Upload 17 files

README.md

@@ -4,7 +4,7 @@ emoji: 🧩🖼️📦
 colorFrom: red
 colorTo: pink
 sdk: gradio
-sdk_version: 4.43.0
+sdk_version: 5.44.1
 app_file: app.py
 pinned: true
 header: mini

app.py

@@ -1,5 +1,6 @@
 import spaces
 import os
+from argparse import ArgumentParser
 from stablepy import (
     Model_Diffusers,
     SCHEDULE_TYPE_OPTIONS,
@@ -23,6 +24,7 @@ from constants import (
     DIFFUSERS_CONTROLNET_MODEL,
     IP_MODELS,
     MODE_IP_OPTIONS,
+    CACHE_HF_ROOT,
 )
 from stablepy.diffusers_vanilla.style_prompt_config import STYLE_NAMES
 import torch
@@ -41,6 +43,7 @@ from utils import (
     progress_step_bar,
     html_template_message,
     escape_html,
+    clear_hf_cache,
 )
 from image_processor import preprocessor_tab
 from datetime import datetime
@@ -53,14 +56,17 @@ from diffusers import FluxPipeline
 # import urllib.parse
 import subprocess
 
-IS_ZEROGPU = os.getenv("SPACES_ZERO_GPU", False)
-if IS_ZEROGPU:
+IS_ZERO_GPU = bool(os.getenv("SPACES_ZERO_GPU"))
+if IS_ZERO_GPU:
     subprocess.run("rm -rf /data-nvme/zerogpu-offload/*", env={}, shell=True)
-    torch.backends.cuda.matmul.allow_tf32 = True
-ImageFile.LOAD_TRUNCATED_IMAGES = True
+IS_GPU_MODE = True if IS_ZERO_GPU else (True if torch.cuda.is_available() else False)
+img_path = "./images/"
+allowed_path = os.path.abspath(img_path)
+delete_cache_time = (9600, 9600) if IS_ZERO_GPU else (86400, 86400)
 
+ImageFile.LOAD_TRUNCATED_IMAGES = True
+torch.backends.cuda.matmul.allow_tf32 = True
 # os.environ["PYTORCH_NO_CUDA_MEMORY_CACHING"] = "1"
-print(os.getenv("SPACES_ZERO_GPU"))
 
 ## BEGIN MOD
 from modutils import (list_uniq, download_private_repo, get_model_id_list, get_tupled_embed_list,
@@ -94,14 +100,11 @@ DOWNLOAD_LORA = ", ".join(DOWNLOAD_LORA_LIST)
 
 # Download stuffs
 for url in [url.strip() for url in DOWNLOAD_MODEL.split(',')]:
-    if not os.path.exists(f"./models/{url.split('/')[-1]}"):
-        download_things(DIRECTORY_MODELS, url, HF_TOKEN, CIVITAI_API_KEY)
+    download_things(DIRECTORY_MODELS, url, HF_TOKEN, CIVITAI_API_KEY)
 for url in [url.strip() for url in DOWNLOAD_VAE.split(',')]:
-    if not os.path.exists(f"./vaes/{url.split('/')[-1]}"):
-        download_things(DIRECTORY_VAES, url, HF_TOKEN, CIVITAI_API_KEY)
+    download_things(DIRECTORY_VAES, url, HF_TOKEN, CIVITAI_API_KEY)
 for url in [url.strip() for url in DOWNLOAD_LORA.split(',')]:
-    if not os.path.exists(f"./loras/{url.split('/')[-1]}"):
-        download_things(DIRECTORY_LORAS, url, HF_TOKEN, CIVITAI_API_KEY)
+    download_things(DIRECTORY_LORAS, url, HF_TOKEN, CIVITAI_API_KEY)
 
 # Download Embeddings
 for url_embed in DOWNLOAD_EMBEDS:
@@ -128,15 +131,16 @@ def get_embed_list(pipeline_name):
 
 print('\033[33m🏁 Download and listing of valid models completed.\033[0m')
 
-flux_repo = "camenduru/FLUX.1-dev-diffusers"
-flux_pipe = FluxPipeline.from_pretrained(
-    flux_repo,
-    transformer=None,
-    torch_dtype=torch.bfloat16,
-)#.to("cuda")
-components = flux_pipe.components
-delete_model(flux_repo)
-# components = None
+components = None
+if IS_ZERO_GPU:
+    flux_repo = "camenduru/FLUX.1-dev-diffusers"
+    flux_pipe = FluxPipeline.from_pretrained(
+        flux_repo,
+        transformer=None,
+        torch_dtype=torch.bfloat16,
+    )#.to("cuda")
+    components = flux_pipe.components
+    delete_model(flux_repo)
 
