Update CFG

Author: Klace

modules/img2img.py

@@ -76,7 +76,7 @@ def process_batch(p, input_dir, output_dir, inpaint_mask_dir, args):
                 processed_image.save(os.path.join(output_dir, filename))
 
 
-def img2img(id_task: str, mode: int, prompt: str, negative_prompt: str, prompt_styles, init_img, sketch, init_img_with_mask, inpaint_color_sketch, inpaint_color_sketch_orig, init_img_inpaint, init_mask_inpaint, steps: int, sampler_index: int, mask_blur: int, mask_alpha: float, inpainting_fill: int, restore_faces: bool, tiling: bool, n_iter: int, batch_size: int, cfg_scale: float, text_cfg_scale: float, denoising_strength: float, seed: int, subseed: int, subseed_strength: float, seed_resize_from_h: int, seed_resize_from_w: int, seed_enable_extras: bool, height: int, width: int, resize_mode: int, inpaint_full_res: bool, inpaint_full_res_padding: int, inpainting_mask_invert: int, img2img_batch_input_dir: str, img2img_batch_output_dir: str, img2img_batch_inpaint_mask_dir: str, override_settings_texts, *args):
+def img2img(id_task: str, mode: int, prompt: str, negative_prompt: str, prompt_styles, init_img, sketch, init_img_with_mask, inpaint_color_sketch, inpaint_color_sketch_orig, init_img_inpaint, init_mask_inpaint, steps: int, sampler_index: int, mask_blur: int, mask_alpha: float, inpainting_fill: int, restore_faces: bool, tiling: bool, n_iter: int, batch_size: int, cfg_scale: float, image_cfg_scale: float, denoising_strength: float, seed: int, subseed: int, subseed_strength: float, seed_resize_from_h: int, seed_resize_from_w: int, seed_enable_extras: bool, height: int, width: int, resize_mode: int, inpaint_full_res: bool, inpaint_full_res_padding: int, inpainting_mask_invert: int, img2img_batch_input_dir: str, img2img_batch_output_dir: str, img2img_batch_inpaint_mask_dir: str, override_settings_texts, *args):
     override_settings = create_override_settings_dict(override_settings_texts)
 
     is_batch = mode == 5
@@ -132,7 +132,7 @@ def img2img(id_task: str, mode: int, prompt: str, negative_prompt: str, prompt_s
         n_iter=n_iter,
         steps=steps,
         cfg_scale=cfg_scale,
-        text_cfg_scale=text_cfg_scale,
+        image_cfg_scale=image_cfg_scale,
         width=width,
         height=height,
         restore_faces=restore_faces,

modules/processing.py

@@ -912,7 +912,7 @@ def save_intermediate(image, index):
 class StableDiffusionProcessingImg2Img(StableDiffusionProcessing):
     sampler = None
 
-    def __init__(self, init_images: list = None, resize_mode: int = 0, denoising_strength: float = 0.75, mask: Any = None, mask_blur: int = 4, inpainting_fill: int = 0, inpaint_full_res: bool = True, inpaint_full_res_padding: int = 0, inpainting_mask_invert: int = 0, text_cfg_scale: float = 7.5, initial_noise_multiplier: float = None, **kwargs):
+    def __init__(self, init_images: list = None, resize_mode: int = 0, denoising_strength: float = 0.75, mask: Any = None, mask_blur: int = 4, inpainting_fill: int = 0, inpaint_full_res: bool = True, inpaint_full_res_padding: int = 0, inpainting_mask_invert: int = 0, image_cfg_scale: float = 7.5, initial_noise_multiplier: float = None, **kwargs):
         super().__init__(**kwargs)
 
         self.init_images = init_images
@@ -927,7 +927,7 @@ def __init__(self, init_images: list = None, resize_mode: int = 0, denoising_str
         self.inpaint_full_res = inpaint_full_res
         self.inpaint_full_res_padding = inpaint_full_res_padding
         self.inpainting_mask_invert = inpainting_mask_invert
-        self.text_cfg_scale=text_cfg_scale
+        self.image_cfg_scale=image_cfg_scale
         self.initial_noise_multiplier = opts.initial_noise_multiplier if initial_noise_multiplier is None else initial_noise_multiplier
         self.mask = None
         self.nmask = None

