Merge branch 'release_candidate'

Protect alphas_cumprod during refiner switchover

CHANGELOG.md

@@ -7,11 +7,11 @@
 * Support for SDXL-Inpaint Model ([#14390](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14390))
 * Use Spandrel for upscaling and face restoration architectures ([#14425](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14425), [#14467](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14467), [#14473](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14473), [#14474](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14474), [#14477](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14477), [#14476](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14476), [#14484](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14484), [#14500](https://github.com/AUTOMATIC1111/stable-difusion-webui/pull/14500), [#14501](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14501), [#14504](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14504), [#14524](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14524), [#14809](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14809))
 * Automatic backwards version compatibility (when loading infotexts from old images with program version specified, will add compatibility settings)
-* Implement zero terminal SNR noise schedule option (**[SEED BREAKING CHANGE](https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Seed-breaking-changes#180-dev-170-225-2024-01-01---zero-terminal-snr-noise-schedule-option)**, [#14145](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14145))
+* Implement zero terminal SNR noise schedule option (**[SEED BREAKING CHANGE](https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Seed-breaking-changes#180-dev-170-225-2024-01-01---zero-terminal-snr-noise-schedule-option)**, [#14145](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14145), [#14979](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14979))
 * Add a [✨] button to run hires fix on selected image in the gallery (with help from [#14598](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14598), [#14626](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14626), [#14728](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14728))
 * [Separate assets repository](https://github.com/AUTOMATIC1111/stable-diffusion-webui-assets); serve fonts locally rather than from google's servers
 * Official LCM Sampler Support ([#14583](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14583))
-* Add support for DAT upscaler models ([#14690](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14690))
+* Add support for DAT upscaler models ([#14690](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14690), [#15039](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/15039))
 * Extra Networks Tree View ([#14588](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14588), [#14900](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14900))
 * NPU Support ([#14801](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14801))
 * Propmpt comments support
@@ -34,13 +34,17 @@
 * Add checkpoint info to csv log file when saving images ([#14663](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14663))
 * Make more columns resizable ([#14740](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14740), [#14884](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14884))
 * Add an option to not overlay original image for inpainting for #14727
-* Add Pad conds v0 option
+* Add Pad conds v0 option to support same generation with DDIM as before 1.6.0
 * Add "Interrupting..." placeholder.
 * Button for refresh extensions list ([#14857](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14857))
 * Add an option to disable normalization after calculating emphasis. ([#14874](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14874))
 * When counting tokens, also include enabled styles (can be disabled in settings to revert to previous behavior)
+* Configuration for the [📂] button for image gallery ([#14947](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14947))
+* Support inference with LyCORIS BOFT networks ([#14871](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14871), [#14973](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14973))
+* support resizable columns for touch (tablets) ([#15002](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/15002))
 
 ### Extensions and API:
+* Removed packages from requirements: basicsr, gfpgan, realesrgan; as well as their dependencies: absl-py, addict, beautifulsoup4, future, gdown, grpcio, importlib-metadata, lmdb, lpips, Markdown, platformdirs, PySocks, soupsieve, tb-nightly, tensorboard-data-server, tomli, Werkzeug, yapf, zipp, soupsieve
 * Enable task ids for API ([#14314](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14314))
 * add override_settings support for infotext API
 * rename generation_parameters_copypaste module to infotext_utils
@@ -55,6 +59,7 @@
 * modules/api/api.py: add api endpoint to refresh embeddings list ([#14715](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14715))
 * set_named_arg ([#14773](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14773))
 * add before_token_counter callback and use it for prompt comments
+* ResizeHandleRow - allow overriden column scale parameter ([#15004](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/15004))
 
 ### Performance
 * Massive performance improvement for extra networks directories with a huge number of files in them in an attempt to tackle #14507 ([#14528](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14528))
@@ -96,6 +101,10 @@
 * Gracefully handle mtime read exception from cache ([#14933](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14933))
 * Only trigger interrupt on `Esc` when interrupt button visible ([#14932](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14932))
 * Disable prompt token counters option actually disables token counting rather than just hiding results.
+* avoid doble upscaling in inpaint ([#14966](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14966))
+* Fix #14591 using translated content to do categories mapping ([#14995](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14995))
+* fix: the `split_threshold` parameter does not work when running Split oversized images ([#15006](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/15006))
+* Fix resize-handle for mobile ([#15010](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/15010), [#15065](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/15065))
 
 ### Other:
 * Assign id for "extra_options". Replace numeric field with slider. ([#14270](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14270))
@@ -116,8 +125,9 @@
 * extensions tab table row hover highlight ([#14885](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14885))
 * Always add timestamp to displayed image ([#14890](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14890))
 * Added core.filemode=false so doesn't track changes in file permission… ([#14930](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14930))
-* Normalize command-line argument paths ([#14934](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14934))
+* Normalize command-line argument paths ([#14934](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14934), [#15035](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/15035))
 * Use original App Title in progress bar ([#14916](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14916))
+* register_tmp_file also for mtime ([#15012](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/15012))
 
 ## 1.7.0
 
@@ -162,6 +172,7 @@
 * add FP32 fallback support on sd_vae_approx ([#14046](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14046))
 * support XYZ scripts / split hires path from unet ([#14126](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14126))
 * allow use of mutiple styles csv files ([#14125](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14125))
+* make extra network card description plaintext by default, with an option (Treat card description as HTML) to re-enable HTML as it was (originally by [#13241](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13241))
 
