Merge pull request #10780 from akx/image-emb-fonts

Mark caption_image_overlay's textfont as deprecated; fix #10778

CHANGELOG.md

@@ -1,4 +1,24 @@
-## Upcoming 1.3.0
+## 1.3.2
+
+### Bug Fixes:
+ * fix files served out of tmp directory even if they are saved to disk
+ * fix postprocessing overwriting parameters
+
+## 1.3.1
+
+### Features:
+ * revert default cross attention optimization to Doggettx
+
+### Bug Fixes:
+ * fix bug: LoRA don't apply on dropdown list sd_lora
+ * fix png info always added even if setting is not enabled
+ * fix some fields not applying in xyz plot
+ * fix "hires. fix" prompt sharing same labels with txt2img_prompt
+ * fix lora hashes not being added properly to infotex if there is only one lora
+ * fix --use-cpu failing to work properly at startup
+ * make --disable-opt-split-attention command line option work again
+
+## 1.3.0
 
 ### Features:
  * add UI to edit defaults

modules/cmd_args.py

@@ -62,7 +62,7 @@ parser.add_argument("--opt-split-attention-invokeai", action='store_true', help=
 parser.add_argument("--opt-split-attention-v1", action='store_true', help="prefer older version of split attention optimization for automatic choice of optimization")
 parser.add_argument("--opt-sdp-attention", action='store_true', help="prefer scaled dot product cross-attention layer optimization for automatic choice of optimization; requires PyTorch 2.*")
 parser.add_argument("--opt-sdp-no-mem-attention", action='store_true', help="prefer scaled dot product cross-attention layer optimization without memory efficient attention for automatic choice of optimization, makes image generation deterministic; requires PyTorch 2.*")
-parser.add_argument("--disable-opt-split-attention", action='store_true', help="does not do anything")
+parser.add_argument("--disable-opt-split-attention", action='store_true', help="prefer no cross-attention layer optimization for automatic choice of optimization")
 parser.add_argument("--disable-nan-check", action='store_true', help="do not check if produced images/latent spaces have nans; useful for running without a checkpoint in CI")
 parser.add_argument("--use-cpu", nargs='+', help="use CPU as torch device for specified modules", default=[], type=str.lower)
 parser.add_argument("--listen", action='store_true', help="launch gradio with 0.0.0.0 as server name, allowing to respond to network requests")

modules/extra_networks.py

@@ -26,7 +26,7 @@ class ExtraNetworkParams:
         self.named = {}
 
         for item in self.items:
-            parts = item.split('=', 2)
+            parts = item.split('=', 2) if isinstance(item, str) else [item]
             if len(parts) == 2:
                 self.named[parts[0]] = parts[1]
             else:

modules/generation_parameters_copypaste.py

@@ -35,7 +35,7 @@ def reset():
 
 
 def quote(text):
-    if ',' not in str(text) and '\n' not in str(text):
+    if ',' not in str(text) and '\n' not in str(text) and ':' not in str(text):
         return text
 
     return json.dumps(text, ensure_ascii=False)

modules/images.py

@@ -488,14 +488,13 @@ def save_image_with_geninfo(image, geninfo, filename, extension=None, existing_p
 
     image_format = Image.registered_extensions()[extension]
 
-    existing_pnginfo = existing_pnginfo or {}
-    if opts.enable_pnginfo:
-        existing_pnginfo['parameters'] = geninfo
-
     if extension.lower() == '.png':
-        pnginfo_data = PngImagePlugin.PngInfo()
-        for k, v in (existing_pnginfo or {}).items():
-            pnginfo_data.add_text(k, str(v))
+        if opts.enable_pnginfo:
+            pnginfo_data = PngImagePlugin.PngInfo()
+            for k, v in (existing_pnginfo or {}).items():
+                pnginfo_data.add_text(k, str(v))
+        else:
+            pnginfo_data = None
 
         image.save(filename, format=image_format, quality=opts.jpeg_quality, pnginfo=pnginfo_data)
 

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/sd_hijack.py

@@ -48,6 +48,11 @@ def apply_optimizations():
 
     undo_optimizations()
 
+    if len(optimizers) == 0:
+        # a script can access the model very early, and optimizations would not be filled by then
+        current_optimizer = None
+        return ''
+
     ldm.modules.diffusionmodules.model.nonlinearity = silu
     ldm.modules.diffusionmodules.openaimodel.th = sd_hijack_unet.th
 
@@ -63,12 +68,15 @@ def apply_optimizations():
 
     if selection == "None":
         matching_optimizer = None
+    elif selection == "Automatic" and shared.cmd_opts.disable_opt_split_attention:
+        matching_optimizer = None
     elif matching_optimizer is None:
         matching_optimizer = optimizers[0]
 
     if matching_optimizer is not None:
-        print(f"Applying optimization: {matching_optimizer.name}")
+        print(f"Applying optimization: {matching_optimizer.name}... ", end='')
         matching_optimizer.apply()
+        print("done.")
         current_optimizer = matching_optimizer
         return current_optimizer.name
     else:
@@ -149,6 +157,13 @@ class StableDiffusionModelHijack:
     def __init__(self):
         self.embedding_db.add_embedding_dir(cmd_opts.embeddings_dir)
 
