lllyasviel/stable-diffusion-webui-forge
1
0
Fork 0
mirror of https://github.com/lllyasviel/stable-diffusion-webui-forge.git synced 2026-04-30 11:11:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #1570 from DenOfEquity/latentmodifier

Browse Source 
fix issue 258 (latent modifier built-in extension)
This commit is contained in:
DenOfEquity
2024-08-29 00:54:04 +01:00
committed by GitHub
parent d339600181 ea6d03bde1
commit 948e91458a
1 changed files with 6 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
extensions-builtin/sd_forge_latent_modifier/lib_latent_modifier/sampler_mega_modifier.py
View File

@@ -824,27 +824,20 @@ def pyramid_noise_like(x, discount=0.9, generator=None, rand_source=random):
import math import math
def dyn_cfg_modifier(conditioning, unconditioning, method, cond_scale, time_mult): def dyn_cfg_modifier(conditioning, unconditioning, method, cond_scale, time_mult):
time = time_mult[0].item()
time_factor = -(math.cos(0.5 * time * math.pi) / 2) + 1
noise_pred = conditioning - unconditioning
noise_pred_magnitude = (torch.linalg.vector_norm(noise_pred, dim=(1)) + 0.0000000001)[:,None]
match method: match method:
case "dyncfg-halfcosine": case "dyncfg-halfcosine":
noise_pred = conditioning - unconditioning
noise_pred_magnitude = (torch.linalg.vector_norm(noise_pred, dim=(1)) + 0.0000000001)[:,None]
time = time_mult.item()
time_factor = -(math.cos(0.5 * time * math.pi) / 2) + 1
noise_pred_timescaled_magnitude = (torch.linalg.vector_norm(noise_pred * time_factor, dim=(1)) + 0.0000000001)[:,None] noise_pred_timescaled_magnitude = (torch.linalg.vector_norm(noise_pred * time_factor, dim=(1)) + 0.0000000001)[:,None]
noise_pred /= noise_pred_magnitude noise_pred /= noise_pred_magnitude
noise_pred *= noise_pred_timescaled_magnitude noise_pred *= noise_pred_timescaled_magnitude
return noise_pred return noise_pred
case "dyncfg-halfcosine-mimic": case "dyncfg-halfcosine-mimic":
noise_pred = conditioning - unconditioning
noise_pred_magnitude = (torch.linalg.vector_norm(noise_pred, dim=(1)) + 0.0000000001)[:,None]
time = time_mult.item()
time_factor = -(math.cos(0.5 * time * math.pi) / 2) + 1
latent = noise_pred latent = noise_pred
mimic_latent = noise_pred * time_factor mimic_latent = noise_pred * time_factor