 ### Extensions and API:
 * update gradio to 3.41.2

extensions-builtin/Lora/network_oft.py

@@ -1,6 +1,5 @@
 import torch
 import network
-from lyco_helpers import factorization
 from einops import rearrange
 
 
@@ -22,20 +21,28 @@ class NetworkModuleOFT(network.NetworkModule):
         self.org_module: list[torch.Module] = [self.sd_module]
 
         self.scale = 1.0
+        self.is_R = False
+        self.is_boft = False
 
-        # kohya-ss
+        # kohya-ss/New LyCORIS OFT/BOFT
         if "oft_blocks" in weights.w.keys():
-            self.is_kohya = True
             self.oft_blocks = weights.w["oft_blocks"] # (num_blocks, block_size, block_size)
-            self.alpha = weights.w["alpha"] # alpha is constraint
+            self.alpha = weights.w.get("alpha", None) # alpha is constraint
             self.dim = self.oft_blocks.shape[0] # lora dim
-        # LyCORIS
+        # Old LyCORIS OFT
         elif "oft_diag" in weights.w.keys():
-            self.is_kohya = False
+            self.is_R = True
             self.oft_blocks = weights.w["oft_diag"]
             # self.alpha is unused
             self.dim = self.oft_blocks.shape[1] # (num_blocks, block_size, block_size)
 
+        # LyCORIS BOFT
+        if self.oft_blocks.dim() == 4:
+            self.is_boft = True
+        self.rescale = weights.w.get('rescale', None)
+        if self.rescale is not None:
+            self.rescale = self.rescale.reshape(-1, *[1]*(self.org_module[0].weight.dim() - 1))
+
         is_linear = type(self.sd_module) in [torch.nn.Linear, torch.nn.modules.linear.NonDynamicallyQuantizableLinear]
         is_conv = type(self.sd_module) in [torch.nn.Conv2d]
         is_other_linear = type(self.sd_module) in [torch.nn.MultiheadAttention] # unsupported
@@ -47,35 +54,64 @@ class NetworkModuleOFT(network.NetworkModule):
         elif is_other_linear:
             self.out_dim = self.sd_module.embed_dim
 
-        if self.is_kohya:
-            self.constraint = self.alpha * self.out_dim
-            self.num_blocks = self.dim
-            self.block_size = self.out_dim // self.dim
-        else:
+        self.num_blocks = self.dim
+        self.block_size = self.out_dim // self.dim
+        self.constraint = (0 if self.alpha is None else self.alpha) * self.out_dim
+        if self.is_R:
             self.constraint = None
-            self.block_size, self.num_blocks = factorization(self.out_dim, self.dim)
+            self.block_size = self.dim
+            self.num_blocks = self.out_dim // self.dim
+        elif self.is_boft:
+            self.boft_m = self.oft_blocks.shape[0]
+            self.num_blocks = self.oft_blocks.shape[1]
+            self.block_size = self.oft_blocks.shape[2]
+            self.boft_b = self.block_size
 
     def calc_updown(self, orig_weight):
         oft_blocks = self.oft_blocks.to(orig_weight.device)
         eye = torch.eye(self.block_size, device=oft_blocks.device)
 
-        if self.is_kohya:
-            block_Q = oft_blocks - oft_blocks.transpose(1, 2) # ensure skew-symmetric orthogonal matrix
-            norm_Q = torch.norm(block_Q.flatten())
-            new_norm_Q = torch.clamp(norm_Q, max=self.constraint.to(oft_blocks.device))
-            block_Q = block_Q * ((new_norm_Q + 1e-8) / (norm_Q + 1e-8))
+        if not self.is_R:
+            block_Q = oft_blocks - oft_blocks.transpose(-1, -2) # ensure skew-symmetric orthogonal matrix
+            if self.constraint != 0:
+                norm_Q = torch.norm(block_Q.flatten())
+                new_norm_Q = torch.clamp(norm_Q, max=self.constraint.to(oft_blocks.device))
+                block_Q = block_Q * ((new_norm_Q + 1e-8) / (norm_Q + 1e-8))
             oft_blocks = torch.matmul(eye + block_Q, (eye - block_Q).float().inverse())
 
         R = oft_blocks.to(orig_weight.device)
 
-        # This errors out for MultiheadAttention, might need to be handled up-stream
-        merged_weight = rearrange(orig_weight, '(k n) ... -> k n ...', k=self.num_blocks, n=self.block_size)
-        merged_weight = torch.einsum(
-            'k n m, k n ... -> k m ...',
-            R,
-            merged_weight
-        )
-        merged_weight = rearrange(merged_weight, 'k m ... -> (k m) ...')
+        if not self.is_boft:
+            # This errors out for MultiheadAttention, might need to be handled up-stream
+            merged_weight = rearrange(orig_weight, '(k n) ... -> k n ...', k=self.num_blocks, n=self.block_size)
+            merged_weight = torch.einsum(
+                'k n m, k n ... -> k m ...',
+                R,
+                merged_weight
+            )
+            merged_weight = rearrange(merged_weight, 'k m ... -> (k m) ...')
+        else:
+            # TODO: determine correct value for scale
+            scale = 1.0
+            m = self.boft_m
+            b = self.boft_b
+            r_b = b // 2
+            inp = orig_weight
+            for i in range(m):
+                bi = R[i] # b_num, b_size, b_size
+                if i == 0:
+                    # Apply multiplier/scale and rescale into first weight
+                    bi = bi * scale + (1 - scale) * eye
+                inp = rearrange(inp, "(c g k) ... -> (c k g) ...", g=2, k=2**i * r_b)
+                inp = rearrange(inp, "(d b) ... -> d b ...", b=b)
+                inp = torch.einsum("b i j, b j ... -> b i ...", bi, inp)
+                inp = rearrange(inp, "d b ... -> (d b) ...")
+                inp = rearrange(inp, "(c k g) ... -> (c g k) ...", g=2, k=2**i * r_b)
+            merged_weight = inp
+
+        # Rescale mechanism
+        if self.rescale is not None:
+            merged_weight = self.rescale.to(merged_weight) * merged_weight
 
