wangbo/ComfyUI
1
0
Fork 0
mirror of https://github.com/comfyanonymous/ComfyUI.git synced 2025-12-18 18:43:05 +08:00
Code Issues Actions 10 Packages Projects Releases Wiki Activity

Merge branch 'master' into dr-support-pip-cm

Browse Source
This commit is contained in:
Dr.Lt.Data 2025-07-20 16:31:14 +09:00
commit b603e034e5
4 changed files with 80 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
.ci/windows_base_files/README_VERY_IMPORTANT.txt
View File

@ -4,6 +4,9 @@ if you have a NVIDIA gpu:
run_nvidia_gpu.bat
if you want to enable the fast fp16 accumulation (faster for fp16 models with slightly less quality):
run_nvidia_gpu_fast_fp16_accumulation.bat
To run it in slow CPU mode:

29
comfy_extras/nodes_cfg.py
View File

@ -40,6 +40,33 @@ class CFGZeroStar:
m.set_model_sampler_post_cfg_function(cfg_zero_star)
return (m, )
class CFGNorm:
@classmethod
def INPUT_TYPES(s):
return {"required": {"model": ("MODEL",),
"strength": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 100.0, "step": 0.01}),
}}
RETURN_TYPES = ("MODEL",)
RETURN_NAMES = ("patched_model",)
FUNCTION = "patch"
CATEGORY = "advanced/guidance"
EXPERIMENTAL = True
def patch(self, model, strength):
m = model.clone()
def cfg_norm(args):
cond_p = args['cond_denoised']
pred_text_ = args["denoised"]
norm_full_cond = torch.norm(cond_p, dim=1, keepdim=True)
norm_pred_text = torch.norm(pred_text_, dim=1, keepdim=True)
scale = (norm_full_cond / (norm_pred_text + 1e-8)).clamp(min=0.0, max=1.0)
return pred_text_ * scale * strength
m.set_model_sampler_post_cfg_function(cfg_norm)
return (m, )
NODE_CLASS_MAPPINGS = {
"CFGZeroStar": CFGZeroStar
"CFGZeroStar": CFGZeroStar,
"CFGNorm": CFGNorm,
}

58
comfy_extras/nodes_custom_sampler.py
View File

@ -301,6 +301,35 @@ class ExtendIntermediateSigmas:
return (extended_sigmas,)
class SamplingPercentToSigma:
@classmethod
def INPUT_TYPES(cls) -> InputTypeDict:
return {
"required": {
"model": (IO.MODEL, {}),
"sampling_percent": (IO.FLOAT, {"default": 0.0, "min": 0.0, "max": 1.0, "step": 0.0001}),
"return_actual_sigma": (IO.BOOLEAN, {"default": False, "tooltip": "Return the actual sigma value instead of the value used for interval checks.\nThis only affects results at 0.0 and 1.0."}),
}
}
RETURN_TYPES = (IO.FLOAT,)
RETURN_NAMES = ("sigma_value",)
CATEGORY = "sampling/custom_sampling/sigmas"
FUNCTION = "get_sigma"
def get_sigma(self, model, sampling_percent, return_actual_sigma):
model_sampling = model.get_model_object("model_sampling")
sigma_val = model_sampling.percent_to_sigma(sampling_percent)
if return_actual_sigma:
if sampling_percent == 0.0:
sigma_val = model_sampling.sigma_max.item()
elif sampling_percent == 1.0:
sigma_val = model_sampling.sigma_min.item()
return (sigma_val,)
class KSamplerSelect:
@classmethod
def INPUT_TYPES(s):
@ -683,9 +712,10 @@ class CFGGuider:
return (guider,)
class Guider_DualCFG(comfy.samplers.CFGGuider):
def set_cfg(self, cfg1, cfg2):
def set_cfg(self, cfg1, cfg2, nested=False):
self.cfg1 = cfg1
self.cfg2 = cfg2
self.nested = nested
def set_conds(self, positive, middle, negative):
middle = node_helpers.conditioning_set_values(middle, {"prompt_type": "negative"})
@ -695,14 +725,20 @@ class Guider_DualCFG(comfy.samplers.CFGGuider):
negative_cond = self.conds.get("negative", None)
middle_cond = self.conds.get("middle", None)
positive_cond = self.conds.get("positive", None)
if model_options.get("disable_cfg1_optimization", False) == False:
if math.isclose(self.cfg2, 1.0):
negative_cond = None
if math.isclose(self.cfg1, 1.0):
middle_cond = None
out = comfy.samplers.calc_cond_batch(self.inner_model, [negative_cond, middle_cond, positive_cond], x, timestep, model_options)
return comfy.samplers.cfg_function(self.inner_model, out[1], out[0], self.cfg2, x, timestep, model_options=model_options, cond=middle_cond, uncond=negative_cond) + (out[2] - out[1]) * self.cfg1
if self.nested:
out = comfy.samplers.calc_cond_batch(self.inner_model, [negative_cond, middle_cond, positive_cond], x, timestep, model_options)
pred_text = comfy.samplers.cfg_function(self.inner_model, out[2], out[1], self.cfg1, x, timestep, model_options=model_options, cond=positive_cond, uncond=middle_cond)
return out[0] + self.cfg2 * (pred_text - out[0])
else:
if model_options.get("disable_cfg1_optimization", False) == False:
if math.isclose(self.cfg2, 1.0):
negative_cond = None
if math.isclose(self.cfg1, 1.0):
middle_cond = None
out = comfy.samplers.calc_cond_batch(self.inner_model, [negative_cond, middle_cond, positive_cond], x, timestep, model_options)
return comfy.samplers.cfg_function(self.inner_model, out[1], out[0], self.cfg2, x, timestep, model_options=model_options, cond=middle_cond, uncond=negative_cond) + (out[2] - out[1]) * self.cfg1
class DualCFGGuider:
@classmethod
@ -714,6 +750,7 @@ class DualCFGGuider:
"negative": ("CONDITIONING", ),
"cfg_conds": ("FLOAT", {"default": 8.0, "min": 0.0, "max": 100.0, "step":0.1, "round": 0.01}),
"cfg_cond2_negative": ("FLOAT", {"default": 8.0, "min": 0.0, "max": 100.0, "step":0.1, "round": 0.01}),
"style": (["regular", "nested"],),
}
}
@ -722,10 +759,10 @@ class DualCFGGuider:
FUNCTION = "get_guider"
CATEGORY = "sampling/custom_sampling/guiders"
def get_guider(self, model, cond1, cond2, negative, cfg_conds, cfg_cond2_negative):
def get_guider(self, model, cond1, cond2, negative, cfg_conds, cfg_cond2_negative, style):
guider = Guider_DualCFG(model)
guider.set_conds(cond1, cond2, negative)
guider.set_cfg(cfg_conds, cfg_cond2_negative)
guider.set_cfg(cfg_conds, cfg_cond2_negative, nested=(style == "nested"))
return (guider,)
class DisableNoise:
@ -879,6 +916,7 @@ NODE_CLASS_MAPPINGS = {
"FlipSigmas": FlipSigmas,
"SetFirstSigma": SetFirstSigma,
"ExtendIntermediateSigmas": ExtendIntermediateSigmas,
"SamplingPercentToSigma": SamplingPercentToSigma,
"CFGGuider": CFGGuider,
"DualCFGGuider": DualCFGGuider,

2
requirements.txt
View File

@ -1,5 +1,5 @@
comfyui-frontend-package==1.23.4
comfyui-workflow-templates==0.1.36
comfyui-workflow-templates==0.1.37
comfyui-embedded-docs==0.2.4
comfyui_manager
torch