Merge branch 'dev' into report-error

extensions-builtin/LDSR/sd_hijack_autoencoder.py

@@ -10,7 +10,7 @@ from contextlib import contextmanager
 from torch.optim.lr_scheduler import LambdaLR
 
 from ldm.modules.ema import LitEma
-from taming.modules.vqvae.quantize import VectorQuantizer2 as VectorQuantizer
+from vqvae_quantize import VectorQuantizer2 as VectorQuantizer
 from ldm.modules.diffusionmodules.model import Encoder, Decoder
 from ldm.util import instantiate_from_config
 

extensions-builtin/LDSR/vqvae_quantize.py

+# Vendored from https://raw.githubusercontent.com/CompVis/taming-transformers/24268930bf1dce879235a7fddd0b2355b84d7ea6/taming/modules/vqvae/quantize.py,
+# where the license is as follows:
+#
+# Copyright (c) 2020 Patrick Esser and Robin Rombach and Björn Ommer
+#
+# Permission is hereby granted, free of charge, to any person obtaining a copy
+# of this software and associated documentation files (the "Software"), to deal
+# in the Software without restriction, including without limitation the rights
+# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+# copies of the Software, and to permit persons to whom the Software is
+# furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in all
+# copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+# EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+# MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+# IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM,
+# DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+# OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE
+# OR OTHER DEALINGS IN THE SOFTWARE./
+
+import torch
+import torch.nn as nn
+import numpy as np
+from einops import rearrange
+
+
+class VectorQuantizer2(nn.Module):
+    """
+    Improved version over VectorQuantizer, can be used as a drop-in replacement. Mostly
+    avoids costly matrix multiplications and allows for post-hoc remapping of indices.
+    """
+
+    # NOTE: due to a bug the beta term was applied to the wrong term. for
+    # backwards compatibility we use the buggy version by default, but you can
+    # specify legacy=False to fix it.
+    def __init__(self, n_e, e_dim, beta, remap=None, unknown_index="random",
+                 sane_index_shape=False, legacy=True):
+        super().__init__()
+        self.n_e = n_e
+        self.e_dim = e_dim
+        self.beta = beta
+        self.legacy = legacy
+
+        self.embedding = nn.Embedding(self.n_e, self.e_dim)
+        self.embedding.weight.data.uniform_(-1.0 / self.n_e, 1.0 / self.n_e)
+
+        self.remap = remap
+        if self.remap is not None:
+            self.register_buffer("used", torch.tensor(np.load(self.remap)))
+            self.re_embed = self.used.shape[0]
+            self.unknown_index = unknown_index  # "random" or "extra" or integer
+            if self.unknown_index == "extra":
+                self.unknown_index = self.re_embed
+                self.re_embed = self.re_embed + 1
+            print(f"Remapping {self.n_e} indices to {self.re_embed} indices. "
+                  f"Using {self.unknown_index} for unknown indices.")
+        else:
+            self.re_embed = n_e
+
+        self.sane_index_shape = sane_index_shape
+
+    def remap_to_used(self, inds):
+        ishape = inds.shape
+        assert len(ishape) > 1
+        inds = inds.reshape(ishape[0], -1)
+        used = self.used.to(inds)
+        match = (inds[:, :, None] == used[None, None, ...]).long()