         updown = merged_weight.to(orig_weight.device) - orig_weight.to(merged_weight.dtype)
         output_shape = orig_weight.shape

javascript/resizeHandle.js

@@ -1,8 +1,8 @@
 (function() {
     const GRADIO_MIN_WIDTH = 320;
-    const GRID_TEMPLATE_COLUMNS = '1fr 16px 1fr';
     const PAD = 16;
     const DEBOUNCE_TIME = 100;
+    const DOUBLE_TAP_DELAY = 200; //ms
 
     const R = {
         tracking: false,
@@ -11,6 +11,7 @@
         leftCol: null,
         leftColStartWidth: null,
         screenX: null,
+        lastTapTime: null,
     };
 
     let resizeTimer;
@@ -23,21 +24,17 @@
     function displayResizeHandle(parent) {
         if (window.innerWidth < GRADIO_MIN_WIDTH * 2 + PAD * 4) {
             parent.style.display = 'flex';
-            if (R.handle != null) {
-                R.handle.style.opacity = '0';
-            }
+            parent.resizeHandle.style.display = "none";
             return false;
         } else {
             parent.style.display = 'grid';
-            if (R.handle != null) {
-                R.handle.style.opacity = '100';
-            }
+            parent.resizeHandle.style.display = "block";
             return true;
         }
     }
 
     function afterResize(parent) {
-        if (displayResizeHandle(parent) && parent.style.gridTemplateColumns != GRID_TEMPLATE_COLUMNS) {
+        if (displayResizeHandle(parent) && parent.style.gridTemplateColumns != parent.style.originalGridTemplateColumns) {
             const oldParentWidth = R.parentWidth;
             const newParentWidth = parent.offsetWidth;
             const widthL = parseInt(parent.style.gridTemplateColumns.split(' ')[0]);
@@ -52,6 +49,14 @@
     }
 
     function setup(parent) {
+
+        function onDoubleClick(evt) {
+            evt.preventDefault();
+            evt.stopPropagation();
+
+            parent.style.gridTemplateColumns = parent.style.originalGridTemplateColumns;
+        }
+
         const leftCol = parent.firstElementChild;
         const rightCol = parent.lastElementChild;
 
@@ -59,63 +64,97 @@
 
         parent.style.display = 'grid';
         parent.style.gap = '0';
-        parent.style.gridTemplateColumns = GRID_TEMPLATE_COLUMNS;
+        const gridTemplateColumns = `${parent.children[0].style.flexGrow}fr ${PAD}px ${parent.children[1].style.flexGrow}fr`;
+        parent.style.gridTemplateColumns = gridTemplateColumns;
+        parent.style.originalGridTemplateColumns = gridTemplateColumns;
 
         const resizeHandle = document.createElement('div');
         resizeHandle.classList.add('resize-handle');
         parent.insertBefore(resizeHandle, rightCol);
+        parent.resizeHandle = resizeHandle;
 
-        resizeHandle.addEventListener('mousedown', (evt) => {
-            if (evt.button !== 0) return;
+        ['mousedown', 'touchstart'].forEach((eventType) => {
+            resizeHandle.addEventListener(eventType, (evt) => {
+                if (eventType.startsWith('mouse')) {
+                    if (evt.button !== 0) return;
+                } else {
+                    if (evt.changedTouches.length !== 1) return;
 
-            evt.preventDefault();
-            evt.stopPropagation();
+                    const currentTime = new Date().getTime();
+                    if (R.lastTapTime && currentTime - R.lastTapTime <= DOUBLE_TAP_DELAY) {
+                        onDoubleClick(evt);
+                        return;
+                    }
 
-            document.body.classList.add('resizing');
+                    R.lastTapTime = currentTime;
+                }
 
-            R.tracking = true;
-            R.parent = parent;
-            R.parentWidth = parent.offsetWidth;
-            R.handle = resizeHandle;
-            R.leftCol = leftCol;
-            R.leftColStartWidth = leftCol.offsetWidth;
-            R.screenX = evt.screenX;
+                evt.preventDefault();
+                evt.stopPropagation();
+
+                document.body.classList.add('resizing');
+
+                R.tracking = true;
+                R.parent = parent;
+                R.parentWidth = parent.offsetWidth;
+                R.leftCol = leftCol;
+                R.leftColStartWidth = leftCol.offsetWidth;
+                if (eventType.startsWith('mouse')) {
+                    R.screenX = evt.screenX;
+                } else {
+                    R.screenX = evt.changedTouches[0].screenX;
+                }
+            });
         });
 