 #######################
 # GUI
@@ -146,10 +150,19 @@ diffusers.utils.logging.set_verbosity(40)
 warnings.filterwarnings(action="ignore", category=FutureWarning, module="diffusers")
 warnings.filterwarnings(action="ignore", category=UserWarning, module="diffusers")
 warnings.filterwarnings(action="ignore", category=FutureWarning, module="transformers")
-## BEGIN MOD
-#logger.setLevel(logging.CRITICAL)
-logger.setLevel(logging.DEBUG)
 
+parser = ArgumentParser(description='DiffuseCraft: Create images from text prompts.', add_help=True)
+parser.add_argument("--share", action="store_true", dest="share_enabled", default=False, help="Enable sharing")
+parser.add_argument('--theme', type=str, default="NoCrypt/miku", help='Set the theme (default: NoCrypt/miku)')
+parser.add_argument("--ssr", action="store_true", default=False, help="Enable SSR (Server-Side Rendering)") #
+parser.add_argument("--log-level", type=str, default="INFO", choices=["DEBUG", "INFO", "WARNING", "ERROR", "CRITICAL"], help="Set logging level (default: INFO)")
+args = parser.parse_args()
+
+logger.setLevel(
+    "INFO" if IS_ZERO_GPU else getattr(logging, args.log_level.upper())
+)
+
+## BEGIN MOD
 from tagger.v2 import V2_ALL_MODELS, v2_random_prompt, v2_upsampling_prompt
 from tagger.utils import (gradio_copy_text, COPY_ACTION_JS, gradio_copy_prompt,
     V2_ASPECT_RATIO_OPTIONS, V2_RATING_OPTIONS, V2_LENGTH_OPTIONS, V2_IDENTITY_OPTIONS)
@@ -167,6 +180,11 @@ def description_ui():
     )
 ## END MOD
 
+def lora_chk(lora_):
+    if isinstance(lora_, str) and lora_.strip() not in ["", "None"]:
+        return lora_
+    return None
+
 class GuiSD:
     def __init__(self, stream=True):
         self.model = None
@@ -182,6 +200,15 @@ class GuiSD:
             removal_candidate = self.inventory.pop(0)
             delete_model(removal_candidate)
 
+        # Cleanup after 60 seconds of inactivity
+        lowPrioCleanup = max((datetime.now() - self.last_load).total_seconds(), 0) > 60
+        if lowPrioCleanup and not self.status_loading and get_used_storage_gb(CACHE_HF_ROOT) > (storage_floor_gb * 2):
+            print("Cleaning up Hugging Face cache...")
+            clear_hf_cache()
+            self.inventory = [
+                m for m in self.inventory if os.path.exists(m)
+            ]
+
     def update_inventory(self, model_name):
         if model_name not in single_file_model_list:
             self.inventory = [
@@ -192,19 +219,24 @@ class GuiSD:
     def load_new_model(self, model_name, vae_model, task, controlnet_model, progress=gr.Progress(track_tqdm=True)):
 
         # download link model > model_name
-        if "http" in model_name: #
+        if model_name.startswith("http"): #
+            yield f"Downloading model: {model_name}"
+            #model_name = download_things(DIRECTORY_MODELS, model_name, HF_TOKEN, CIVITAI_API_KEY)
             model_name, model_type = download_link_model(model_name, DIRECTORY_MODELS) #
+            if not model_name:
+                raise ValueError("Error retrieving model information from URL")
             is_link_model = True #
         else: is_link_model = False #
 
