ltx workflow

LTX Workflow is a platform that provides direct downloads for the LTX 2.3 AI video generation model files and tools to instantly generate custom ComfyUI workflow JSON configurations for text-to-video and image-to-video projects.

The VRAM Adapter is a tool where you select your GPU's video memory (e.g., 16GB). It then displays and recommends the most suitable LTX 2.3 model variants, like FP8-quantized versions, optimized to run within your system's constraints.

This is the required VAE (Variational Autoencoder) file for the LTX 2.3 model. It is essential for all setups and must be downloaded first, regardless of which main checkpoint model you choose, as it decodes the generated latents into video.

You select parameters like model, resolution, frame count, and input a prompt. The tool then generates a complete JSON file representing a ComfyUI node graph. You drag this file into your ComfyUI interface to load the pre-configured workflow instantly.

FP8 (8-bit floating point) models are quantized versions of the full LTX 2.3 checkpoint. They significantly reduce VRAM requirements (down to ~16GB), making the model accessible to more users, though they may require newer GPUs (RTX 40xx+) for optimal speed.

No, the website is not affiliated with Lightricks, the creators of LTX 2.3. It is an independent resource hub that curates and provides tools for the community using the official, open-source models and software.

Distilled models are smaller, optimized versions trained from the larger developer ('dev') model. They offer good quality with lower resource needs. The 'dev' model is the full, base model, often used when applying custom LoRA weights for fine-tuning.

Basic functions like browsing models and viewing resources are free. However, features like downloading generated workflow JSON files and accessing the advanced 'Director Mode' prompt generator require signing in to the platform.

The spatial upscaler is an optional model component used in a two-stage pipeline. After the main video is generated at a base resolution, this upscaler can be applied to increase the output's spatial resolution (e.g., 2x) for higher detail.

The website has a dedicated 'Tutorials' section under Resources, and a 'Guide' page. These provide step-by-step instructions on installing models, setting up ComfyUI nodes, and running workflows to help users get started.