-        resizeHandle.addEventListener('dblclick', (evt) => {
-            evt.preventDefault();
-            evt.stopPropagation();
-
-            parent.style.gridTemplateColumns = GRID_TEMPLATE_COLUMNS;
-        });
+        resizeHandle.addEventListener('dblclick', onDoubleClick);
 
         afterResize(parent);
     }
 
-    window.addEventListener('mousemove', (evt) => {
-        if (evt.button !== 0) return;
+    ['mousemove', 'touchmove'].forEach((eventType) => {
+        window.addEventListener(eventType, (evt) => {
+            if (eventType.startsWith('mouse')) {
+                if (evt.button !== 0) return;
+            } else {
+                if (evt.changedTouches.length !== 1) return;
+            }
 
-        if (R.tracking) {
-            evt.preventDefault();
-            evt.stopPropagation();
+            if (R.tracking) {
+                if (eventType.startsWith('mouse')) {
+                    evt.preventDefault();
+                }
+                evt.stopPropagation();
 
-            const delta = R.screenX - evt.screenX;
-            const leftColWidth = Math.max(Math.min(R.leftColStartWidth - delta, R.parent.offsetWidth - GRADIO_MIN_WIDTH - PAD), GRADIO_MIN_WIDTH);
-            setLeftColGridTemplate(R.parent, leftColWidth);
-        }
+                let delta = 0;
+                if (eventType.startsWith('mouse')) {
+                    delta = R.screenX - evt.screenX;
+                } else {
+                    delta = R.screenX - evt.changedTouches[0].screenX;
+                }
+                const leftColWidth = Math.max(Math.min(R.leftColStartWidth - delta, R.parent.offsetWidth - GRADIO_MIN_WIDTH - PAD), GRADIO_MIN_WIDTH);
+                setLeftColGridTemplate(R.parent, leftColWidth);
+            }
+        });
     });
 
-    window.addEventListener('mouseup', (evt) => {
-        if (evt.button !== 0) return;
+    ['mouseup', 'touchend'].forEach((eventType) => {
+        window.addEventListener(eventType, (evt) => {
+            if (eventType.startsWith('mouse')) {
+                if (evt.button !== 0) return;
+            } else {
+                if (evt.changedTouches.length !== 1) return;
+            }
 
-        if (R.tracking) {
-            evt.preventDefault();
-            evt.stopPropagation();
+            if (R.tracking) {
+                evt.preventDefault();
+                evt.stopPropagation();
 
-            R.tracking = false;
+                R.tracking = false;
 
-            document.body.classList.remove('resizing');
-        }
+                document.body.classList.remove('resizing');
+            }
+        });
     });
 
 

javascript/settings.js

@@ -55,8 +55,8 @@ onOptionsChanged(function() {
     });
 
     opts._categories.forEach(function(x) {
-        var section = x[0];
-        var category = x[1];
+        var section = localization[x[0]] ?? x[0];
+        var category = localization[x[1]] ?? x[1];
 
         var span = document.createElement('SPAN');
         span.textContent = category;

modules/cmd_args.py

@@ -53,6 +53,7 @@ parser.add_argument("--gfpgan-models-path", type=normalized_filepath, help="Path
 parser.add_argument("--esrgan-models-path", type=normalized_filepath, help="Path to directory with ESRGAN model file(s).", default=os.path.join(models_path, 'ESRGAN'))
 parser.add_argument("--bsrgan-models-path", type=normalized_filepath, help="Path to directory with BSRGAN model file(s).", default=os.path.join(models_path, 'BSRGAN'))
 parser.add_argument("--realesrgan-models-path", type=normalized_filepath, help="Path to directory with RealESRGAN model file(s).", default=os.path.join(models_path, 'RealESRGAN'))
+parser.add_argument("--dat-models-path", type=normalized_filepath, help="Path to directory with DAT model file(s).", default=os.path.join(models_path, 'DAT'))
 parser.add_argument("--clip-models-path", type=normalized_filepath, help="Path to directory with CLIP model file(s).", default=None)
 parser.add_argument("--xformers", action='store_true', help="enable xformers for cross attention layers")
 parser.add_argument("--force-enable-xformers", action='store_true', help="enable xformers for cross attention layers regardless of whether the checking code thinks you can run it; do not make bug reports if this fails to work")

modules/paths_internal.py

@@ -7,7 +7,7 @@ import shlex
 from pathlib import Path
 
 
-normalized_filepath = lambda filepath: str(Path(filepath).resolve())
+normalized_filepath = lambda filepath: str(Path(filepath).absolute())
 
 commandline_args = os.environ.get('COMMANDLINE_ARGS', "")
 sys.argv += shlex.split(commandline_args)

modules/processing.py

@@ -74,16 +74,18 @@ def uncrop(image, dest_size, paste_loc):
 
 def apply_overlay(image, paste_loc, overlay):
     if overlay is None:
-        return image
+        return image, image.copy()
 
     if paste_loc is not None:
         image = uncrop(image, (overlay.width, overlay.height), paste_loc)
 
+    original_denoised_image = image.copy()
+
     image = image.convert('RGBA')
     image.alpha_composite(overlay)
     image = image.convert('RGB')
 