modules/sd_samplers_common.py

@@ -39,6 +39,7 @@ def single_sample_to_image(sample, approximation=None):
         x_sample = processing.decode_first_stage(shared.sd_model, sample.unsqueeze(0))[0]
 
     x_sample = torch.clamp((x_sample + 1.0) / 2.0, min=0.0, max=1.0)
+    x_sample = 255.0 * rearrange(x_sample, "1 c h w -> h w c")
     x_sample = 255. * np.moveaxis(x_sample.cpu().numpy(), 0, 2)
     x_sample = x_sample.astype(np.uint8)
 

modules/sd_samplers_kdiffusion.py

@@ -58,18 +58,17 @@ def __init__(self, model):
         self.init_latent = None
         self.step = 0
 
-    def combine_denoised(self, x_out, conds_list, uncond, cond_scale, text_scale):
+    def combine_denoised(self, x_out, conds_list, uncond, cond_scale, image_scale):
         denoised_uncond = x_out[-uncond.shape[0]:]
         denoised = torch.clone(denoised_uncond)
 
         for i, conds in enumerate(conds_list):
             for cond_index, weight in conds:
-                #denoised[i] += (x_out[cond_index] - denoised_uncond[i]) * (weight * cond_scale)
-                denoised[i] += text_scale * (x_out[cond_index] - denoised_uncond[i]) + cond_scale * (denoised_uncond[i] - x_out[cond_index])
+                denoised[i] += cond_scale * (x_out[cond_index] - denoised_uncond[i]) + image_scale * (denoised_uncond[i] - x_out[cond_index])
 
         return denoised
 
-    def forward(self, x, sigma, uncond, cond, cond_scale, image_cond, text_scale):
+    def forward(self, x, sigma, uncond, cond, cond_scale, image_cond, image_scale):
         if state.interrupted or state.skipped:
             raise sd_samplers_common.InterruptedException
 
@@ -116,7 +115,7 @@ def forward(self, x, sigma, uncond, cond, cond_scale, image_cond, text_scale):
         elif opts.live_preview_content == "Negative prompt":
             sd_samplers_common.store_latent(x_out[-uncond.shape[0]:])
 
-        denoised = self.combine_denoised(x_out, conds_list, uncond, cond_scale, text_scale)
+        denoised = self.combine_denoised(x_out, conds_list, uncond, cond_scale, image_scale)
 
         if self.mask is not None:
             denoised = self.init_latent * self.mask + self.nmask * denoised
@@ -293,7 +292,7 @@ def sample_img2img(self, p, x, noise, conditioning, unconditional_conditioning,
             'image_cond': image_conditioning, 
             'uncond': unconditional_conditioning, 
             'cond_scale': p.cfg_scale,
-            'text_scale': p.text_cfg_scale
+            'image_scale': p.image_cfg_scale
         }, disable=False, callback=self.callback_state, **extra_params_kwargs))
 
         return samples

modules/ui.py

@@ -765,8 +765,8 @@ def copy_image(img):
 
                     elif category == "cfg":
                         with FormGroup():
-                            cfg_scale = gr.Slider(minimum=1.0, maximum=30.0, step=0.5, label='CFG Scale', value=7.0, elem_id="img2img_cfg_scale")
-                            text_cfg_scale = gr.Slider(minimum=1.0, maximum=30.0, step=0.5, label='Text CFG Scale', value=7.5, elem_id="img2img_cfg_scale")
+                            cfg_scale = gr.Slider(minimum=1.0, maximum=30.0, step=0.5, label='CFG Scale', value=7.5, elem_id="img2img_cfg_scale")
+                            image_cfg_scale = gr.Slider(minimum=1.0, maximum=30.0, step=0.5, label='Image CFG Scale', value=1.5, elem_id="img2img_cfg_scale")
                             denoising_strength = gr.Slider(minimum=0.0, maximum=1.0, step=0.01, label='Denoising strength', value=0.75, elem_id="img2img_denoising_strength")
 
                     elif category == "seed":
@@ -862,7 +862,7 @@ def select_img2img_tab(tab):
                     batch_count,
                     batch_size,
                     cfg_scale,
-                    text_cfg_scale,
+                    image_cfg_scale,
                     denoising_strength,
                     seed,
                     subseed, subseed_strength, seed_resize_from_h, seed_resize_from_w, seed_checkbox,