diff options
| -rw-r--r-- | README.md | 6 | ||||
| -rw-r--r-- | modules/processing.py | 19 | ||||
| -rw-r--r-- | modules/sd_samplers.py | 5 | ||||
| -rw-r--r-- | script.js | 23 | ||||
| -rw-r--r-- | scripts/poor_mans_outpainting.py | 32 |
5 files changed, 62 insertions, 23 deletions
@@ -42,9 +42,6 @@ A browser interface based on Gradio library for Stable Diffusion. You need [python](https://www.python.org/downloads/windows/) and [git](https://git-scm.com/download/win)
installed to run this, and an NVidia videocard.
-I tested the installation to work Windows with Python 3.8.10, and with Python 3.10.6. You may be able
-to have success with different versions.
-
You need `model.ckpt`, Stable Diffusion model checkpoint, a big file containing the neural network weights. You
can obtain it from the following places:
- [official download](https://huggingface.co/CompVis/stable-diffusion-v-1-4-original)
@@ -60,10 +57,9 @@ as model if it has .pth extension. Grab models from the [Model Database](https:/ - install [Python 3.10.6](https://www.python.org/downloads/windows/) and check "Add Python to PATH" during installation. You must install this exact version.
- install [git](https://git-scm.com/download/win)
-- install [CUDA 11.3](https://developer.nvidia.com/cuda-11.3.0-download-archive?target_os=Windows&target_arch=x86_64)
- place `model.ckpt` into webui directory, next to `webui.bat`.
- _*(optional)*_ place `GFPGANv1.3.pth` into webui directory, next to `webui.bat`.
-- run `webui.bat` from Windows Explorer.
+- run `webui.bat` from Windows Explorer. Run it as normal user, ***not*** as administrator.
#### Troublehooting:
diff --git a/modules/processing.py b/modules/processing.py index 49474b73..73b060f4 100644 --- a/modules/processing.py +++ b/modules/processing.py @@ -52,7 +52,7 @@ class StableDiffusionProcessing: self.overlay_images = overlay_images
self.paste_to = None
- def init(self):
+ def init(self, seed):
pass
def sample(self, x, conditioning, unconditional_conditioning):
@@ -155,7 +155,7 @@ def process_images(p: StableDiffusionProcessing) -> Processed: precision_scope = torch.autocast if cmd_opts.precision == "autocast" else contextlib.nullcontext
ema_scope = (contextlib.nullcontext if cmd_opts.lowvram else p.sd_model.ema_scope)
with torch.no_grad(), precision_scope("cuda"), ema_scope():
- p.init()
+ p.init(seed=all_seeds[0])
if state.job_count == -1:
state.job_count = p.n_iter
@@ -240,7 +240,7 @@ def process_images(p: StableDiffusionProcessing) -> Processed: class StableDiffusionProcessingTxt2Img(StableDiffusionProcessing):
sampler = None
- def init(self):
+ def init(self, seed):
self.sampler = samplers[self.sampler_index].constructor(self.sd_model)
def sample(self, x, conditioning, unconditional_conditioning):
@@ -320,7 +320,7 @@ class StableDiffusionProcessingImg2Img(StableDiffusionProcessing): self.mask = None
self.nmask = None
- def init(self):
+ def init(self, seed):
self.sampler = samplers_for_img2img[self.sampler_index].constructor(self.sd_model)
crop_region = None
@@ -347,11 +347,13 @@ class StableDiffusionProcessingImg2Img(StableDiffusionProcessing): else:
self.image_mask = images.resize_image(self.resize_mode, self.image_mask, self.width, self.height)
np_mask = np.array(self.image_mask)
- np_mask = 255 - np.clip((255 - np_mask.astype(np.float)) * 2, 0, 255).astype(np.uint8)
+ np_mask = np.clip((np_mask.astype(np.float)) * 2, 0, 255).astype(np.uint8)
self.mask_for_overlay = Image.fromarray(np_mask)