-    return image
+    return image, original_denoised_image
 
 def create_binary_mask(image, round=True):
     if image.mode == 'RGBA' and image.getextrema()[-1] != (255, 255):
@@ -913,33 +915,7 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed:
             if p.n_iter > 1:
                 shared.state.job = f"Batch {n+1} out of {p.n_iter}"
 
-            def rescale_zero_terminal_snr_abar(alphas_cumprod):
-                alphas_bar_sqrt = alphas_cumprod.sqrt()
-
-                # Store old values.
-                alphas_bar_sqrt_0 = alphas_bar_sqrt[0].clone()
-                alphas_bar_sqrt_T = alphas_bar_sqrt[-1].clone()
-
-                # Shift so the last timestep is zero.
-                alphas_bar_sqrt -= (alphas_bar_sqrt_T)
-
-                # Scale so the first timestep is back to the old value.
-                alphas_bar_sqrt *= alphas_bar_sqrt_0 / (alphas_bar_sqrt_0 - alphas_bar_sqrt_T)
-
-                # Convert alphas_bar_sqrt to betas
-                alphas_bar = alphas_bar_sqrt**2  # Revert sqrt
-                alphas_bar[-1] = 4.8973451890853435e-08
-                return alphas_bar
-
-            if hasattr(p.sd_model, 'alphas_cumprod') and hasattr(p.sd_model, 'alphas_cumprod_original'):
-                p.sd_model.alphas_cumprod = p.sd_model.alphas_cumprod_original.to(shared.device)
-
-                if opts.use_downcasted_alpha_bar:
-                    p.extra_generation_params['Downcast alphas_cumprod'] = opts.use_downcasted_alpha_bar
-                    p.sd_model.alphas_cumprod = p.sd_model.alphas_cumprod.half().to(shared.device)
-                if opts.sd_noise_schedule == "Zero Terminal SNR":
-                    p.extra_generation_params['Noise Schedule'] = opts.sd_noise_schedule
-                    p.sd_model.alphas_cumprod = rescale_zero_terminal_snr_abar(p.sd_model.alphas_cumprod).to(shared.device)
+            sd_models.apply_alpha_schedule_override(p.sd_model, p)
 
             with devices.without_autocast() if devices.unet_needs_upcast else devices.autocast():
                 samples_ddim = p.sample(conditioning=p.c, unconditional_conditioning=p.uc, seeds=p.seeds, subseeds=p.subseeds, subseed_strength=p.subseed_strength, prompts=p.prompts)
@@ -1021,7 +997,7 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed:
 
                 if p.color_corrections is not None and i < len(p.color_corrections):
                     if save_samples and opts.save_images_before_color_correction:
-                        image_without_cc = apply_overlay(image, p.paste_to, overlay_image)
+                        image_without_cc, _ = apply_overlay(image, p.paste_to, overlay_image)
                         images.save_image(image_without_cc, p.outpath_samples, "", p.seeds[i], p.prompts[i], opts.samples_format, info=infotext(i), p=p, suffix="-before-color-correction")
                     image = apply_color_correction(p.color_corrections[i], image)
 
@@ -1029,12 +1005,7 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed:
                 # that is being composited over the original image,
                 # we need to keep the original image around
                 # and use it in the composite step.
-                original_denoised_image = image.copy()
-
-                if p.paste_to is not None:
-                    original_denoised_image = uncrop(original_denoised_image, (overlay_image.width, overlay_image.height), p.paste_to)
-
-                image = apply_overlay(image, p.paste_to, overlay_image)
+                image, original_denoised_image = apply_overlay(image, p.paste_to, overlay_image)
 
                 if p.scripts is not None:
                     pp = scripts.PostprocessImageArgs(image)

modules/sd_models.py

@@ -15,6 +15,7 @@ from ldm.util import instantiate_from_config
 
 from modules import paths, shared, modelloader, devices, script_callbacks, sd_vae, sd_disable_initialization, errors, hashes, sd_models_config, sd_unet, sd_models_xl, cache, extra_networks, processing, lowvram, sd_hijack, patches
 from modules.timer import Timer
+from modules.shared import opts
 import tomesd
 import numpy as np
 
@@ -427,6 +428,8 @@ def load_model_weights(model, checkpoint_info: CheckpointInfo, state_dict, timer
         devices.dtype_unet = torch.float16
         timer.record("apply half()")
 
+    apply_alpha_schedule_override(model)
+
     for module in model.modules():
         if hasattr(module, 'fp16_weight'):
             del module.fp16_weight
@@ -550,6 +553,48 @@ def repair_config(sd_config):
         sd_config.model.params.noise_aug_config.params.clip_stats_path = sd_config.model.params.noise_aug_config.params.clip_stats_path.replace("checkpoints/karlo_models", karlo_path)
 