+        new = match.argmax(-1)
+        unknown = match.sum(2) < 1
+        if self.unknown_index == "random":
+            new[unknown] = torch.randint(0, self.re_embed, size=new[unknown].shape).to(device=new.device)
+        else:
+            new[unknown] = self.unknown_index
+        return new.reshape(ishape)
+
+    def unmap_to_all(self, inds):
+        ishape = inds.shape
+        assert len(ishape) > 1
+        inds = inds.reshape(ishape[0], -1)
+        used = self.used.to(inds)
+        if self.re_embed > self.used.shape[0]:  # extra token
+            inds[inds >= self.used.shape[0]] = 0  # simply set to zero
+        back = torch.gather(used[None, :][inds.shape[0] * [0], :], 1, inds)
+        return back.reshape(ishape)
+
+    def forward(self, z, temp=None, rescale_logits=False, return_logits=False):
+        assert temp is None or temp == 1.0, "Only for interface compatible with Gumbel"
+        assert rescale_logits is False, "Only for interface compatible with Gumbel"
+        assert return_logits is False, "Only for interface compatible with Gumbel"
+        # reshape z -> (batch, height, width, channel) and flatten
+        z = rearrange(z, 'b c h w -> b h w c').contiguous()
+        z_flattened = z.view(-1, self.e_dim)
+        # distances from z to embeddings e_j (z - e)^2 = z^2 + e^2 - 2 e * z
+
+        d = torch.sum(z_flattened ** 2, dim=1, keepdim=True) + \
+            torch.sum(self.embedding.weight ** 2, dim=1) - 2 * \
+            torch.einsum('bd,dn->bn', z_flattened, rearrange(self.embedding.weight, 'n d -> d n'))
+
+        min_encoding_indices = torch.argmin(d, dim=1)
+        z_q = self.embedding(min_encoding_indices).view(z.shape)
+        perplexity = None
+        min_encodings = None
+
+        # compute loss for embedding
+        if not self.legacy:
+            loss = self.beta * torch.mean((z_q.detach() - z) ** 2) + \
+                   torch.mean((z_q - z.detach()) ** 2)
+        else:
+            loss = torch.mean((z_q.detach() - z) ** 2) + self.beta * \
+                   torch.mean((z_q - z.detach()) ** 2)
+
+        # preserve gradients
+        z_q = z + (z_q - z).detach()
+
+        # reshape back to match original input shape
+        z_q = rearrange(z_q, 'b h w c -> b c h w').contiguous()
+
+        if self.remap is not None:
+            min_encoding_indices = min_encoding_indices.reshape(z.shape[0], -1)  # add batch axis
+            min_encoding_indices = self.remap_to_used(min_encoding_indices)
+            min_encoding_indices = min_encoding_indices.reshape(-1, 1)  # flatten
+
+        if self.sane_index_shape:
+            min_encoding_indices = min_encoding_indices.reshape(
+                z_q.shape[0], z_q.shape[2], z_q.shape[3])
+
+        return z_q, loss, (perplexity, min_encodings, min_encoding_indices)
+
+    def get_codebook_entry(self, indices, shape):
+        # shape specifying (batch, height, width, channel)
+        if self.remap is not None:
+            indices = indices.reshape(shape[0], -1)  # add batch axis
+            indices = self.unmap_to_all(indices)
+            indices = indices.reshape(-1)  # flatten again
+
+        # get quantized latent vectors
+        z_q = self.embedding(indices)
+
+        if shape is not None:
+            z_q = z_q.view(shape)
+            # reshape back to match original input shape
+            z_q = z_q.permute(0, 3, 1, 2).contiguous()
+
+        return z_q