-        self.update_storage_models()
+        if IS_ZERO_GPU:
+            self.update_storage_models()
 
         vae_model = vae_model if vae_model != "None" else None
         model_type = get_model_type(model_name) if not is_link_model else model_type #
         dtype_model = torch.bfloat16 if model_type == "FLUX" else torch.float16
 
         if not os.path.exists(model_name):
-            print("debug", model_name, vae_model, task, controlnet_model)
+            logger.debug(f"model_name={model_name}, vae_model={vae_model}, task={task}, controlnet_model={controlnet_model}")
             _ = download_diffuser_repo(
                 repo_name=model_name,
                 model_type=model_type,
@@ -249,10 +281,10 @@ class GuiSD:
                     type_model_precision=dtype_model,
                     retain_task_model_in_cache=False,
                     controlnet_model=controlnet_model,
-                    device="cpu",
+                    device="cpu" if IS_ZERO_GPU else None,
                     env_components=components,
                 )
-                self.model.advanced_params(image_preprocessor_cuda_active=True)
+                self.model.advanced_params(image_preprocessor_cuda_active=IS_GPU_MODE)
             else:
                 if self.model.base_model_id != model_name:
                     load_now_time = datetime.now()
@@ -262,7 +294,8 @@ class GuiSD:
                         print("Waiting for the previous model's time ops...")
                         time.sleep(9 - elapsed_time)
 
-                self.model.device = torch.device("cpu")
+                if IS_ZERO_GPU:
+                    self.model.device = torch.device("cpu")
                 self.model.load_pipe(
                     model_name,
                     task_name=TASK_STABLEPY[task],
@@ -425,7 +458,7 @@ class GuiSD:
                                   lora_scale3, lora4, lora_scale4, lora5, lora_scale5, lora6, lora_scale6, lora7, lora_scale7)
 ## END MOD
 
-        print("Config model:", model_name, vae_model, loras_list)
+        logger.debug(f"Config model: {model_name}, {vae_model}, {loras_list}")
 
         task = TASK_STABLEPY[task]
 
@@ -523,19 +556,19 @@ class GuiSD:
             "distance_threshold": distance_threshold,
             "recolor_gamma_correction": float(recolor_gamma_correction),
             "tile_blur_sigma": int(tile_blur_sigma),
-            "lora_A": lora1 if lora1 != "None" else None,
+            "lora_A": lora_chk(lora1),
             "lora_scale_A": lora_scale1,
-            "lora_B": lora2 if lora2 != "None" else None,
+            "lora_B": lora_chk(lora2),
             "lora_scale_B": lora_scale2,
-            "lora_C": lora3 if lora3 != "None" else None,
+            "lora_C": lora_chk(lora3),
             "lora_scale_C": lora_scale3,
-            "lora_D": lora4 if lora4 != "None" else None,
+            "lora_D": lora_chk(lora4),
             "lora_scale_D": lora_scale4,
-            "lora_E": lora5 if lora5 != "None" else None,
+            "lora_E": lora_chk(lora5),
             "lora_scale_E": lora_scale5,
-            "lora_F": lora6 if lora6 != "None" else None,
+            "lora_F": lora_chk(lora6),
             "lora_scale_F": lora_scale6,
-            "lora_G": lora7 if lora7 != "None" else None,
+            "lora_G": lora_chk(lora7),
             "lora_scale_G": lora_scale7,
 ## BEGIN MOD
             "textual_inversion": get_embed_list(self.model.class_name) if textual_inversion else [],
@@ -595,11 +628,11 @@ class GuiSD:
         # kwargs for diffusers pipeline
         if guidance_rescale:
             pipe_params["guidance_rescale"] = guidance_rescale
-
-        self.model.device = torch.device("cuda:0")
-        if hasattr(self.model.pipe, "transformer") and loras_list != ["None"] * self.model.num_loras:
-            self.model.pipe.transformer.to(self.model.device)
-            print("transformer to cuda")
+        if IS_ZERO_GPU:
+            self.model.device = torch.device("cuda:0")
+            if hasattr(self.model.pipe, "transformer") and loras_list != ["None"] * self.model.num_loras:
+                self.model.pipe.transformer.to(self.model.device)
+                logger.debug("transformer to cuda")
 
         actual_progress = 0
         info_images = gr.update()
@@ -629,7 +662,7 @@ class GuiSD:
 
                 download_links = "<br>".join(
                     [
-                        f'<a href="{path.replace("/images/", "/file=/home/user/app/images/")}" download="{os.path.basename(path)}">Download Image {i + 1}</a>'
+                        f'<a href="{path.replace("/images/", f"/gradio_api/file={allowed_path}/")}" download="{os.path.basename(path)}">Download Image {i + 1}</a>'
                         for i, path in enumerate(image_path)
                     ]
                 )
@@ -741,7 +774,8 @@ def sd_gen_generate_pipeline(*args):
 
