diff options
| -rw-r--r-- | javascript/hints.js | 1 | ||||
| -rw-r--r-- | modules/processing.py | 127 | ||||
| -rw-r--r-- | modules/shared.py | 1 | ||||
| -rw-r--r-- | scripts/xy_grid.py | 2 |
4 files changed, 79 insertions, 52 deletions
diff --git a/javascript/hints.js b/javascript/hints.js index 6ddd6aec..623bc25c 100644 --- a/javascript/hints.js +++ b/javascript/hints.js @@ -75,6 +75,7 @@ titles = { "Create style": "Save current prompts as a style. If you add the token {prompt} to the text, the style use that as placeholder for your prompt when you use the style in the future.", "Checkpoint name": "Loads weights from checkpoint before making images. You can either use hash or a part of filename (as seen in settings) for checkpoint name. Recommended to use with Y axis for less switching.", + "Inpainting conditioning mask strength": "Only applies to inpainting models. Determines how strongly to mask off the original image for inpainting and img2img. 1.0 means fully masked, which is the default behaviour. 0.0 means a fully unmasked conditioning. Lower values will help preserve the overall composition of the image, but will struggle with large changes.", "vram": "Torch active: Peak amount of VRAM used by Torch during generation, excluding cached data.\nTorch reserved: Peak amount of VRAM allocated by Torch, including all active and cached data.\nSys VRAM: Peak amount of VRAM allocation across all applications / total GPU VRAM (peak utilization%).", diff --git a/modules/processing.py b/modules/processing.py index 4efba946..548eec29 100644 --- a/modules/processing.py +++ b/modules/processing.py @@ -129,6 +129,73 @@ class StableDiffusionProcessing(): self.all_seeds = None
self.all_subseeds = None
+ def txt2img_image_conditioning(self, x, width=None, height=None):
+ if self.sampler.conditioning_key not in {'hybrid', 'concat'}:
+ # Dummy zero conditioning if we're not using inpainting model.
+ # Still takes up a bit of memory, but no encoder call.
+ # Pretty sure we can just make this a 1x1 image since its not going to be used besides its batch size.
+ return torch.zeros(
+ x.shape[0], 5, 1, 1,
+ dtype=x.dtype,
+ device=x.device
+ )
+
+ height = height or self.height
+ width = width or self.width
+
+ # The "masked-image" in this case will just be all zeros since the entire image is masked.
+ image_conditioning = torch.zeros(x.shape[0], 3, height, width, device=x.device)
+ image_conditioning = self.sd_model.get_first_stage_encoding(self.sd_model.encode_first_stage(image_conditioning))
+
+ # Add the fake full 1s mask to the first dimension.
+ image_conditioning = torch.nn.functional.pad(image_conditioning, (0, 0, 0, 0, 1, 0), value=1.0)
+ image_conditioning = image_conditioning.to(x.dtype)
+
+ return image_conditioning
+
+ def img2img_image_conditioning(self, source_image, latent_image, image_mask = None):
+ if self.sampler.conditioning_key not in {'hybrid', 'concat'}:
+ # Dummy zero conditioning if we're not using inpainting model.
+ return torch.zeros(
+ latent_image.shape[0], 5, 1, 1,
+ dtype=latent_image.dtype,
+ device=latent_image.device
+ )
+
+ # Handle the different mask inputs
+ if image_mask is not None:
+ if torch.is_tensor(image_mask):
+ conditioning_mask = image_mask
+ else:
+ conditioning_mask = np.array(image_mask.convert("L"))
+ conditioning_mask = conditioning_mask.astype(np.float32) / 255.0
+ conditioning_mask = torch.from_numpy(conditioning_mask[None, None])
+
+ # Inpainting model uses a discretized mask as input, so we round to either 1.0 or 0.0
+ conditioning_mask = torch.round(conditioning_mask)
+ else:
+ conditioning_mask = torch.ones(1, 1, *source_image.shape[-2:])
+
+ # Create another latent image, this time with a masked version of the original input.
+ # Smoothly interpolate between the masked and unmasked latent conditioning image using a parameter.
+ conditioning_mask = conditioning_mask.to(source_image.device)
+ conditioning_image = torch.lerp(
+ source_image,
+ source_image * (1.0 - conditioning_mask),
+ getattr(self, "inpainting_mask_weight", shared.opts.inpainting_mask_weight)
+ )
+
+ # Encode the new masked image using first stage of network.