javascript/imageviewerGamepad.js

+let gamepads = [];
+
 window.addEventListener('gamepadconnected', (e) => {
     const index = e.gamepad.index;
     let isWaiting = false;
-    setInterval(async() => {
+    gamepads[index] = setInterval(async() => {
         if (!opts.js_modal_lightbox_gamepad || isWaiting) return;
         const gamepad = navigator.getGamepads()[index];
         const xValue = gamepad.axes[0];
@@ -24,6 +26,10 @@ window.addEventListener('gamepadconnected', (e) => {
     }, 10);
 });
 
+window.addEventListener('gamepaddisconnected', (e) => {
+    clearInterval(gamepads[e.gamepad.index]);
+});
+
 /*
 Primarily for vr controller type pointer devices.
 I use the wheel event because there's currently no way to do it properly with web xr.

modules/api/api.py

@@ -24,6 +24,7 @@ from modules.textual_inversion.preprocess import preprocess
 from modules.hypernetworks.hypernetwork import create_hypernetwork, train_hypernetwork
 from PIL import PngImagePlugin,Image
 from modules.sd_models import checkpoints_list, unload_model_weights, reload_model_weights
+from modules.sd_vae import vae_dict
 from modules.sd_models_config import find_checkpoint_config_near_filename
 from modules.realesrgan_model import get_realesrgan_models
 from modules import devices
@@ -190,6 +191,7 @@ class Api:
         self.add_api_route("/sdapi/v1/samplers", self.get_samplers, methods=["GET"], response_model=List[models.SamplerItem])
         self.add_api_route("/sdapi/v1/upscalers", self.get_upscalers, methods=["GET"], response_model=List[models.UpscalerItem])
         self.add_api_route("/sdapi/v1/sd-models", self.get_sd_models, methods=["GET"], response_model=List[models.SDModelItem])
+        self.add_api_route("/sdapi/v1/sd-vae", self.get_sd_vaes, methods=["GET"], response_model=List[models.SDVaeItem])
         self.add_api_route("/sdapi/v1/hypernetworks", self.get_hypernetworks, methods=["GET"], response_model=List[models.HypernetworkItem])
         self.add_api_route("/sdapi/v1/face-restorers", self.get_face_restorers, methods=["GET"], response_model=List[models.FaceRestorerItem])
         self.add_api_route("/sdapi/v1/realesrgan-models", self.get_realesrgan_models, methods=["GET"], response_model=List[models.RealesrganItem])
@@ -542,6 +544,9 @@ class Api:
     def get_sd_models(self):
         return [{"title": x.title, "model_name": x.model_name, "hash": x.shorthash, "sha256": x.sha256, "filename": x.filename, "config": find_checkpoint_config_near_filename(x)} for x in checkpoints_list.values()]
 
+    def get_sd_vaes(self):
+        return [{"model_name": x, "filename": vae_dict[x]} for x in vae_dict.keys()]
+
     def get_hypernetworks(self):
         return [{"name": name, "path": shared.hypernetworks[name]} for name in shared.hypernetworks]
 

modules/api/models.py

@@ -249,6 +249,10 @@ class SDModelItem(BaseModel):
     filename: str = Field(title="Filename")
     config: Optional[str] = Field(title="Config file")
 
+class SDVaeItem(BaseModel):
+    model_name: str = Field(title="Model Name")
+    filename: str = Field(title="Filename")
+
 class HypernetworkItem(BaseModel):
     name: str = Field(title="Name")
     path: Optional[str] = Field(title="Path")

modules/cmd_args.py

@@ -11,7 +11,7 @@ parser.add_argument("--skip-python-version-check", action='store_true', help="la
 parser.add_argument("--skip-torch-cuda-test", action='store_true', help="launch.py argument: do not check if CUDA is able to work properly")
 parser.add_argument("--reinstall-xformers", action='store_true', help="launch.py argument: install the appropriate version of xformers even if you have some version already installed")
 parser.add_argument("--reinstall-torch", action='store_true', help="launch.py argument: install the appropriate version of torch even if you have some version already installed")
-parser.add_argument("--update-check", action='store_true', help="launch.py argument: chck for updates at startup")
+parser.add_argument("--update-check", action='store_true', help="launch.py argument: check for updates at startup")
 parser.add_argument("--test-server", action='store_true', help="launch.py argument: configure server for testing")
 parser.add_argument("--skip-prepare-environment", action='store_true', help="launch.py argument: skip all environment preparation")
 parser.add_argument("--skip-install", action='store_true', help="launch.py argument: skip installation of packages")

modules/extensions.py

 import os
 import threading
 
-import git
-
 from modules import shared
 from modules.errors import print_error
+from modules.gitpython_hack import Repo
 from modules.paths_internal import extensions_dir, extensions_builtin_dir, script_path  # noqa: F401
 
 extensions = []
@@ -53,7 +52,7 @@ class Extension:
         repo = None
         try:
             if os.path.exists(os.path.join(self.path, ".git")):
-                repo = git.Repo(self.path)
+                repo = Repo(self.path)
         except Exception:
             print_error(f"Error reading github repository info from {self.path}", exc_info=True)
 