 @spaces.GPU(duration=15)
 def process_upscale(image, upscaler_name, upscaler_size):
-    if image is None: return None
+    if image is None:
+        return None
 
     from stablepy.diffusers_vanilla.utils import save_pil_image_with_metadata
     from stablepy import load_upscaler_model
@@ -758,7 +792,7 @@ def process_upscale(image, upscaler_name, upscaler_size):
 
         name_upscaler = f"./{DIRECTORY_UPSCALERS}/{name_upscaler.split('/')[-1]}"
 
-    scaler_beta = load_upscaler_model(model=name_upscaler, tile=0, tile_overlap=8, device="cuda", half=True)
+    scaler_beta = load_upscaler_model(model=name_upscaler, tile=(0 if IS_ZERO_GPU else 192), tile_overlap=8, device=("cuda" if IS_GPU_MODE else "cpu"), half=IS_GPU_MODE)
     image_up = scaler_beta.upscale(image, upscaler_size, True)
 
     image_path = save_pil_image_with_metadata(image_up, f'{os.getcwd()}/up_images', exif_image)
@@ -767,8 +801,8 @@ def process_upscale(image, upscaler_name, upscaler_size):
 
 
 # https://huggingface.co/spaces/BestWishYsh/ConsisID-preview-Space/discussions/1#674969a022b99c122af5d407
-dynamic_gpu_duration.zerogpu = True
-sd_gen_generate_pipeline.zerogpu = True
+# dynamic_gpu_duration.zerogpu = True
+# sd_gen_generate_pipeline.zerogpu = True
 sd_gen = GuiSD()
 
 
@@ -786,7 +820,7 @@ CSS ="""
 .desc [src$='#float'] { float: right; margin: 20px; }
 """
 
-with gr.Blocks(theme='NoCrypt/miku@>=1.2.2', elem_id="main", fill_width=True, css=CSS, delete_cache=(60, 3600)) as app:
+with gr.Blocks(theme=args.theme, elem_id="main", fill_width=True, fill_height=False, css=CSS) as app:
     gr.Markdown("# 🧩 DiffuseCraft Mod", elem_classes="title")
     gr.Markdown("This space is a modification of [r3gm's DiffuseCraft](https://huggingface.co/spaces/r3gm/DiffuseCraft).", elem_classes="info")
     with gr.Column():
@@ -912,7 +946,7 @@ with gr.Blocks(theme='NoCrypt/miku@>=1.2.2', elem_id="main", fill_width=True, cs
                         with gr.Row():
                             seed_gui = gr.Number(minimum=-1, maximum=2**32-1, value=-1, label="Seed")
                             pag_scale_gui = gr.Slider(minimum=0.0, maximum=10.0, step=0.1, value=0.0, label="PAG Scale")
-                            num_images_gui = gr.Slider(minimum=1, maximum=5, step=1, value=1, label="Images")
+                            num_images_gui = gr.Slider(minimum=1, maximum=(5 if IS_ZERO_GPU else 20), step=1, value=1, label="Images")
                             clip_skip_gui = gr.Checkbox(value=False, label="Layer 2 Clip Skip")
                             free_u_gui = gr.Checkbox(value=False, label="FreeU")
                         with gr.Row():
@@ -939,7 +973,22 @@ with gr.Blocks(theme='NoCrypt/miku@>=1.2.2', elem_id="main", fill_width=True, cs
                                     "Schedule type": gr.update(value="Automatic"),
                                     "PAG": gr.update(value=.0),
                                     "FreeU": gr.update(value=False),
+                                    "Hires upscaler": gr.update(),
+                                    "Hires upscale": gr.update(),
+                                    "Hires steps": gr.update(),
+                                    "Hires denoising strength": gr.update(),
+                                    "Hires CFG": gr.update(),
+                                    "Hires sampler": gr.update(),
+                                    "Hires schedule type": gr.update(),
+                                    "Image resolution": gr.update(value=1024),
+                                    "Strength": gr.update(),
                                 }
+
+                                # Generate up to 7 LoRAs
+                                for i in range(1, 8):
+                                    valid_receptors[f"Lora_{i}"] = gr.update()
+                                    valid_receptors[f"Lora_scale_{i}"] = gr.update()
+
                                 valid_keys = list(valid_receptors.keys())
 
                                 parameters = extract_parameters(base_prompt)
@@ -953,6 +1002,36 @@ with gr.Blocks(theme='NoCrypt/miku@>=1.2.2', elem_id="main", fill_width=True, cs
                                             parameters["Sampler"] = value_sampler
                                             parameters["Schedule type"] = s_type
 