self.overlay_images = []
+ latent_mask = self.latent_mask if self.latent_mask is not None else self.image_mask
+
imgs = []
for img in self.init_images:
image = img.convert("RGB")
@@ -361,7 +363,7 @@ class StableDiffusionProcessingImg2Img(StableDiffusionProcessing): if self.image_mask is not None:
if self.inpainting_fill != 1:
- image = fill(image, self.mask_for_overlay)
+ image = fill(image, latent_mask)
image_masked = Image.new('RGBa', (image.width, image.height))
image_masked.paste(image.convert("RGBA").convert("RGBa"), mask=ImageOps.invert(self.mask_for_overlay.convert('L')))
@@ -394,17 +396,18 @@ class StableDiffusionProcessingImg2Img(StableDiffusionProcessing): self.init_latent = self.sd_model.get_first_stage_encoding(self.sd_model.encode_first_stage(image))
if self.image_mask is not None:
- init_mask = self.latent_mask if self.latent_mask is not None else self.image_mask
+ init_mask = latent_mask
latmask = init_mask.convert('RGB').resize((self.init_latent.shape[3], self.init_latent.shape[2]))
latmask = np.moveaxis(np.array(latmask, dtype=np.float64), 2, 0) / 255
latmask = latmask[0]
+ latmask = np.around(latmask)
latmask = np.tile(latmask[None], (4, 1, 1))
self.mask = torch.asarray(1.0 - latmask).to(shared.device).type(self.sd_model.dtype)
self.nmask = torch.asarray(latmask).to(shared.device).type(self.sd_model.dtype)
if self.inpainting_fill == 2:
- self.init_latent = self.init_latent * self.mask + create_random_tensors(self.init_latent.shape[1:], [self.seed + x + 1 for x in range(self.init_latent.shape[0])]) * self.nmask
+ self.init_latent = self.init_latent * self.mask + create_random_tensors(self.init_latent.shape[1:], [seed + x + 1 for x in range(self.init_latent.shape[0])]) * self.nmask
elif self.inpainting_fill == 3:
self.init_latent = self.init_latent * self.mask
diff --git a/modules/sd_samplers.py b/modules/sd_samplers.py index e8bc5be2..140b5dea 100644 --- a/modules/sd_samplers.py +++ b/modules/sd_samplers.py @@ -58,7 +58,10 @@ def p_sample_ddim_hook(sampler_wrapper, x_dec, cond, ts, *args, **kwargs): img_orig = sampler_wrapper.sampler.model.q_sample(sampler_wrapper.init_latent, ts)
x_dec = img_orig * sampler_wrapper.mask + sampler_wrapper.nmask * x_dec
- store_latent(x_dec)
+ store_latent(sampler_wrapper.init_latent * sampler_wrapper.mask + sampler_wrapper.nmask * x_dec)
+
+ else:
+ store_latent(x_dec)
return sampler_wrapper.orig_p_sample_ddim(x_dec, cond, ts, *args, **kwargs)
@@ -103,11 +103,31 @@ function addTitles(root){ } +tabNames = {"txt2img": 1, "img2img": 1, "Extras": 1, "PNG Info": 1, "Settings": 1} +processedTabs = {} + document.addEventListener("DOMContentLoaded", function() { var mutationObserver = new MutationObserver(function(m){ addTitles(gradioApp()); + + // fix for gradio breaking when you switch away from tab with mask + gradioApp().querySelectorAll('button').forEach(function(button){ + title = button.textContent.trim() + if(processedTabs[title]) return + if(tabNames[button.textContent.trim()]==null) return; + processedTabs[title]=1 + + button.onclick = function(){ + mask_buttons = gradioApp().querySelectorAll('#img2maskimg button'); + if(mask_buttons.length == 2){ + mask_buttons[1].click(); + } + } + }) }); mutationObserver.observe( gradioApp(), { childList:true, subtree:true }) + + }); function selected_gallery_index(){ @@ -150,6 +170,5 @@ function submit(){ for(var i=0;i<arguments.length;i++){ res.push(arguments[i]) } - console.log(res) return res -}
