The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.
Make perp neg take a conditioning input instead of a CLIP one.
comfyanonymous9cad2f06ff93
diff --git a/comfy_extras/nodes_perpneg.py b/comfy_extras/nodes_perpneg.pyindex 36f2eb01a..0c5ccb77a 100644--- a/comfy_extras/nodes_perpneg.py+++ b/comfy_extras/nodes_perpneg.py@@ -9,7 +9,7 @@ class PerpNeg:@classmethoddef INPUT_TYPES(s):return {"required": {"model": ("MODEL", ),- "clip": ("CLIP", ),+ "empty_conditioning": ("CONDITIONING", ),"neg_scale": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 100.0}),}}RETURN_TYPES = ("MODEL",)@@ -17,13 +17,9 @@ class PerpNeg:CATEGORY = "_for_testing"- def patch(self, model, clip, neg_scale):+ def patch(self, model, empty_conditioning, neg_scale):m = model.clone()-- tokens = clip.tokenize("")- nocond, nocond_pooled = clip.encode_from_tokens(tokens, return_pooled=True)- nocond = [[nocond, {"pooled_output": nocond_pooled}]]- nocond = comfy.sample.convert_cond(nocond)+ nocond = comfy.sample.convert_cond(empty_conditioning)def cfg_function(args):model = args["model"]@@ -33,9 +29,9 @@ class PerpNeg:x = args["input"]sigma = args["sigma"]model_options = args["model_options"]-+(noise_pred_nocond, _) = comfy.samplers.calc_cond_uncond_batch(model, nocond, None, x, sigma, model_options)-+pos = noise_pred_pos - noise_pred_nocondneg = noise_pred_neg - noise_pred_nocondperp = ((torch.mul(pos, neg).sum())/(torch.norm(neg)**2)) * neg