> ## 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.
# Quick Start
> Get Scope running and generate your first AI video
# Get Started with Scope
Daydream Scope is an open-source tool for running and customizing real-time interactive generative AI pipelines and models.
Watch this 10-minute walkthrough covering the main Scope features and how to get your first generation running:
Follow the steps below to install Scope and create your first generation.
***
## Prerequisites
**For Desktop App or Local Installation:**
* NVIDIA GPU with ≥24GB VRAM (RTX 4090/5090 or similar)
* CUDA 12.8+ drivers
* Windows, Mac\*, or Linux
**\*Mac / No powerful GPU?** Mac does not currently support NVIDIA GPUs, but Daydream Scope Cloud lets you run inference remotely - right from the app - without needing a local GPU. This means Mac users and anyone without a dedicated NVIDIA card can still use most of Scope's capabilities. See the [Remote Inference guide](/scope/guides/remote-inference) to get started.
**For Cloud (RunPod):**
* RunPod account with credits
* Similar GPU requirements apply to your instance selection
View detailed hardware specs, pipeline-specific VRAM needs, and software dependencies
**Krea Realtime** requires ≥32GB VRAM and uses the **fp8 quantized model** at that tier (e.g. RTX 5090). For higher resolutions without quantization, ≥40GB VRAM is recommended (e.g. H100, RTX 6000 Pro). A 24GB GPU like the RTX 4090 cannot run Krea Realtime - use LongLive or StreamDiffusion V2 instead.
***
## Step 1: Install Scope
Choose your installation method:
**Best for:** Windows and Mac users who want the easiest installation experience
The Daydream Scope desktop app is an Electron-based application that provides the simplest way to get Scope running on your machine. It's available for both Windows and macOS (Apple Silicon).
Download the latest Windows installer (.exe). This link always points to the most recent release.
Download the latest macOS installer (.dmg) for Apple Silicon. This link always points to the most recent release.
**Windows:** Run the downloaded `.exe` file and follow the standard Windows installation prompts.
**Mac:** Open the downloaded `.dmg` file and drag Daydream Scope to your Applications folder.
**Windows:** Launch Daydream Scope from your Start menu or desktop shortcut.
**Mac:** Launch Daydream Scope from your Applications folder or via Spotlight.
Visit the [Daydream Scope releases page](https://github.com/daydreamlive/scope/releases) on GitHub. Select the release you want, expand the **Assets** section at the bottom, and download the file ending in `.exe` (Windows) or `.dmg` (Mac).
**Best for:** Developers who want to build on top of Scope - custom nodes, API integration, or running bleeding-edge code from the `main` branch before it ships in a release. You get full access to the Python environment and source code, but you manage dependencies (UV, Node.js, CUDA drivers) yourself. There are no auto-updates.
Verify that your NVIDIA drivers support CUDA 12.8 or higher:
```bash theme={null}
nvidia-smi
```
The output should show your GPU and a CUDA version of at least 12.8.
You'll need:
* **UV** (Python package manager)
* **Node.js** and **npm**
If you don't have these installed, visit their respective websites for installation instructions.
```bash theme={null}
git clone git@github.com:daydreamlive/scope.git
cd scope
```
```bash theme={null}
uv run build
```
This installs all required Python packages including Torch and FlashAttention. First-time install may take a while.
```bash theme={null}
uv run daydream-scope
```
On the first run, model weights will download automatically to `~/.daydream-scope/models`.
Open your browser and navigate to **[http://localhost:8000](http://localhost:8000)**
Local install is the recommended path for node development. See the node development guide to get started.
**Best for:** Researchers and developers without access to local high-end GPUs
RunPod is a third-party cloud GPU service. We've created a template to make deployment simple.
You'll need a RunPod account with credits. Create an account at [runpod.io](https://runpod.io) and add funds to get started.
**Prefer video?** Watch the RunPod tutorial:
Click to deploy: [**Deploy Daydream Scope on RunPod**](https://console.runpod.io/deploy?template=aca8mw9ivw\&ref=5k8hxjq3)
Required for TURN server functionality (WebRTC in cloud environments):
1. Create account at [huggingface.co](https://huggingface.co)
2. Navigate to **Settings > Access Tokens**
3. Create a token with **read** permissions
4. Copy the token
HuggingFace provides 10GB of free streaming per month via Cloudflare TURN servers. For more details on authentication methods and troubleshooting, see the [HuggingFace Auth](/scope/guides/huggingface) guide.
