpt-4o-mini")
    
    agent = Agent(
        edge=getstream.Edge(),
        agent_user=User(name="Visual Guide", id="agent"),
        instructions="You guide users through different visual environments.",
        llm=llm,
        processors=[processor]
    )
    
    # Register a function to change the video style
    @llm.register_function(
        description="Changes the visual style of the video background"
    )
    async def change_style(prompt: str) -> str:
        await processor.update_prompt(prompt)
        return f"Video style changed to: {prompt}"
    
    return agent
Now your agent can dynamically change the video style based on the conversation: 
User: "Can you make it look like a cyberpunk city?"
Agent: *calls change_style("Cyberpunk city with neon lights")*
       "I've transformed the scene into a cyberpunk city!"

​
Advanced Configuration

​
Mirror Mode

Control whether the video is mirrored (useful for front-facing cameras): 
processor = decart.RestylingProcessor(
    initial_prompt="Watercolor painting style",
    mirror=True  # Enable mirroring
)

# Or update dynamically
await processor.set_mirror(False)

​
Custom Resolution

Adjust the output video resolution based on your needs: 
processor = decart.RestylingProcessor(
    initial_prompt="Oil painting style",
    width=1920,   # Full HD width
    height=1080   # Full HD height
)
Higher resolutions require more bandwidth and processing power. The default 1280x720 provides a good balance between quality and performance.

​
Prompt Engineering Tips

For best results with Decart video transformations:
  • Be specific: “Studio Ghibli animation with soft colors” works better than just “animated”
  • Include style details: Mention artistic styles, color palettes, or visual effects
  • Use descriptive language: “Cyberpunk city with neon lights and rain” creates more vivid results
  • Experiment: Try different prompts to find what works best for your use case
Example prompts:
  • "Studio Ghibli animation style with vibrant colors"
  • "Cyberpunk city with neon lights and holographic displays"
  • "Watercolor painting with soft pastel colors"
  • "Retro 80s aesthetic with VHS effects"
  • "Fantasy medieval castle with magical atmosphere"

​
Troubleshooting

​
Connection Issues

If you experience connection problems:
  • Verify your Decart API key is valid
  • Ensure network access to Decart’s servers
  • Check firewall settings for WebSocket traffic

​
No Transformed Video

  • Check browser console for errors
  • Verify Decart API key has proper permissions
  • Ensure the processor is added to the agent’s processors list