@@ -92,7 +91,7 @@ class Extension:
         return res
 
     def check_updates(self):
-        repo = git.Repo(self.path)
+        repo = Repo(self.path)
         for fetch in repo.remote().fetch(dry_run=True):
             if fetch.flags != fetch.HEAD_UPTODATE:
                 self.can_update = True
@@ -114,7 +113,7 @@ class Extension:
         self.status = "latest"
 
     def fetch_and_reset_hard(self, commit='origin'):
-        repo = git.Repo(self.path)
+        repo = Repo(self.path)
         # Fix: `error: Your local changes to the following files would be overwritten by merge`,
         # because WSL2 Docker set 755 file permissions instead of 644, this results to the error.
         repo.git.fetch(all=True)

modules/gitpython_hack.py

+from __future__ import annotations
+
+import io
+import subprocess
+
+import git
+
+
+class Git(git.Git):
+    """
+    Git subclassed to never use persistent processes.
+    """
+
+    def _get_persistent_cmd(self, attr_name, cmd_name, *args, **kwargs):
+        raise NotImplementedError(f"Refusing to use persistent process: {attr_name} ({cmd_name} {args} {kwargs})")
+
+    def get_object_header(self, ref: str | bytes) -> tuple[str, str, int]:
+        ret = subprocess.check_output(
+            [self.GIT_PYTHON_GIT_EXECUTABLE, "cat-file", "--batch-check"],
+            input=self._prepare_ref(ref),
+            cwd=self._working_dir,
+            timeout=2,
+        )
+        return self._parse_object_header(ret)
+
+    def stream_object_data(self, ref: str) -> tuple[str, str, int, "Git.CatFileContentStream"]:
+        # Not really streaming, per se; this buffers the entire object in memory.
+        # Shouldn't be a problem for our use case, since we're only using this for
+        # object headers (commit objects).
+        ret = subprocess.check_output(
+            [self.GIT_PYTHON_GIT_EXECUTABLE, "cat-file", "--batch"],
+            input=self._prepare_ref(ref),
+            cwd=self._working_dir,
+            timeout=30,
+        )
+        bio = io.BytesIO(ret)
+        hexsha, typename, size = self._parse_object_header(bio.readline())
+        return (hexsha, typename, size, self.CatFileContentStream(size, bio))
+
+
+class Repo(git.Repo):
+    GitCommandWrapperType = Git

modules/launch_utils.py

@@ -230,13 +230,11 @@ def prepare_environment():
     openclip_package = os.environ.get('OPENCLIP_PACKAGE', "https://github.com/mlfoundations/open_clip/archive/bb6e834e9c70d9c27d0dc3ecedeebeaeb1ffad6b.zip")
 
     stable_diffusion_repo = os.environ.get('STABLE_DIFFUSION_REPO', "https://github.com/Stability-AI/stablediffusion.git")
-    taming_transformers_repo = os.environ.get('TAMING_TRANSFORMERS_REPO', "https://github.com/CompVis/taming-transformers.git")
     k_diffusion_repo = os.environ.get('K_DIFFUSION_REPO', 'https://github.com/crowsonkb/k-diffusion.git')
     codeformer_repo = os.environ.get('CODEFORMER_REPO', 'https://github.com/sczhou/CodeFormer.git')
     blip_repo = os.environ.get('BLIP_REPO', 'https://github.com/salesforce/BLIP.git')
 