 
+def rescale_zero_terminal_snr_abar(alphas_cumprod):
+    alphas_bar_sqrt = alphas_cumprod.sqrt()
+
+    # Store old values.
+    alphas_bar_sqrt_0 = alphas_bar_sqrt[0].clone()
+    alphas_bar_sqrt_T = alphas_bar_sqrt[-1].clone()
+
+    # Shift so the last timestep is zero.
+    alphas_bar_sqrt -= (alphas_bar_sqrt_T)
+
+    # Scale so the first timestep is back to the old value.
+    alphas_bar_sqrt *= alphas_bar_sqrt_0 / (alphas_bar_sqrt_0 - alphas_bar_sqrt_T)
+
+    # Convert alphas_bar_sqrt to betas
+    alphas_bar = alphas_bar_sqrt ** 2  # Revert sqrt
+    alphas_bar[-1] = 4.8973451890853435e-08
+    return alphas_bar
+
+
+def apply_alpha_schedule_override(sd_model, p=None):
+    """
+    Applies an override to the alpha schedule of the model according to settings.
+    - downcasts the alpha schedule to half precision
+    - rescales the alpha schedule to have zero terminal SNR
+    """
+
+    if not hasattr(sd_model, 'alphas_cumprod') or not hasattr(sd_model, 'alphas_cumprod_original'):
+        return
+
+    sd_model.alphas_cumprod = sd_model.alphas_cumprod_original.to(shared.device)
+
+    if opts.use_downcasted_alpha_bar:
+        if p is not None:
+            p.extra_generation_params['Downcast alphas_cumprod'] = opts.use_downcasted_alpha_bar
+        sd_model.alphas_cumprod = sd_model.alphas_cumprod.half().to(shared.device)
+
+    if opts.sd_noise_schedule == "Zero Terminal SNR":
+        if p is not None:
+            p.extra_generation_params['Noise Schedule'] = opts.sd_noise_schedule
+        sd_model.alphas_cumprod = rescale_zero_terminal_snr_abar(sd_model.alphas_cumprod).to(shared.device)
+
+
 sd1_clip_weight = 'cond_stage_model.transformer.text_model.embeddings.token_embedding.weight'
 sd2_clip_weight = 'cond_stage_model.model.transformer.resblocks.0.attn.in_proj_weight'
 sdxl_clip_weight = 'conditioner.embedders.1.model.ln_final.weight'

modules/sd_samplers_cfg_denoiser.py

@@ -220,10 +220,10 @@ class CFGDenoiser(torch.nn.Module):
 
         self.padded_cond_uncond = False
         self.padded_cond_uncond_v0 = False
-        if shared.opts.pad_cond_uncond and tensor.shape[1] != uncond.shape[1]:
-            tensor, uncond = self.pad_cond_uncond(tensor, uncond)
-        elif shared.opts.pad_cond_uncond_v0 and tensor.shape[1] != uncond.shape[1]:
+        if shared.opts.pad_cond_uncond_v0 and tensor.shape[1] != uncond.shape[1]:
             tensor, uncond = self.pad_cond_uncond_v0(tensor, uncond)
+        elif shared.opts.pad_cond_uncond and tensor.shape[1] != uncond.shape[1]:
+            tensor, uncond = self.pad_cond_uncond(tensor, uncond)
 
         if tensor.shape[1] == uncond.shape[1] or skip_uncond:
             if is_edit_model:

modules/shared_options.py

@@ -211,7 +211,7 @@ options_templates.update(options_section(('optimizations', "Optimizations", "sd"
     "token_merging_ratio_img2img": OptionInfo(0.0, "Token merging ratio for img2img", gr.Slider, {"minimum": 0.0, "maximum": 0.9, "step": 0.1}).info("only applies if non-zero and overrides above"),
     "token_merging_ratio_hr": OptionInfo(0.0, "Token merging ratio for high-res pass", gr.Slider, {"minimum": 0.0, "maximum": 0.9, "step": 0.1}, infotext='Token merging ratio hr').info("only applies if non-zero and overrides above"),
     "pad_cond_uncond": OptionInfo(False, "Pad prompt/negative prompt", infotext='Pad conds').info("improves performance when prompt and negative prompt have different lengths; changes seeds"),
-    "pad_cond_uncond_v0": OptionInfo(False, "Pad prompt/negative prompt (v0)", infotext='Pad conds v0').info("alternative implementation for the above; used prior to 1.6.0 for DDIM sampler; ignored if the above is set; changes seeds"),
+    "pad_cond_uncond_v0": OptionInfo(False, "Pad prompt/negative prompt (v0)", infotext='Pad conds v0').info("alternative implementation for the above; used prior to 1.6.0 for DDIM sampler; overrides the above if set; WARNING: truncates negative prompt if it's too long; changes seeds"),
     "persistent_cond_cache": OptionInfo(True, "Persistent cond cache").info("do not recalculate conds from prompts if prompts have not changed since previous calculation"),
     "batch_cond_uncond": OptionInfo(True, "Batch cond/uncond").info("do both conditional and unconditional denoising in one batch; uses a bit more VRAM during sampling, but improves speed; previously this was controlled by --always-batch-cond-uncond comandline argument"),
     "fp8_storage": OptionInfo("Disable", "FP8 weight", gr.Radio, {"choices": ["Disable", "Enable for SDXL", "Enable"]}).info("Use FP8 to store Linear/Conv layers' weight. Require pytorch>=2.1.0."),
@@ -254,6 +254,7 @@ options_templates.update(options_section(('extra_networks', "Extra Networks", "s
     "extra_networks_card_height": OptionInfo(0, "Card height for Extra Networks").info("in pixels"),
     "extra_networks_card_text_scale": OptionInfo(1.0, "Card text scale", gr.Slider, {"minimum": 0.0, "maximum": 2.0, "step": 0.01}).info("1 = original size"),
     "extra_networks_card_show_desc": OptionInfo(True, "Show description on card"),
+    "extra_networks_card_description_is_html": OptionInfo(False, "Treat card description as HTML"),
     "extra_networks_card_order_field": OptionInfo("Path", "Default order field for Extra Networks cards", gr.Dropdown, {"choices": ['Path', 'Name', 'Date Created', 'Date Modified']}).needs_reload_ui(),
     "extra_networks_card_order": OptionInfo("Ascending", "Default order for Extra Networks cards", gr.Dropdown, {"choices": ['Ascending', 'Descending']}).needs_reload_ui(),
     "extra_networks_tree_view_default_enabled": OptionInfo(False, "Enables the Extra Networks directory tree view by default").needs_reload_ui(),
@@ -284,6 +285,7 @@ options_templates.update(options_section(('ui_gallery', "Gallery", "ui"), {
     "sd_webui_modal_lightbox_icon_opacity": OptionInfo(1, "Full page image viewer: control icon unfocused opacity", gr.Slider, {"minimum": 0.0, "maximum": 1, "step": 0.01}, onchange=shared.reload_gradio_theme).info('for mouse only').needs_reload_ui(),
     "sd_webui_modal_lightbox_toolbar_opacity": OptionInfo(0.9, "Full page image viewer: tool bar opacity", gr.Slider, {"minimum": 0.0, "maximum": 1, "step": 0.01}, onchange=shared.reload_gradio_theme).info('for mouse only').needs_reload_ui(),
     "gallery_height": OptionInfo("", "Gallery height", gr.Textbox).info("can be any valid CSS value, for example 768px or 20em").needs_reload_ui(),
+    "open_dir_button_choice": OptionInfo("Subdirectory", "What directory the [📂] button opens", gr.Radio, {"choices": ["Output Root", "Subdirectory", "Subdirectory (even temp dir)"]}),
 }))
 