Choose a GPU with:
* **Minimum:** ≥24GB VRAM
* **Recommended:** NVIDIA RTX 4090/5090 or similar
* **CUDA:** 12.8+ support
1. Click **"Edit Template"**
2. Add variable: `HF_TOKEN` = your HuggingFace token
3. Click **"Save"**
Click **"Deploy On-Demand"** and wait for deployment to complete.
Open the app at **port 8000**: `https://your-instance-id.runpod.io:8000`
***
## Step 2: Your First Generation
Once Scope is running, open the interface at `localhost:8000` (or your RunPod URL).
### Text-to-Video
The LongLive pipeline is pre-selected with a prompt describing a 3D animated panda walking through a park. Just hit **play** - you'll see the generation running in real-time, frame by frame, with no render queue or progress bar.
Stop the generation, try changing the prompt to something completely different, and hit play again:
* "a dragon flying through clouds over a volcano"
* "a robot walking on mars"
* "an astronaut floating through a neon city"
The generation adapts in real-time - you're steering it live, not queueing up renders.
### Video-to-Video
Now try switching the input mode from **Text** to **Video**. A looping cat test video is loaded by default. Hit play and watch the model transform the video based on your prompt while preserving its structure and motion.
Experiment with different prompts to see how the same source video gets reinterpreted:
* "a cow sitting in the grass"
* "a fish sitting in the grass"
* "a dragon sitting in the grass"
You can also use your **webcam** as a live input, load your own **video file**, or receive video from other applications via **Spout** (Windows only).
### Explore Community Projects
See what others are creating with Scope:
Live camera feed restyled in real-time using VACE on an iPad
A seeded case study exploring StreamDiffusion to Scope transition
Explore more creations, download timelines, and share your own work on the Daydream Community Hub
***
## Step 3: Next Steps
Now that you're generating, here are some things to try:
Add style adapters to transform your generations - from photorealistic to Pixar with a single file
Guide generation with reference images and control videos for character consistency
Share real-time video with TouchDesigner, Unity, and other Windows applications
### Go Deeper
Ready to build programmatically? Scope exposes a powerful API for integration into your own applications.
Set up the server, connect via WebRTC, and control generation in real-time
Explore each pipeline's capabilities, parameters, and hardware requirements
### Supported Pipelines
Scope ships with five built-in autoregressive video diffusion pipelines: [StreamDiffusion V2](/scope/reference/pipelines/streamdiffusion-v2), [LongLive](/scope/reference/pipelines/longlive), [Krea Realtime](/scope/reference/pipelines/krea-realtime), [RewardForcing](/scope/reference/pipelines/reward-forcing), and [MemFlow](/scope/reference/pipelines/memflow). Four run on 24GB GPUs, while Krea Realtime needs 32GB+ for its larger 14B model. A sixth pipeline, [LTX 2.3](/scope/reference/pipelines/ltx-2-3), is installed as a plugin and adds synchronized audio-video generation on 24GB cards.
Compare all pipelines - features, VRAM requirements, and use cases at a glance
***
## Step 4: Connect, Share & Contribute
Browse creations, download timelines, and share your work
Have questions or want to connect with others? Join our friendly community
Report issues, suggest features, or contribute code
***
## Troubleshooting
Run `nvidia-smi` and verify CUDA version is ≥ 12.8. Update your NVIDIA drivers if needed.
* Ensure UV, Node.js, and npm are properly installed
* Try clearing the cache: `uv cache clean && uv run build`
Install the Python development package:
```bash theme={null}
# Debian/Ubuntu
sudo apt-get install python3-dev
```
* Check your internet connection
* Verify disk space in `~/.daydream-scope/models`
* Model downloads can be large - be patient on first run
* Verify the instance is fully deployed in RunPod dashboard
* Ensure you're accessing port 8000
* Check that `HF_TOKEN` is correctly set
* Verify your `HF_TOKEN` is valid with read permissions
* Try redeploying the instance with the correct token