Quantized versions of the FastH3 V2 model have been released for ComfyUI

Developers presented FastH3-V2_GGUFs — a set of quantized weights for the FastVideo FastH3 generative model, optimized for operation in the ComfyUI environment. The release includes three compression options: Q3_K_M, Q4_K_M, and Q5_K_M, which allows running the model on consumer equipment with limited video memory. The FastH3 8-Step V2 model is focused on generating video content and requires the use of auxiliary components for full-fledged operation. In particular, for the correct functioning of the system, integration with Qwen3-VL is necessary, as well as the connection of the corresponding VAE models. The use of the GGUF format provides compatibility with a wide range of local rendering tools. Quantization allows significantly reducing the requirements for hardware resources while maintaining the basic characteristics of generation. The Q4_K_M version is a balanced option that provides an optimal ratio between processing speed and the quality of the final video stream. The model showed a result of 8 points on the internal evaluation system with a generation time of 2 hours on the test configuration. This release simplifies the implementation of heavy video models in ComfyUI users' local workflows. The availability of various levels of quantization gives the opportunity to flexibly adjust the system's performance depending on the available amount of VRAM, which is critical for working with real-time video generation.