+    def apply_optimizations(self):
+        try:
+            self.optimization_method = apply_optimizations()
+        except Exception as e:
+            errors.display(e, "applying cross attention optimization")
+            undo_optimizations()
+
     def hijack(self, m):
         if type(m.cond_stage_model) == xlmr.BertSeriesModelWithTransformation:
             model_embeddings = m.cond_stage_model.roberta.embeddings
@@ -168,11 +183,7 @@ class StableDiffusionModelHijack:
         if m.cond_stage_key == "edit":
             sd_hijack_unet.hijack_ddpm_edit()
 
-        try:
-            self.optimization_method = apply_optimizations()
-        except Exception as e:
-            errors.display(e, "applying cross attention optimization")
-            undo_optimizations()
+        self.apply_optimizations()
 
         self.clip = m.cond_stage_model
 

modules/sd_hijack_optimizations.py

@@ -59,7 +59,7 @@ class SdOptimizationSdpNoMem(SdOptimization):
     name = "sdp-no-mem"
     label = "scaled dot product without memory efficient attention"
     cmd_opt = "opt_sdp_no_mem_attention"
-    priority = 90
+    priority = 80
 
     def is_available(self):
         return hasattr(torch.nn.functional, "scaled_dot_product_attention") and callable(torch.nn.functional.scaled_dot_product_attention)
@@ -73,7 +73,7 @@ class SdOptimizationSdp(SdOptimizationSdpNoMem):
     name = "sdp"
     label = "scaled dot product"
     cmd_opt = "opt_sdp_attention"
-    priority = 80
+    priority = 70
 
     def apply(self):
         ldm.modules.attention.CrossAttention.forward = scaled_dot_product_attention_forward
@@ -116,7 +116,7 @@ class SdOptimizationInvokeAI(SdOptimization):
 class SdOptimizationDoggettx(SdOptimization):
     name = "Doggettx"
     cmd_opt = "opt_split_attention"
-    priority = 20
+    priority = 90
 
     def apply(self):
         ldm.modules.attention.CrossAttention.forward = split_cross_attention_forward

modules/sd_models.py

@@ -313,8 +313,6 @@ def load_model_weights(model, checkpoint_info: CheckpointInfo, state_dict, timer
 
         timer.record("apply half()")
 
-    devices.dtype = torch.float32 if shared.cmd_opts.no_half else torch.float16
-    devices.dtype_vae = torch.float32 if shared.cmd_opts.no_half or shared.cmd_opts.no_half_vae else torch.float16
     devices.dtype_unet = model.model.diffusion_model.dtype
     devices.unet_needs_upcast = shared.cmd_opts.upcast_sampling and devices.dtype == torch.float16 and devices.dtype_unet == torch.float16
 

modules/shared.py

@@ -6,6 +6,7 @@ import threading
 import time
 
 import gradio as gr
+import torch
 import tqdm
 
 import modules.interrogate
@@ -76,6 +77,9 @@ cmd_opts.disable_extension_access = (cmd_opts.share or cmd_opts.listen or cmd_op
 devices.device, devices.device_interrogate, devices.device_gfpgan, devices.device_esrgan, devices.device_codeformer = \
     (devices.cpu if any(y in cmd_opts.use_cpu for y in [x, 'all']) else devices.get_optimal_device() for x in ['sd', 'interrogate', 'gfpgan', 'esrgan', 'codeformer'])
 
+devices.dtype = torch.float32 if cmd_opts.no_half else torch.float16
+devices.dtype_vae = torch.float32 if cmd_opts.no_half or cmd_opts.no_half_vae else torch.float16
+
 device = devices.device
 weight_load_location = None if cmd_opts.lowram else "cpu"
 

modules/textual_inversion/image_embedding.py

@@ -1,8 +1,10 @@
 import base64
 import json
+import warnings
+
 import numpy as np
 import zlib
-from PIL import Image, ImageDraw, ImageFont
+from PIL import Image, ImageDraw
 import torch
 
 
@@ -129,14 +131,17 @@ def extract_image_data_embed(image):
 
 
 def caption_image_overlay(srcimage, title, footerLeft, footerMid, footerRight, textfont=None):
+    from modules.images import get_font
+    if textfont:
+        warnings.warn(
+            'passing in a textfont to caption_image_overlay is deprecated and does nothing',
+            DeprecationWarning,
+            stacklevel=2,
+        )
     from math import cos
 
     image = srcimage.copy()
     fontsize = 32
-    if textfont is None:
-        from modules.images import get_font
-        textfont = get_font(fontsize)
-
     factor = 1.5
     gradient = Image.new('RGBA', (1, image.size[1]), color=(0, 0, 0, 0))
     for y in range(image.size[1]):
@@ -147,12 +152,12 @@ def caption_image_overlay(srcimage, title, footerLeft, footerMid, footerRight, t
 
     draw = ImageDraw.Draw(image)
 
-    font = ImageFont.truetype(textfont, fontsize)
+    font = get_font(fontsize)
     padding = 10
 
