ComfyUI beginner setup with LTX 2.3 and Flux Klein

This beginner tutorial covers ComfyUI installation and local AI image and video workflows. The presenter describes ComfyUI as free and open source, with node connections that control the generation pipeline. The setup uses the Windows portable build and launches through run_nvidia_gpu.bat into a browser on localhost.

The NVIDIA-sponsored guide recommends at least 8 GB of VRAM for many image models and some video models with optimization flags, with 12 to 16 GB as a more comfortable range. These are the speaker's recommendations rather than guarantees for every workflow.

Installation covers model folders, ComfyUI Manager, and custom nodes. The LTX 2.3 example introduces subgraphs, video previews through Video Helper Suite, and H.264 MP4 output at 24 fps. The presenter warns that complex dialogue can become gibberish and demonstrates bypassing prompt rewriting when preserving dialogue matters. A short explanation connects noise, conditioning, sampling, and VAE decoding.

The upscaling section installs NVIDIA RTX Video Super Resolution nodes, GPU-specific Video Effects software, and requirements before restarting ComfyUI. For images, the guide uses FP8 distilled Flux Klein models for editing and reference-based style transfer, then demonstrates Flux 1 inpainting. It also explains that closed-source API nodes require an API key and still consume credits.