+                                params_lora = []
+                                if ">" in parameters["prompt"] and "<" in parameters["prompt"]:
+                                    params_lora = re.findall(r'<lora:[^>]+>', parameters["prompt"])
+                                if "Loras" in parameters:
+                                    params_lora += re.findall(r'<lora:[^>]+>', parameters["Loras"])
+
+                                if params_lora:
+                                    parsed_params = []
+                                    for tag_l in params_lora:
+                                        try:
+                                            inner = tag_l.strip("<>")        # remove < >
+                                            _, data_l = inner.split(":", 1)  # remove the "lora:" part
+                                            parts_l = data_l.split(":")
+
+                                            name_l = parts_l[0]
+                                            weight_l = float(parts_l[1]) if len(parts_l) > 1 else 1.0  # default weight = 1.0
+
+                                            parsed_params.append((name_l, weight_l))
+                                        except Exception as e:
+                                            print(f"Error parsing LoRA tag {tag_l}: {e}")
+
+                                    num_lora = 1
+                                    for parsed_l, parsed_s in parsed_params:
+                                        filtered_loras = [m for m in lora_model_list if parsed_l in m]
+                                        if filtered_loras:
+                                            parameters[f"Lora_{num_lora}"] = filtered_loras[0]
+                                            parameters[f"Lora_scale_{num_lora}"] = parsed_s
+                                            num_lora += 1
+
+                                # continue = discard new value
                                 for key, val in parameters.items():
                                     # print(val)
                                     if key in valid_keys:
@@ -960,9 +1039,12 @@ with gr.Blocks(theme='NoCrypt/miku@>=1.2.2', elem_id="main", fill_width=True, cs
                                             if key == "Sampler":
                                                 if val not in scheduler_names:
                                                     continue
-                                            if key == "Schedule type":
+                                            if key in ["Schedule type", "Hires schedule type"]:
                                                 if val not in SCHEDULE_TYPE_OPTIONS:
-                                                    val = "Automatic"
+                                                    continue
+                                            if key == "Hires sampler":
+                                                if val not in POST_PROCESSING_SAMPLER:
+                                                    continue
                                             elif key == "Clip skip":
                                                 if "," in str(val):
                                                     val = val.replace(",", "")
@@ -970,15 +1052,15 @@ with gr.Blocks(theme='NoCrypt/miku@>=1.2.2', elem_id="main", fill_width=True, cs
                                                     val = True
                                             if key == "prompt":
                                                 if ">" in val and "<" in val:
-                                                    val = re.sub(r'<[^>]+>', '', val)
+                                                    val = re.sub(r'<[^>]+>', '', val)  # Delete html and loras
                                                     print("Removed LoRA written in the prompt")
                                             if key in ["prompt", "neg_prompt"]:
                                                 val = re.sub(r'\s+', ' ', re.sub(r',+', ',', val)).strip()
-                                            if key in ["Steps", "width", "height", "Seed"]:
+                                            if key in ["Steps", "width", "height", "Seed", "Hires steps", "Image resolution"]:
                                                 val = int(val)
                                             if key == "FreeU":
                                                 val = True
-                                            if key in ["CFG scale", "PAG"]:
+                                            if key in ["CFG scale", "PAG", "Hires upscale", "Hires denoising strength", "Hires CFG", "Strength"]:
                                                 val = float(val)
                                             if key == "Model":
                                                 filtered_models = [m for m in model_list if val in m]
@@ -986,8 +1068,12 @@ with gr.Blocks(theme='NoCrypt/miku@>=1.2.2', elem_id="main", fill_width=True, cs
                                                     val = filtered_models[0]
                                                 else:
                                                     val = name_model
+                                            if key == "Hires upscaler":
+                                                if val not in UPSCALER_KEYS:
+                                                    continue
                                             if key == "Seed":
                                                 continue
+
                                             valid_receptors[key] = gr.update(value=val)
                                             # print(val, type(val))
                                             # print(valid_receptors)
@@ -995,24 +1081,6 @@ with gr.Blocks(theme='NoCrypt/miku@>=1.2.2', elem_id="main", fill_width=True, cs
                                             print(str(e))
                                 return [value for value in valid_receptors.values()]
 
