The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.
float16 inference is a bit broken on mochi.
comfyanonymous9ee0a6553ab9
diff --git a/comfy/supported_models.py b/comfy/supported_models.pyindex 57099082a..8300c4c34 100644--- a/comfy/supported_models.py+++ b/comfy/supported_models.py@@ -686,7 +686,7 @@ class GenmoMochi(supported_models_base.BASE):memory_usage_factor = 2.0 #TODO- supported_inference_dtypes = [torch.float16, torch.bfloat16, torch.float32]+ supported_inference_dtypes = [torch.bfloat16, torch.float32]vae_key_prefix = ["vae."]text_encoder_key_prefix = ["text_encoders."]