From 94b573b0bd2bcd5a1d16cb788c86e940a04b61d9 Mon Sep 17 00:00:00 2001 From: Cherrytest Date: Mon, 6 Oct 2025 16:17:10 +0000 Subject: [PATCH] Create README.md --- Wan22Animate/README.md | 10 ++++++++++ 1 file changed, 10 insertions(+) create mode 100644 Wan22Animate/README.md diff --git a/Wan22Animate/README.md b/Wan22Animate/README.md new file mode 100644 index 0000000..09137ce --- /dev/null +++ b/Wan22Animate/README.md @@ -0,0 +1,10 @@ +# WanAnimate fp8_scaled models + +# v2 bugfix update for native workflows + +The initial uploads has the face encoder layers in bf16, as back then I was unsure if they can be quantized. + +When running these models in the WanVideoWrapper those layers are used in the base_precision, but when running these +in native ComfyUI, they get downscaled to normal fp8, which seems to introduce pretty bad noise in the outputs in a grid pattern. + +The v2 update addresses this by using fp8 scaled on the face encoder layers as well. For the wrapper you can keep using the initial models. \ No newline at end of file