\ No newline at end of file +} diff --git a/scripts/poor_mans_outpainting.py b/scripts/poor_mans_outpainting.py index 08171877..c029c67f 100644 --- a/scripts/poor_mans_outpainting.py +++ b/scripts/poor_mans_outpainting.py @@ -21,7 +21,7 @@ class Script(scripts.Script): if not is_img2img:
return None
- pixels = gr.Slider(label="Pixels to expand", minimum=8, maximum=128, step=8)
+ pixels = gr.Slider(label="Pixels to expand", minimum=8, maximum=256, step=8, value=128)
mask_blur = gr.Slider(label='Mask blur', minimum=0, maximum=64, step=1, value=4, visible=False)
inpainting_fill = gr.Radio(label='Masked content', choices=['fill', 'original', 'latent noise', 'latent nothing'], value='fill', type="index", visible=False)
direction = gr.CheckboxGroup(label="Outpainting direction", choices=['left', 'right', 'up', 'down'], value=['left', 'right', 'up', 'down'])
@@ -32,7 +32,7 @@ class Script(scripts.Script): initial_seed = None
initial_info = None
- p.mask_blur = mask_blur
+ p.mask_blur = mask_blur * 2
p.inpainting_fill = inpainting_fill
p.inpaint_full_res = False
@@ -47,11 +47,14 @@ class Script(scripts.Script): if left > 0:
left = left * (target_w - init_img.width) // (left + right)
- right = target_w - init_img.width - left
+ if right > 0:
+ right = target_w - init_img.width - left
if up > 0:
up = up * (target_h - init_img.height) // (up + down)
- down = target_h - init_img.height - up
+
+ if down > 0:
+ down = target_h - init_img.height - up
img = Image.new("RGB", (target_w, target_h))
img.paste(init_img, (left, up))
@@ -67,13 +70,18 @@ class Script(scripts.Script): latent_mask = Image.new("L", (img.width, img.height), "white")
latent_draw = ImageDraw.Draw(latent_mask)
- latent_draw.rectangle((left + left//2, up + up//2, mask.width - right - right//2, mask.height - down - down//2), fill="black")
+ latent_draw.rectangle((
+ left + (mask_blur//2 if left > 0 else 0),
+ up + (mask_blur//2 if up > 0 else 0),
+ mask.width - right - (mask_blur//2 if right > 0 else 0),
+ mask.height - down - (mask_blur//2 if down > 0 else 0)
+ ), fill="black")
processing.torch_gc()
grid = images.split_grid(img, tile_w=p.width, tile_h=p.height, overlap=pixels)
grid_mask = images.split_grid(mask, tile_w=p.width, tile_h=p.height, overlap=pixels)
- grid_latent_mask = images.split_grid(mask, tile_w=p.width, tile_h=p.height, overlap=pixels)
+ grid_latent_mask = images.split_grid(latent_mask, tile_w=p.width, tile_h=p.height, overlap=pixels)
p.n_iter = 1
p.batch_size = 1
@@ -85,8 +93,13 @@ class Script(scripts.Script): work_latent_mask = []
work_results = []
- for (_, _, row), (_, _, row_mask), (_, _, row_latent_mask) in zip(grid.tiles, grid_mask.tiles, grid_latent_mask.tiles):
+ for (y, h, row), (_, _, row_mask), (_, _, row_latent_mask) in zip(grid.tiles, grid_mask.tiles, grid_latent_mask.tiles):
for tiledata, tiledata_mask, tiledata_latent_mask in zip(row, row_mask, row_latent_mask):
+ x, w = tiledata[0:2]
+
+ if x >= left and x+w <= img.width - right and y >= up and y+h <= img.height - down:
+ continue
+
work.append(tiledata[2])
work_mask.append(tiledata_mask[2])
work_latent_mask.append(tiledata_latent_mask[2])
@@ -115,6 +128,11 @@ class Script(scripts.Script): image_index = 0
for y, h, row in grid.tiles:
for tiledata in row:
+ x, w = tiledata[0:2]
+
+ if x >= left and x+w <= img.width - right and y >= up and y+h <= img.height - down:
+ continue
+
tiledata[2] = work_results[image_index] if image_index < len(work_results) else Image.new("RGB", (p.width, p.height))
image_index += 1
|