LTX-2 inference with ComfyUI¶
A streamlined and automated environment for running ComfyUI with LTX-2 video models, optimized for use on RunPod
bf16¶
fp8¶
🔧 Features¶
- Automatic model and LoRA provisioning via environment variables or lora-manager.
- Supports advanced workflows for video generation and enhancement using pre-installed custom nodes.
- Compatible with high-performance NVIDIA GPUs (CUDA 12.8).
- Compiled attentions and GPU accelerations.
- Loads models/workflows dependent on available VRAM (see Hardware requirements).
- Latent preview enabled for both samplers.
- Based on Kijai's workflow.
- i2v and t2v.
🔧 Built-in authentication¶
- ComfyUI
- Code Server
- Hugging Face API
- CivitAI API