From my personal experience the quality difference with Comfy-Org FP8 isn't huge at first glance, but the subtle improvements in details grew on me over time, now I clearly notice them when using <a target="_blank" rel="ugc" href="https://huggingface.co/Kijai/WanVideo_comfy_fp8_scaled/discussions">Kijai quantizations</a>. It's hard to explain exactly, but those small enhancements make a real difference and makes you go back to it. For <a rel="ugc" href="https://civarchive.com/search/models?modelType=Checkpoint&amp;sortBy=models_v9&amp;query=WAN2.2">NSFW</a> videos, I used SmoothMix High noise with one LightX2V set at 0.23 strength to eliminate bounce issues. Use ComfyUI default workflow or WanVideoWrapper in the Zip file: <a target="_blank" rel="ugc" href="https://civarchive.com/api/download/models/2544663?type=Training%20Data">I2V</a> | <a target="_blank" rel="ugc" href="https://civarchive.com/api/download/models/2545187?type=Training%20Data">T2V</a><ul><li>In practice: can use a bit more VRAM at load time in some workflows, but once running it’s often competitive or slightly faster than the default Comfy‑Org workflow. Quality is extremely close to FP16, especially in subtle detail, texture and motion.</li><li>“better fp8 scaled models (when measured against fp16)” with a <a target="_blank" rel="ugc" href="https://github.com/Tencent-Hunyuan/HunyuanVideo/blob/main/hyvideo/modules/fp8_optimization.py">quantization method</a> inspired by Tencent’s HunyuanVideo FP8 code.</li><li>large linear weights in FP8, many small tensors (biases, norms, embeddings, head, modulation) kept in FP32 for extra stability.</li><li>Designed and tested together with the <a target="_blank" rel="ugc" href="https://github.com/kijai/ComfyUI-WanVideoWrapper">ComfyUI‑WanVideoWrapper</a>, but also works with native WanVideo nodes.</li></ul><h3 id="comfy-org-or-kj-which-one-should-you-use">Comfy-Org or KJ, Which one should you use?</h3><ul><li>Use Kijai FP8 if:<ul><li>You care about squeezing every bit of quality out of the FP8 and value subtle detail, texture and motion coherence across many renders.</li><li>You’re using ComfyUI‑WanVideoWrapper or don’t mind experimenting a bit with workflows.</li><li>You can tolerate slightly higher VRAM use during load in exchange for potential quality/speed wins in actual generation.</li></ul></li><li>Use Comfy‑Org FP8 if:<ul><li>You want the “official” default that matches ComfyUI Wan2.2 tutorials and examples.</li><li>You prefer a stable, well‑documented setup that “just works” with native WanVideo nodes.</li><li>You mainly care about solid quality + VRAM savings, without chasing the last 1-2% of detail.</li></ul></li></ul>Think of Comfy‑Org FP8 as a reliable baseline, and Kijai FP8 as an enthusiast‑tuned variant with extra precision in key places and a slight edge in subtle detail,<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/72cc2722-4568-4331-bd29-d6f6f74c9327/width=525/72cc2722-4568-4331-bd29-d6f6f74c9327.jpeg" />

WAN2.2 FP8 KJ

Wan2_2-T2V-A14B_HIGH_fp8_e4m3fn_scaled_KJ.

wan22FP8KJ_t2vA14BHIGH_trainingData.zip

wan22FP8KJ_t2vA14BHIGH.safetensors

Files with SHA256: d4db18190303e0ba27d6c6318ffe8ee943a2cda32608028e9842db2857f7ee8e