     stable_diffusion_commit_hash = os.environ.get('STABLE_DIFFUSION_COMMIT_HASH', "cf1d67a6fd5ea1aa600c4df58e5b47da45f6bdbf")
-    taming_transformers_commit_hash = os.environ.get('TAMING_TRANSFORMERS_COMMIT_HASH', "24268930bf1dce879235a7fddd0b2355b84d7ea6")
     k_diffusion_commit_hash = os.environ.get('K_DIFFUSION_COMMIT_HASH', "c9fe758757e022f05ca5a53fa8fac28889e4f1cf")
     codeformer_commit_hash = os.environ.get('CODEFORMER_COMMIT_HASH', "c5b4593074ba6214284d6acd5f1719b6c5d739af")
     blip_commit_hash = os.environ.get('BLIP_COMMIT_HASH', "48211a1594f1321b00f14c9f7a5b4813144b2fb9")
@@ -287,7 +285,6 @@ def prepare_environment():
     os.makedirs(os.path.join(script_path, dir_repos), exist_ok=True)
 
     git_clone(stable_diffusion_repo, repo_dir('stable-diffusion-stability-ai'), "Stable Diffusion", stable_diffusion_commit_hash)
-    git_clone(taming_transformers_repo, repo_dir('taming-transformers'), "Taming Transformers", taming_transformers_commit_hash)
     git_clone(k_diffusion_repo, repo_dir('k-diffusion'), "K-diffusion", k_diffusion_commit_hash)
     git_clone(codeformer_repo, repo_dir('CodeFormer'), "CodeFormer", codeformer_commit_hash)
     git_clone(blip_repo, repo_dir('BLIP'), "BLIP", blip_commit_hash)

modules/paths.py

@@ -20,7 +20,6 @@ assert sd_path is not None, f"Couldn't find Stable Diffusion in any of: {possibl
 
 path_dirs = [
     (sd_path, 'ldm', 'Stable Diffusion', []),
-    (os.path.join(sd_path, '../taming-transformers'), 'taming', 'Taming Transformers', []),
     (os.path.join(sd_path, '../CodeFormer'), 'inference_codeformer.py', 'CodeFormer', []),
     (os.path.join(sd_path, '../BLIP'), 'models/blip.py', 'BLIP', []),
     (os.path.join(sd_path, '../k-diffusion'), 'k_diffusion/sampling.py', 'k_diffusion', ["atstart"]),

modules/processing.py

@@ -321,14 +321,13 @@ class StableDiffusionProcessing:
         have been used before. The second element is where the previously
         computed result is stored.
         """
-
-        if cache[0] is not None and (required_prompts, steps) == cache[0]:
+        if cache[0] is not None and (required_prompts, steps, opts.CLIP_stop_at_last_layers, shared.sd_model.sd_checkpoint_info) == cache[0]:
             return cache[1]
 
         with devices.autocast():
             cache[1] = function(shared.sd_model, required_prompts, steps)
 
-        cache[0] = (required_prompts, steps)
+        cache[0] = (required_prompts, steps, opts.CLIP_stop_at_last_layers, shared.sd_model.sd_checkpoint_info)
         return cache[1]
 
     def setup_conds(self):

modules/shared.py

@@ -416,12 +416,12 @@ options_templates.update(options_section(('sd', "Stable Diffusion"), {
     "comma_padding_backtrack": OptionInfo(20, "Prompt word wrap length limit", gr.Slider, {"minimum": 0, "maximum": 74, "step": 1}).info("in tokens - for texts shorter than specified, if they don't fit into 75 token limit, move them to the next 75 token chunk"),
     "CLIP_stop_at_last_layers": OptionInfo(1, "Clip skip", gr.Slider, {"minimum": 1, "maximum": 12, "step": 1}).link("wiki", "https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#clip-skip").info("ignore last layers of CLIP nrtwork; 1 ignores none, 2 ignores one layer"),
     "upcast_attn": OptionInfo(False, "Upcast cross attention layer to float32"),
-    "randn_source": OptionInfo("GPU", "Random number generator source.", gr.Radio, {"choices": ["GPU", "CPU"]}).info("changes seeds drastically; use CPU to produce the same picture across different vidocard vendors"),
+    "randn_source": OptionInfo("GPU", "Random number generator source.", gr.Radio, {"choices": ["GPU", "CPU"]}).info("changes seeds drastically; use CPU to produce the same picture across different videocard vendors"),
 }))
 