 options_templates.update(options_section(('ui_alternatives', "UI alternatives", "ui"), {

modules/ui_common.py

@@ -9,7 +9,7 @@ import sys
 import gradio as gr
 import subprocess as sp
 
-from modules import call_queue, shared
+from modules import call_queue, shared, ui_tempdir
 from modules.infotext_utils import image_from_url_text
 import modules.images
 from modules.ui_components import ToolButton
@@ -164,29 +164,43 @@ class OutputPanel:
 def create_output_panel(tabname, outdir, toprow=None):
     res = OutputPanel()
 
-    def open_folder(f):
+    def open_folder(f, images=None, index=None):
+        if shared.cmd_opts.hide_ui_dir_config:
+            return
+
+        try:
+            if 'Sub' in shared.opts.open_dir_button_choice:
+                image_dir = os.path.split(images[index]["name"].rsplit('?', 1)[0])[0]
+                if 'temp' in shared.opts.open_dir_button_choice or not ui_tempdir.is_gradio_temp_path(image_dir):
+                    f = image_dir
+        except Exception:
+            pass
+
         if not os.path.exists(f):
-            print(f'Folder "{f}" does not exist. After you create an image, the folder will be created.')
+            msg = f'Folder "{f}" does not exist. After you create an image, the folder will be created.'
+            print(msg)
+            gr.Info(msg)
             return
         elif not os.path.isdir(f):
-            print(f"""
+            msg = f"""
 WARNING
 An open_folder request was made with an argument that is not a folder.
 This could be an error or a malicious attempt to run code on your computer.
 Requested path was: {f}
-""", file=sys.stderr)
+"""
+            print(msg, file=sys.stderr)
+            gr.Warning(msg)
             return
 
-        if not shared.cmd_opts.hide_ui_dir_config:
-            path = os.path.normpath(f)
-            if platform.system() == "Windows":
-                os.startfile(path)
-            elif platform.system() == "Darwin":
-                sp.Popen(["open", path])
-            elif "microsoft-standard-WSL2" in platform.uname().release:
-                sp.Popen(["wsl-open", path])
-            else:
-                sp.Popen(["xdg-open", path])
+        path = os.path.normpath(f)
+        if platform.system() == "Windows":
+            os.startfile(path)
+        elif platform.system() == "Darwin":
+            sp.Popen(["open", path])
+        elif "microsoft-standard-WSL2" in platform.uname().release:
+            sp.Popen(["wsl-open", path])
+        else:
+            sp.Popen(["xdg-open", path])
 
     with gr.Column(elem_id=f"{tabname}_results"):
         if toprow:
@@ -213,8 +227,12 @@ Requested path was: {f}
                     res.button_upscale = ToolButton('✨', elem_id=f'{tabname}_upscale', tooltip="Create an upscaled version of the current image using hires fix settings.")
 
             open_folder_button.click(
-                fn=lambda: open_folder(shared.opts.outdir_samples or outdir),
-                inputs=[],
+                fn=lambda images, index: open_folder(shared.opts.outdir_samples or outdir, images, index),
+                _js="(y, w) => [y, selected_gallery_index()]",
+                inputs=[
+                    res.gallery,
+                    open_folder_button,  # placeholder for index
+                ],
                 outputs=[],
             )
 

modules/ui_extra_networks.py

@@ -289,12 +289,16 @@ class ExtraNetworksPage:
                 }
             )
 
+        description = (item.get("description", "") or "" if shared.opts.extra_networks_card_show_desc else "")
+        if not shared.opts.extra_networks_card_description_is_html:
+            description = html.escape(description)
+
         # Some items here might not be used depending on HTML template used.
         args = {
             "background_image": background_image,
             "card_clicked": onclick,
             "copy_path_button": btn_copy_path,
-            "description": (item.get("description", "") or "" if shared.opts.extra_networks_card_show_desc else ""),
+            "description": description,
             "edit_button": btn_edit_item,
             "local_preview": quote_js(item["local_preview"]),
             "metadata_button": btn_metadata,
@@ -472,7 +476,7 @@ class ExtraNetworksPage:
 
         return f"<ul class='tree-list tree-list--tree'>{res}</ul>"
 
-    def create_card_view_html(self, tabname: str) -> str:
+    def create_card_view_html(self, tabname: str, *, none_message) -> str:
         """Generates HTML for the network Card View section for a tab.
 
