Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.daydream.live/llms.txt

Use this file to discover all available pages before exploring further.

​
Sending parameters to a running pipeline

After establishing a WebRTC connection, you can send real-time parameter updates via the data channel. This allows dynamic control of generation without reconnecting.

​
Overview

Parameters are sent as JSON messages through the WebRTC data channel. Updates take effect on the next chunk that is generated.

​
Setting Up the Data Channel

The data channel is created when establishing the WebRTC connection: 
// Create data channel before creating offer
const dataChannel = pc.createDataChannel("parameters", { ordered: true });

dataChannel.onopen = () => {
  console.log("Data channel ready for parameter updates");
};

dataChannel.onmessage = (event) => {
  const data = JSON.parse(event.data);

  // Handle notifications from server
  if (data.type === "stream_stopped") {
    console.log("Stream stopped:", data.error_message);
    pc.close();
  }
};

​
Sending Parameter Updates

function sendParameters(params) {
  if (dataChannel.readyState === "open") {
    dataChannel.send(JSON.stringify(params));
  } else {
    console.warn("Data channel not ready");
  }
}

// Example: Update prompt
sendParameters({
  prompts: [{ text: "A cat playing piano", weight: 1.0 }]
});

// Example: Update multiple parameters
sendParameters({
  prompts: [{ text: "A sunset over the ocean", weight: 1.0 }],
  denoising_step_list: [800, 600, 400],
  noise_scale: 0.7
});

​
Available Parameters

​
Prompts

Control what is being generated: 
// Single prompt
sendParameters({
  prompts: [{ text: "A beautiful forest", weight: 1.0 }]
});

// Blended prompts (spatial blending within frame)
sendParameters({
  prompts: [
    { text: "A sunny day", weight: 0.7 },
    { text: "A rainy day", weight: 0.3 }
  ],
  prompt_interpolation_method: "linear"  // or "slerp"
});
ParameterTypeDefaultDescription
promptsarray-Array of { text: string, weight: number }
prompt_interpolation_methodstring"linear""linear" or "slerp" for blending

​
Prompt Transitions

Smoothly transition between prompts over multiple frames: 
sendParameters({
  transition: {
    target_prompts: [
      { text: "A night sky with stars", weight: 1.0 }
    ],
    num_steps: 8,  // Transition over 8 chunks
    temporal_interpolation_method: "linear"
  }
});
ParameterTypeDefaultDescription
transition.target_promptsarray-Target prompts to transition to
transition.num_stepsint4Frames to transition over (0 = instant)
transition.temporal_interpolation_methodstring"linear""linear" or "slerp"

​
Denoising Steps

Control quality vs speed tradeoff: 
// More steps = higher quality, slower
sendParameters({
  denoising_step_list: [1000, 750, 500, 250]
});

// Fewer steps = faster, lower quality
sendParameters({
  denoising_step_list: [700, 400]
});
ParameterTypeDescription
denoising_step_listarrayDescending timesteps (e.g., [1000, 750, 500])

​
Noise Control

sendParameters({
  noise_scale: 0.8,      // 0.0-1.0, amount of noise
  noise_controller: true  // Auto-adjust based on motion
});
ParameterTypeRangeDescription
noise_scalefloat0.0-1.0Manual noise amount
noise_controllerbool-Enable automatic noise adjustment

​
Cache Control

// Enable automatic cache management
sendParameters({
  manage_cache: true
});

// Or manual cache reset
sendParameters({
  reset_cache: true  // Trigger one-time cache reset
});
ParameterTypeDescription
manage_cacheboolAuto cache management
reset_cacheboolForce cache reset (one-shot)

​
Playback Control

// Pause generation
sendParameters({ paused: true });

// Resume generation
sendParameters({ paused: false });

​
LoRA Scale Updates

Update LoRA adapter scales at runtime (requires lora_merge_mode: "runtime_peft" at load time): 
sendParameters({
  lora_scales: [
    { path: "/path/to/style.safetensors", scale: 0.5 },
    { path: "/path/to/character.safetensors", scale: 1.2 }
  ]
});

​
VACE Parameters

Control strength of visual conditioning: 
sendParameters({
  vace_ref_images: ["/path/to/reference.png"],
  vace_context_scale: 1.0  // 0.0-2.0
});
See VACE for detailed VACE usage.

​
Spout (Windows)

Enable Spout output for sending frames to external applications: 
sendParameters({
  spout_sender: {
    enabled: true,
    name: "ScopeOutput"
  }
});

// Receive from Spout instead of WebRTC input
sendParameters({
  spout_receiver: {
    enabled: true,
    name: "ExternalApp"
  }
});

​
Initial Parameters

You can also send initial parameters when establishing the WebRTC connection: 
const response = await fetch("http://localhost:8000/api/v1/webrtc/offer", {
  method: "POST",
  headers: { "Content-Type": "application/json" },
  body: JSON.stringify({
    sdp: offer.sdp,
    type: offer.type,
    initialParameters: {
      prompts: [{ text: "Initial prompt", weight: 1.0 }],
      denoising_step_list: [1000, 750, 500, 250],
      manage_cache: true
    }
  })
});

​
See Also

Receive Video

Set up WebRTC connection

VACE

Reference image conditioning

Load Pipeline

Configure pipeline at load time

Spout Guide

Using Spout for external apps