 options_templates.update(options_section(('optimizations', "Optimizations"), {
     "cross_attention_optimization": OptionInfo("Automatic", "Cross attention optimization", gr.Dropdown, lambda: {"choices": shared_items.cross_attention_optimizations()}),
-    "s_min_uncond": OptionInfo(0, "Negative Guidance minimum sigma", gr.Slider, {"minimum": 0.0, "maximum": 4.0, "step": 0.01}).link("PR", "https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/9177").info("skip negative prompt for some steps when the image is almost ready; 0=disable, higher=faster"),
+    "s_min_uncond": OptionInfo(0.0, "Negative Guidance minimum sigma", gr.Slider, {"minimum": 0.0, "maximum": 4.0, "step": 0.01}).link("PR", "https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/9177").info("skip negative prompt for some steps when the image is almost ready; 0=disable, higher=faster"),
     "token_merging_ratio": OptionInfo(0.0, "Token merging ratio", gr.Slider, {"minimum": 0.0, "maximum": 0.9, "step": 0.1}).link("PR", "https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/9256").info("0=disable, higher=faster"),
     "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}).info("only applies if non-zero and overrides above"),

modules/ui.py

@@ -504,10 +504,10 @@ def create_ui():
                             with FormRow(elem_id="txt2img_hires_fix_row4", variant="compact", visible=opts.hires_fix_show_prompts) as hr_prompts_container:
                                 with gr.Column(scale=80):
                                     with gr.Row():
-                                        hr_prompt = gr.Textbox(label="Prompt", elem_id="hires_prompt", show_label=False, lines=3, placeholder="Prompt for hires fix pass.\nLeave empty to use the same prompt as in first pass.", elem_classes=["prompt"])
+                                        hr_prompt = gr.Textbox(label="Hires prompt", elem_id="hires_prompt", show_label=False, lines=3, placeholder="Prompt for hires fix pass.\nLeave empty to use the same prompt as in first pass.", elem_classes=["prompt"])
                                 with gr.Column(scale=80):
                                     with gr.Row():
-                                        hr_negative_prompt = gr.Textbox(label="Negative prompt", elem_id="hires_neg_prompt", show_label=False, lines=3, placeholder="Negative prompt for hires fix pass.\nLeave empty to use the same negative prompt as in first pass.", elem_classes=["prompt"])
+                                        hr_negative_prompt = gr.Textbox(label="Hires negative prompt", elem_id="hires_neg_prompt", show_label=False, lines=3, placeholder="Negative prompt for hires fix pass.\nLeave empty to use the same negative prompt as in first pass.", elem_classes=["prompt"])
 
                     elif category == "batch":
                         if not opts.dimensions_and_batch_together:

modules/ui_extensions.py

@@ -487,8 +487,14 @@ def refresh_available_extensions_from_data(hide_tags, sort_column, filter_text="
 
 
 def preload_extensions_git_metadata():
+    t0 = time.time()
     for extension in extensions.extensions:
         extension.read_info_from_repo()
+    print(
+        f"preload_extensions_git_metadata for "
+        f"{len(extensions.extensions)} extensions took "
+        f"{time.time() - t0:.2f}s"
+    )
 
 
 def create_ui():

modules/upscaler.py

@@ -53,8 +53,8 @@ class Upscaler:
 
     def upscale(self, img: PIL.Image, scale, selected_model: str = None):
         self.scale = scale
-        dest_w = int(img.width * scale)
-        dest_h = int(img.height * scale)
+        dest_w = round((img.width * scale - 4) / 8) * 8
+        dest_h = round((img.height * scale - 4) / 8) * 8
 
         for _ in range(3):
             shape = (img.width, img.height)

webui-user.sh

@@ -36,7 +36,6 @@
 
 # Fixed git commits
 #export STABLE_DIFFUSION_COMMIT_HASH=""
-#export TAMING_TRANSFORMERS_COMMIT_HASH=""
 #export CODEFORMER_COMMIT_HASH=""
 #export BLIP_COMMIT_HASH=""