-                            set_params_gui.click(
-                                run_set_params_gui, [prompt_gui, model_name_gui], [
-                                    prompt_gui,
-                                    neg_prompt_gui,
-                                    steps_gui,
-                                    img_width_gui,
-                                    img_height_gui,
-                                    seed_gui,
-                                    sampler_gui,
-                                    cfg_gui,
-                                    clip_skip_gui,
-                                    model_name_gui,
-                                    schedule_type_gui,
-                                    pag_scale_gui,
-                                    free_u_gui,
-                                ],
-                            )
-
                             def run_clear_prompt_gui():
                                 return gr.update(value=""), gr.update(value="")
                             clear_prompt_gui.click(
@@ -1030,8 +1098,9 @@ with gr.Blocks(theme='NoCrypt/miku@>=1.2.2', elem_id="main", fill_width=True, cs
                             return gr.Dropdown(label=label, choices=get_all_lora_tupled_list(), value="", allow_custom_value=True, elem_classes="lora", min_width=320, visible=visible)
 
                         def lora_scale_slider(label, visible=True):
-                            return gr.Slider(minimum=-2, maximum=2, step=0.01, value=1.00, label=label, visible=visible)
-                        
+                            val_lora = 8 if IS_ZERO_GPU else 8 #
+                            return gr.Slider(minimum=-val_lora, maximum=val_lora, step=0.01, value=0.33, label=label, visible=visible)
+                    
                         def lora_textbox(label):
                             return gr.Textbox(label=label, info="Example of prompt:", value="None", show_copy_button=True, interactive=False, visible=False)
                         
@@ -1077,16 +1146,16 @@ with gr.Blocks(theme='NoCrypt/miku@>=1.2.2', elem_id="main", fill_width=True, cs
                                         lora5_copy_gui = gr.Button(value="Copy example to prompt", visible=False)
                                     lora5_desc_gui = gr.Markdown(value="", visible=False)
                             with gr.Column():
-                                lora6_gui = lora_dropdown("LoRA6", visible=False)
-                                lora_scale_6_gui = lora_scale_slider("LoRA Scale 6", visible=False)
+                                lora6_gui = lora_dropdown("LoRA6", visible=(not IS_ZERO_GPU))
+                                lora_scale_6_gui = lora_scale_slider("LoRA Scale 6", visible=(not IS_ZERO_GPU))
                                 with gr.Row():
                                     with gr.Group():
                                         lora6_info_gui = lora_textbox("LoRA6 prompts")
                                         lora6_copy_gui = gr.Button(value="Copy example to prompt", visible=False)
                                     lora6_desc_gui = gr.Markdown(value="", visible=False)
                             with gr.Column():
-                                lora7_gui = lora_dropdown("LoRA7", visible=False)
-                                lora_scale_7_gui = lora_scale_slider("LoRA Scale 7", visible=False)
+                                lora7_gui = lora_dropdown("LoRA7", visible=(not IS_ZERO_GPU))
+                                lora_scale_7_gui = lora_scale_slider("LoRA Scale 7", visible=(not IS_ZERO_GPU))
                                 with gr.Row():
                                     with gr.Group():
                                         lora7_info_gui = lora_textbox("LoRA7 prompts")
@@ -1108,7 +1177,7 @@ with gr.Blocks(theme='NoCrypt/miku@>=1.2.2', elem_id="main", fill_width=True, cs
                             search_civitai_result_lora = gr.Dropdown(label="Search Results", choices=[("", "")], value="", allow_custom_value=True, visible=False)
                             with gr.Row():
                                 text_lora = gr.Textbox(label="LoRA's download URL", placeholder="https://civitai.com/api/download/models/28907", info="It has to be .safetensors files, and you can also download them from Hugging Face.", lines=1, scale=4)
-                                romanize_text = gr.Checkbox(value=False, label="Transliterate name", scale=1, visible=False)
+                                romanize_text = gr.Checkbox(value=False, label="Transliterate name", visible=(not IS_ZERO_GPU))
                             button_lora = gr.Button("Get and Refresh the LoRA Lists")
                             new_lora_status = gr.HTML()
                         with gr.Accordion("From Local", open=True, visible=True):
@@ -1119,8 +1188,8 @@ with gr.Blocks(theme='NoCrypt/miku@>=1.2.2', elem_id="main", fill_width=True, cs
                         with gr.Accordion("Hires fix", open=False, visible=True) as menu_hires:
                             upscaler_model_path_gui = gr.Dropdown(label="Upscaler", choices=UPSCALER_KEYS, value=UPSCALER_KEYS[0])
                             with gr.Row():
-                                upscaler_increases_size_gui = gr.Slider(minimum=1.1, maximum=6., step=0.1, value=1.0, label="Upscale by")
-                                upscaler_tile_size_gui = gr.Slider(minimum=0, maximum=512, step=16, value=0, label="Upscaler Tile Size", info="0 = no tiling")
+                                upscaler_increases_size_gui = gr.Slider(minimum=1.1, maximum=6., step=0.1, value=1.2, label="Upscale by")
+                                upscaler_tile_size_gui = gr.Slider(minimum=0, maximum=512, step=16, value=(0 if IS_ZERO_GPU else 192), label="Upscaler Tile Size", info="0 = no tiling")
                                 upscaler_tile_overlap_gui = gr.Slider(minimum=0, maximum=48, step=1, value=8, label="Upscaler Tile Overlap")
                             with gr.Row():
                                 hires_steps_gui = gr.Slider(minimum=0, value=30, maximum=100, step=1, label="Hires Steps")
@@ -1230,8 +1299,8 @@ with gr.Blocks(theme='NoCrypt/miku@>=1.2.2', elem_id="main", fill_width=True, cs
                                 )
 