+ conditioning_image = self.sd_model.get_first_stage_encoding(self.sd_model.encode_first_stage(conditioning_image))
+
+ # Create the concatenated conditioning tensor to be fed to `c_concat`
+ conditioning_mask = torch.nn.functional.interpolate(conditioning_mask, size=latent_image.shape[-2:])
+ conditioning_mask = conditioning_mask.expand(conditioning_image.shape[0], -1, -1, -1)
+ image_conditioning = torch.cat([conditioning_mask, conditioning_image], dim=1)
+ image_conditioning = image_conditioning.to(shared.device).type(self.sd_model.dtype)
+
+ return image_conditioning
+
def init(self, all_prompts, all_seeds, all_subseeds):
pass
@@ -571,37 +638,16 @@ class StableDiffusionProcessingTxt2Img(StableDiffusionProcessing): self.truncate_x = int(self.firstphase_width - firstphase_width_truncated) // opt_f
self.truncate_y = int(self.firstphase_height - firstphase_height_truncated) // opt_f
- def create_dummy_mask(self, x, width=None, height=None):
- if self.sampler.conditioning_key in {'hybrid', 'concat'}:
- height = height or self.height
- width = width or self.width
-
- # The "masked-image" in this case will just be all zeros since the entire image is masked.
- image_conditioning = torch.zeros(x.shape[0], 3, height, width, device=x.device)
- image_conditioning = self.sd_model.get_first_stage_encoding(self.sd_model.encode_first_stage(image_conditioning))
-
- # Add the fake full 1s mask to the first dimension.
- image_conditioning = torch.nn.functional.pad(image_conditioning, (0, 0, 0, 0, 1, 0), value=1.0)
- image_conditioning = image_conditioning.to(x.dtype)
-
- else:
- # Dummy zero conditioning if we're not using inpainting model.
- # Still takes up a bit of memory, but no encoder call.
- # Pretty sure we can just make this a 1x1 image since its not going to be used besides its batch size.
- image_conditioning = torch.zeros(x.shape[0], 5, 1, 1, dtype=x.dtype, device=x.device)
-
- return image_conditioning
-
def sample(self, conditioning, unconditional_conditioning, seeds, subseeds, subseed_strength):
self.sampler = sd_samplers.create_sampler_with_index(sd_samplers.samplers, self.sampler_index, self.sd_model)
if not self.enable_hr:
x = create_random_tensors([opt_C, self.height // opt_f, self.width // opt_f], seeds=seeds, subseeds=subseeds, subseed_strength=self.subseed_strength, seed_resize_from_h=self.seed_resize_from_h, seed_resize_from_w=self.seed_resize_from_w, p=self)
- samples = self.sampler.sample(self, x, conditioning, unconditional_conditioning, image_conditioning=self.create_dummy_mask(x))
+ samples = self.sampler.sample(self, x, conditioning, unconditional_conditioning, image_conditioning=self.txt2img_image_conditioning(x))
return samples
x = create_random_tensors([opt_C, self.firstphase_height // opt_f, self.firstphase_width // opt_f], seeds=seeds, subseeds=subseeds, subseed_strength=self.subseed_strength, seed_resize_from_h=self.seed_resize_from_h, seed_resize_from_w=self.seed_resize_from_w, p=self)
- samples = self.sampler.sample(self, x, conditioning, unconditional_conditioning, image_conditioning=self.create_dummy_mask(x, self.firstphase_width, self.firstphase_height))
+ samples = self.sampler.sample(self, x, conditioning, unconditional_conditioning, image_conditioning=self.txt2img_image_conditioning(x, self.firstphase_width, self.firstphase_height))
samples = samples[:, :, self.truncate_y//2:samples.shape[2]-self.truncate_y//2, self.truncate_x//2:samples.shape[3]-self.truncate_x//2]
@@ -638,7 +684,12 @@ class StableDiffusionProcessingTxt2Img(StableDiffusionProcessing): x = None
devices.torch_gc()
- samples = self.sampler.sample_img2img(self, samples, noise, conditioning, unconditional_conditioning, steps=self.steps, image_conditioning=self.create_dummy_mask(samples))
+ image_conditioning = self.img2img_image_conditioning(
+ decoded_samples,
+ samples,
+ decoded_samples.new_ones(decoded_samples.shape[0], 1, decoded_samples.shape[2], decoded_samples.shape[3])
+ )
+ samples = self.sampler.sample_img2img(self, samples, noise, conditioning, unconditional_conditioning, steps=self.steps, image_conditioning=image_conditioning)
return samples
@@ -770,33 +821,7 @@ class StableDiffusionProcessingImg2Img(StableDiffusionProcessing): elif self.inpainting_fill == 3:
self.init_latent = self.init_latent * self.mask
- if self.sampler.conditioning_key in {'hybrid', 'concat'}:
- if self.image_mask is not None:
- conditioning_mask = np.array(self.image_mask.convert("L"))
- conditioning_mask = conditioning_mask.astype(np.float32) / 255.0
- conditioning_mask = torch.from_numpy(conditioning_mask[None, None])
-
- # Inpainting model uses a discretized mask as input, so we round to either 1.0 or 0.0
- conditioning_mask = torch.round(conditioning_mask)
- else:
- conditioning_mask = torch.ones(1, 1, *image.shape[-2:])
-
- # Create another latent image, this time with a masked version of the original input.