     _, _, w, h = draw.textbbox((0, 0), title, font=font)
     fontsize = min(int(fontsize * (((image.size[0]*0.75)-(padding*4))/w)), 72)
-    font = ImageFont.truetype(textfont, fontsize)
+    font = get_font(fontsize)
     _, _, w, h = draw.textbbox((0, 0), title, font=font)
     draw.text((padding, padding), title, anchor='lt', font=font, fill=(255, 255, 255, 230))
 
@@ -163,7 +168,7 @@ def caption_image_overlay(srcimage, title, footerLeft, footerMid, footerRight, t
     _, _, w, h = draw.textbbox((0, 0), footerRight, font=font)
     fontsize_right = min(int(fontsize * (((image.size[0]/3)-(padding))/w)), 72)
 
-    font = ImageFont.truetype(textfont, min(fontsize_left, fontsize_mid, fontsize_right))
+    font = get_font(min(fontsize_left, fontsize_mid, fontsize_right))
 
     draw.text((padding, image.size[1]-padding),               footerLeft, anchor='ls', font=font, fill=(255, 255, 255, 230))
     draw.text((image.size[0]/2, image.size[1]-padding),       footerMid, anchor='ms', font=font, fill=(255, 255, 255, 230))

modules/ui.py

@@ -505,10 +505,10 @@ 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_tempdir.py

@@ -3,7 +3,7 @@ import tempfile
 from collections import namedtuple
 from pathlib import Path
 
-import gradio as gr
+import gradio.components
 
 from PIL import PngImagePlugin
 
@@ -31,13 +31,16 @@ def check_tmp_file(gradio, filename):
     return False
 
 
-def save_pil_to_file(pil_image, dir=None):
+def save_pil_to_file(self, pil_image, dir=None):
     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)
+        filename = already_saved_as
 
-        file_obj = Savedfile(f'{already_saved_as}?{os.path.getmtime(already_saved_as)}')
-        return file_obj
+        if not shared.opts.save_images_add_number:
+            filename += f'?{os.path.getmtime(already_saved_as)}'
+
+        return filename
 
     if shared.opts.temp_dir != "":
         dir = shared.opts.temp_dir
@@ -51,11 +54,11 @@ def save_pil_to_file(pil_image, dir=None):
 
     file_obj = tempfile.NamedTemporaryFile(delete=False, suffix=".png", dir=dir)
     pil_image.save(file_obj, pnginfo=(metadata if use_metadata else None))
-    return file_obj
+    return file_obj.name
 
 
 # override save to file function so that it also writes PNG info
-gr.processing_utils.save_pil_to_file = save_pil_to_file
+gradio.components.IOComponent.pil_to_temp_file = save_pil_to_file
 
 
 def on_tmpdir_changed():

requirements_versions.txt

@@ -3,7 +3,7 @@ transformers==4.25.1
 accelerate==0.18.0
 basicsr==1.4.2
 gfpgan==1.3.8
-gradio==3.31.0
+gradio==3.32.0
 numpy==1.23.5
 Pillow==9.5.0
 realesrgan==0.3.0

style.css

@@ -756,13 +756,22 @@ footer {
 .extra-network-cards .card .metadata-button, .extra-network-thumbs .card .metadata-button{
     display: none;
     position: absolute;
-    right: 0;
     color: white;
+    right: 0;
+}
+.extra-network-cards .card .metadata-button {
     text-shadow: 2px 2px 3px black;
     padding: 0.25em;
     font-size: 22pt;
     width: 1.5em;
 }
+.extra-network-thumbs .card .metadata-button {
+    text-shadow: 1px 1px 2px black;
+    padding: 0;
+    font-size: 16pt;
+    width: 1em;
+    top: -0.25em;
+}
 .extra-network-cards .card:hover .metadata-button, .extra-network-thumbs .card:hover .metadata-button{
     display: inline-block;
 }
@@ -787,6 +796,13 @@ footer {
     position: relative;
 }
 
+.extra-network-thumbs .card .preview{
+    position: absolute;
+    object-fit: cover;
+    width: 100%;
+    height:100%;
+}
+
 .extra-network-thumbs .card:hover .additional a {
     display: inline-block;
 }

webui.py

@@ -291,9 +291,20 @@ def initialize_rest(*, reload_script_modules=False):
     modules.sd_hijack.list_optimizers()
     startup_timer.record("scripts list_optimizers")
 
-    # load model in parallel to other startup stuff
-    # (when reloading, this does nothing)
-    Thread(target=lambda: shared.sd_model).start()
+    def load_model():
+        """
+        Accesses shared.sd_model property to load model.
+        After it's available, if it has been loaded before this access by some extension,
+        its optimization may be None because the list of optimizaers has neet been filled
+        by that time, so we apply optimization again.
+        """
+
+        shared.sd_model  # noqa: B018
+
+        if modules.sd_hijack.current_optimizer is None:
+            modules.sd_hijack.apply_optimizations()
+
+    Thread(target=load_model).start()
 
     shared.reload_hypernetworks()
     startup_timer.record("reload hypernetworks")