                             with gr.Row():
-                                value_threshold_gui = gr.Slider(minimum=1, maximum=2.0, step=0.01, value=0.1, label="'MLSD' Hough value threshold")
-                                distance_threshold_gui = gr.Slider(minimum=1, maximum=20.0, step=0.01, value=0.1, label="'MLSD' Hough distance threshold")
+                                value_threshold_gui = gr.Slider(minimum=0.0, maximum=2.0, step=0.01, value=0.1, label="'MLSD' Hough value threshold")
+                                distance_threshold_gui = gr.Slider(minimum=0.0, maximum=20.0, step=0.01, value=0.1, label="'MLSD' Hough distance threshold")
                                 recolor_gamma_correction_gui = gr.Number(minimum=0., maximum=25., value=1., step=0.001, label="'RECOLOR' gamma correction")
                                 tile_blur_sigma_gui = gr.Number(minimum=0, maximum=100, value=9, step=1, label="'TILE' blur sigma")
 
@@ -1296,20 +1365,62 @@ with gr.Blocks(theme='NoCrypt/miku@>=1.2.2', elem_id="main", fill_width=True, cs
                                 hires_before_adetailer_gui = gr.Checkbox(value=False, label="Hires Before Adetailer")
                                 hires_after_adetailer_gui = gr.Checkbox(value=True, label="Hires After Adetailer")
                                 generator_in_cpu_gui = gr.Checkbox(value=False, label="Generator in CPU")
-
+                            with gr.Column(visible=(not IS_ZERO_GPU)):
+                                image_storage_location_gui = gr.Textbox(value=img_path, label="Image Storage Location")
+                                disable_progress_bar_gui = gr.Checkbox(value=False, label="Disable Progress Bar")
+                                leave_progress_bar_gui = gr.Checkbox(value=True, label="Leave Progress Bar")
+                        
                         with gr.Accordion("More settings", open=False, visible=False):
                             loop_generation_gui = gr.Slider(minimum=1, value=1, label="Loop Generation")
                             retain_task_cache_gui = gr.Checkbox(value=True, label="Retain task model in cache")
-                            leave_progress_bar_gui = gr.Checkbox(value=True, label="Leave Progress Bar")
-                            disable_progress_bar_gui = gr.Checkbox(value=False, label="Disable Progress Bar")
                             display_images_gui = gr.Checkbox(value=False, label="Display Images")
                             image_previews_gui = gr.Checkbox(value=True, label="Image Previews")
-                            image_storage_location_gui = gr.Textbox(value="./images", label="Image Storage Location")
                             retain_compel_previous_load_gui = gr.Checkbox(value=False, label="Retain Compel Previous Load")
                             retain_detailfix_model_previous_load_gui = gr.Checkbox(value=False, label="Retain Detailfix Model Previous Load")
                             retain_hires_model_previous_load_gui = gr.Checkbox(value=False, label="Retain Hires Model Previous Load")
                             xformers_memory_efficient_attention_gui = gr.Checkbox(value=False, label="Xformers Memory Efficient Attention")
 