- conditioning_mask = conditioning_mask.to(image.device)
- conditioning_image = image * (1.0 - conditioning_mask)
- conditioning_image = self.sd_model.get_first_stage_encoding(self.sd_model.encode_first_stage(conditioning_image))
-
- # Create the concatenated conditioning tensor to be fed to `c_concat`
- conditioning_mask = torch.nn.functional.interpolate(conditioning_mask, size=self.init_latent.shape[-2:])
- conditioning_mask = conditioning_mask.expand(conditioning_image.shape[0], -1, -1, -1)
- self.image_conditioning = torch.cat([conditioning_mask, conditioning_image], dim=1)
- self.image_conditioning = self.image_conditioning.to(shared.device).type(self.sd_model.dtype)
- else:
- self.image_conditioning = torch.zeros(
- self.init_latent.shape[0], 5, 1, 1,
- dtype=self.init_latent.dtype,
- device=self.init_latent.device
- )
+ self.image_conditioning = self.img2img_image_conditioning(image, self.init_latent, self.image_mask)
def sample(self, conditioning, unconditional_conditioning, seeds, subseeds, subseed_strength):
diff --git a/modules/shared.py b/modules/shared.py index 1a9b8289..7c428d90 100644 --- a/modules/shared.py +++ b/modules/shared.py @@ -267,6 +267,7 @@ options_templates.update(options_section(('sd', "Stable Diffusion"), { "sd_checkpoint_cache": OptionInfo(0, "Checkpoints to cache in RAM", gr.Slider, {"minimum": 0, "maximum": 10, "step": 1}),
"sd_hypernetwork": OptionInfo("None", "Hypernetwork", gr.Dropdown, lambda: {"choices": ["None"] + [x for x in hypernetworks.keys()]}, refresh=reload_hypernetworks),
"sd_hypernetwork_strength": OptionInfo(1.0, "Hypernetwork strength", gr.Slider, {"minimum": 0.0, "maximum": 1.0, "step": 0.001}),
+ "inpainting_mask_weight": OptionInfo(1.0, "Inpainting conditioning mask strength", gr.Slider, {"minimum": 0.0, "maximum": 1.0, "step": 0.01}),
"img2img_color_correction": OptionInfo(False, "Apply color correction to img2img results to match original colors."),
"save_images_before_color_correction": OptionInfo(False, "Save a copy of image before applying color correction to img2img results"),
"img2img_fix_steps": OptionInfo(False, "With img2img, do exactly the amount of steps the slider specifies (normally you'd do less with less denoising)."),
diff --git a/scripts/xy_grid.py b/scripts/xy_grid.py index eff0c942..f5255786 100644 --- a/scripts/xy_grid.py +++ b/scripts/xy_grid.py @@ -153,7 +153,6 @@ def str_permutations(x): """dummy function for specifying it in AxisOption's type when you want to get a list of permutations"""
return x
-
AxisOption = namedtuple("AxisOption", ["label", "type", "apply", "format_value", "confirm"])
AxisOptionImg2Img = namedtuple("AxisOptionImg2Img", ["label", "type", "apply", "format_value", "confirm"])
@@ -178,6 +177,7 @@ axis_options = [ AxisOption("Eta", float, apply_field("eta"), format_value_add_label, None),
AxisOption("Clip skip", int, apply_clip_skip, format_value_add_label, None),
AxisOption("Denoising", float, apply_field("denoising_strength"), format_value_add_label, None),
+ AxisOption("Cond. Image Mask Weight", float, apply_field("inpainting_mask_weight"), format_value_add_label, None),
]
|