         This HTML goes into the `extra-networks-pane.html` <div> with
@@ -480,6 +484,7 @@ class ExtraNetworksPage:
 
         Args:
             tabname: The name of the active tab.
+            none_message: HTML text to show when there are no cards.
 
         Returns:
             HTML formatted string.
@@ -490,24 +495,28 @@ class ExtraNetworksPage:
 
         if res == "":
             dirs = "".join([f"<li>{x}</li>" for x in self.allowed_directories_for_previews()])
-            res = shared.html("extra-networks-no-cards.html").format(dirs=dirs)
+            res = none_message or shared.html("extra-networks-no-cards.html").format(dirs=dirs)
 
         return res
 
-    def create_html(self, tabname):
+    def create_html(self, tabname, *, empty=False):
         """Generates an HTML string for the current pane.
 
         The generated HTML uses `extra-networks-pane.html` as a template.
 
         Args:
             tabname: The name of the active tab.
+            empty: create an empty HTML page with no items
 
         Returns:
             HTML formatted string.
         """
         self.lister.reset()
         self.metadata = {}
-        self.items = {x["name"]: x for x in self.list_items()}
+
+        items_list = [] if empty else self.list_items()
+        self.items = {x["name"]: x for x in items_list}
+
         # Populate the instance metadata for each item.
         for item in self.items.values():
             metadata = item.get("metadata")
@@ -536,7 +545,7 @@ class ExtraNetworksPage:
                 "tree_view_btn_extra_class": tree_view_btn_extra_class,
                 "tree_view_div_extra_class": tree_view_div_extra_class,
                 "tree_html": self.create_tree_view_html(tabname),
-                "items_html": self.create_card_view_html(tabname),
+                "items_html": self.create_card_view_html(tabname, none_message="Loading..." if empty else None),
             }
         )
 
@@ -655,7 +664,7 @@ def create_ui(interface: gr.Blocks, unrelated_tabs, tabname):
                 pass
 
             elem_id = f"{tabname}_{page.extra_networks_tabname}_cards_html"
-            page_elem = gr.HTML('Loading...', elem_id=elem_id)
+            page_elem = gr.HTML(page.create_html(tabname, empty=True), elem_id=elem_id)
             ui.pages.append(page_elem)
             editor = page.create_user_metadata_editor(ui, tabname)
             editor.create_ui()

modules/ui_tempdir.py

@@ -35,7 +35,9 @@ def save_pil_to_file(self, pil_image, dir=None, format="png"):
     already_saved_as = getattr(pil_image, 'already_saved_as', None)
     if already_saved_as and os.path.isfile(already_saved_as):
         register_tmp_file(shared.demo, already_saved_as)
-        return f'{already_saved_as}?{os.path.getmtime(already_saved_as)}'
+        filename_with_mtime = f'{already_saved_as}?{os.path.getmtime(already_saved_as)}'
+        register_tmp_file(shared.demo, filename_with_mtime)
+        return filename_with_mtime
 
     if shared.opts.temp_dir != "":
         dir = shared.opts.temp_dir
@@ -81,3 +83,18 @@ def cleanup_tmpdr():
 
             filename = os.path.join(root, name)
             os.remove(filename)
+
+
+def is_gradio_temp_path(path):
+    """
+    Check if the path is a temp dir used by gradio
+    """
+    path = Path(path)
+    if shared.opts.temp_dir and path.is_relative_to(shared.opts.temp_dir):
+        return True
+    if gradio_temp_dir := os.environ.get("GRADIO_TEMP_DIR"):
+        if path.is_relative_to(gradio_temp_dir):
+            return True
+    if path.is_relative_to(Path(tempfile.gettempdir()) / "gradio"):
+        return True
+    return False

script.js

@@ -167,7 +167,7 @@ document.addEventListener('keydown', function(e) {
         const lightboxModal = document.querySelector('#lightboxModal');
         if (!globalPopup || globalPopup.style.display === 'none') {
             if (document.activeElement === lightboxModal) return;
-            if (interruptButton.style.display !== 'none') {
+            if (interruptButton.style.display === 'block') {
                 interruptButton.click();
                 e.preventDefault();
             }

scripts/postprocessing_split_oversized.py

@@ -61,7 +61,7 @@ class ScriptPostprocessingSplitOversized(scripts_postprocessing.ScriptPostproces
             ratio = (pp.image.height * width) / (pp.image.width * height)
             inverse_xy = True
 
-        if ratio >= 1.0 and ratio > split_threshold:
+        if ratio >= 1.0 or ratio > split_threshold:
             return
 
         result, *others = split_pic(pp.image, inverse_xy, width, height, overlap_ratio)