| |
|
|
| import gradio as gr |
| import torch |
| import torch.nn.functional as F |
|
|
| from modules import scripts |
| from modules.infotext_utils import PasteField |
| from modules.shared import opts |
|
|
|
|
| class ScriptMahiro(scripts.ScriptBuiltinUI): |
| section = "cfg" |
| create_group = False |
| sorting_priority = 1 |
|
|
| def title(self): |
| return "MaHiRo" |
|
|
| def show(self, is_img2img): |
| return scripts.AlwaysVisible if opts.show_mahiro else None |
|
|
| def ui(self, is_img2img): |
| enable = gr.Checkbox( |
| value=False, |
| label="MaHiRo", |
| elem_id=f"{'img2img' if is_img2img else 'txt2img'}_enable_mahiro", |
| scale=1, |
| ) |
| self.infotext_fields = [PasteField(enable, "MaHiRo", api="mahiro")] |
| return [enable] |
|
|
| def after_extra_networks_activate(self, p, enable, *args, **kwargs): |
| if enable: |
| p.extra_generation_params.update({"MaHiRo": enable}) |
| p.mahiro = True |
|
|
| def process_before_every_sampling(self, p, *args, **kwargs): |
| if str(getattr(p, "mahiro", False)) != "True": |
| return |
|
|
| @torch.inference_mode() |
| def mahiro_normd(args: dict): |
| scale: float = args["cond_scale"] |
| cond_p: torch.Tensor = args["cond_denoised"] |
| uncond_p: torch.Tensor = args["uncond_denoised"] |
| leap = cond_p * scale |
| u_leap = uncond_p * scale |
| cfg: torch.Tensor = args["denoised"] |
| merge = (leap + cfg) / 2 |
| normu = torch.sqrt(u_leap.abs()) * u_leap.sign() |
| normm = torch.sqrt(merge.abs()) * merge.sign() |
| sim = F.cosine_similarity(normu, normm).mean() |
| simsc = 2 * (sim + 1) |
| wm = (simsc * cfg + (4 - simsc) * leap) / 4 |
| return wm |
|
|
| unet = p.sd_model.forge_objects.unet.clone() |
| unet.set_model_sampler_post_cfg_function(mahiro_normd) |
| p.sd_model.forge_objects.unet = unet |
|
|
| print("using MaHiRo") |
|
|
