WanVideo Comfy Fp8 Scaled is an FP8 scaled and quantized version of the WanVideo 14B text-to-video model hosted on Hugging Face. It is designed for integration with ComfyUI workflows and provides improved performance over FP16 versions in certain scenarios.
The model originates from quantization code published by Tencent Hunyuan. It supports 14B-T2V comparison tests without additional LoRAs at settings such as 25 steps and 832x480x81 resolution. Tests indicate better results than the FP16 counterpart when measured in those conditions.
It is delivered as a single-file diffusion model under the Apache 2.0 license. The repository explicitly states it can be used with the ComfyUI-WanVideoWrapper in its latest version and with ComfyUI native WanVideo nodes. No code snippets for direct library usage are provided on the page, directing users to check repository files and library documentation instead.
The model is intended for local video generation tasks that rely on the named ComfyUI components. It is published by the Hugging Face user Kijai.
WanVideo Comfy Fp8 Scaled is an Other AI project. It focuses on generating high-quality videos locally using quantized models within ComfyUI. WanVideo Comfy Fp8 Scaled is an open-source project aimed at developers. WanVideo Comfy Fp8 Scaled is open source under the Open Source license. WanVideo Comfy Fp8 Scaled is available on the web, and it can be self-hosted.
Kijai builds and maintains WanVideo Comfy Fp8 Scaled, and it first shipped in 2024. The project is developed in the open on GitHub with 12.3k stars and 2 commits in the last 90 days. Key capabilities include Video Generation, FP8 Quantization, and ComfyUI Integration.
Summary written by a language model from the project’s public pages.
What PulseGate has recorded for this listing
Closest matches by what these projects do