ComfyUI Extension: LTX-2 Microscope

Authored by dorpxam

Created

Updated

5 stars

Run ComfyUI workflows without the setup

No installs, no CUDA version roulette, no GPU sitting idle on your bill. Bring a workflow and run it in the browser.

48-Layers Latent Previewer for Lightrick's LTX-2 model.

Looking for a different extension?

Custom Nodes (0)

    README

    LTX-2 Microscope

    48-Layers Latent Previewer for Lightricks LTX-2 Video Transformer.

    [!CAUTION] For now, this project is not compatible to the Node 2.0 (UI) of ComfyUI.

    Screenshot of LTX-2 Microscope.

    Description

    LTX-2 Microscope is a technical diagnostic tool for ComfyUI designed to visualize the internal latent states across all 48 layers of the LTX-2 transformer model during the sampling process. It provides a real-time 8x6 grid overlay to analyze how noise transforms into signal through each layer of the architecture.

    Features

    • View Modes: Supports conditioning, unconditioning, and differential visualization.
    • Normalization: Dynamic range remapping via none, min_max, z_score, or percentile modes.
    • Colormaps: Integrated support for 11 perceptual colormaps (Cividis, Inferno, Viridis, Magma, etc.).
    • Animation Control:
      • Realtime: Automated looping playback with FPS control and freeze toggle.
      • Frame-by-Frame: Manual navigation slider for precise temporal latent analysis.

    Installation

    No external dependencies required. This extension uses native ComfyUI and Python libraries.

    1. Navigate to ComfyUI/custom_nodes.
    2. Clone the repository:
    git clone https://github.com/dorpxam/ComfyUI-LTX2-Microscope
    1. Restart ComfyUI.

    Usage

    The node can be inserted anywhere in the workflow where a MODEL link is available.

    1. Connect the model input/output.
    2. The UI will automatically synchronize with the LTX-2 sampling process.
    3. Use the RESET button to clear the local buffer and return to the idle state.

    License

    Apache 2.0

    Run ComfyUI workflows without the setup

    No installs, no CUDA version roulette, no GPU sitting idle on your bill. Bring a workflow and run it in the browser.

    Learn more