+                        set_params_gui.click(
+                            run_set_params_gui, [prompt_gui, model_name_gui], [
+                                prompt_gui,
+                                neg_prompt_gui,
+                                steps_gui,
+                                img_width_gui,
+                                img_height_gui,
+                                seed_gui,
+                                sampler_gui,
+                                cfg_gui,
+                                clip_skip_gui,
+                                model_name_gui,
+                                schedule_type_gui,
+                                pag_scale_gui,
+                                free_u_gui,
+                                upscaler_model_path_gui,
+                                upscaler_increases_size_gui,
+                                hires_steps_gui,
+                                hires_denoising_strength_gui,
+                                hires_guidance_scale_gui,
+                                hires_sampler_gui,
+                                hires_schedule_type_gui,
+                                image_resolution_gui,
+                                strength_gui,
+                                lora1_gui,
+                                lora_scale_1_gui,
+                                lora2_gui,
+                                lora_scale_2_gui,
+                                lora3_gui,
+                                lora_scale_3_gui,
+                                lora4_gui,
+                                lora_scale_4_gui,
+                                lora5_gui,
+                                lora_scale_5_gui,
+                                lora6_gui,
+                                lora_scale_6_gui,
+                                lora7_gui,
+                                lora_scale_7_gui,
+                            ],
+                        )
+
             with gr.Accordion("Examples and help", open=True, visible=True) as menu_example:
                 gr.Examples(
                     examples=EXAMPLES_GUI,
@@ -1337,36 +1448,8 @@ with gr.Blocks(theme='NoCrypt/miku@>=1.2.2', elem_id="main", fill_width=True, cs
 
         with gr.Tab("Inpaint mask maker", render=True):
 
-            def create_mask_now(img, invert):            
-                import numpy as np
-                import time
-
-                time.sleep(0.5)
-
-                transparent_image = img["layers"][0]
-
-                # Extract the alpha channel
-                alpha_channel = np.array(transparent_image)[:, :, 3]
-
-                # Create a binary mask by thresholding the alpha channel
-                binary_mask = alpha_channel > 1
-
-                if invert:
-                    print("Invert")
-                    # Invert the binary mask so that the drawn shape is white and the rest is black
-                    binary_mask = np.invert(binary_mask)
-
-                # Convert the binary mask to a 3-channel RGB mask
-                rgb_mask = np.stack((binary_mask,) * 3, axis=-1)
-
-                # Convert the mask to uint8
-                rgb_mask = rgb_mask.astype(np.uint8) * 255
-
-                return img["background"], rgb_mask
-
             with gr.Row():
                 with gr.Column(scale=2):
-                    # image_base = gr.ImageEditor(label="Base image", show_label=True, brush=gr.Brush(colors=["#000000"]))
                     image_base = gr.ImageEditor(
                         sources=["upload", "clipboard"],
                         # crop_size="1:1",
@@ -1383,10 +1466,21 @@ with gr.Blocks(theme='NoCrypt/miku@>=1.2.2', elem_id="main", fill_width=True, cs
                                 # "hsl(360, 120, 120)" # in fact any valid colorstring
                             ]
                         ),
-                        eraser=gr.Eraser(default_size="16")
+                        eraser=gr.Eraser(default_size="16"),
+                        render=True,
+                        visible=False,
+                        interactive=False,
                     )
+
+                    show_canvas = gr.Button("SHOW INPAINT CANVAS")
+
+                    def change_visibility_canvas():
+                        return gr.update(visible=True, interactive=True), gr.update(visible=False)
+                    show_canvas.click(change_visibility_canvas, [], [image_base, show_canvas])
+
                     invert_mask = gr.Checkbox(value=False, label="Invert mask")
                     btn = gr.Button("Create mask")
+
                 with gr.Column(scale=1):
                     img_source = gr.Image(interactive=False)
                     img_result = gr.Image(label="Mask image", show_label=True, interactive=False)
@@ -1780,6 +1874,13 @@ with gr.Blocks(theme='NoCrypt/miku@>=1.2.2', elem_id="main", fill_width=True, cs
     gr.LoginButton()
     gr.DuplicateButton(value="Duplicate Space for private use (This demo does not work on CPU. Requires GPU Space)")
 
-app.queue()
-app.launch(show_error=True, debug=True) # allowed_paths=["./images/"], show_error=True, debug=True
+if __name__ == "__main__":
+    app.queue()
+    app.launch(
+        show_error=True,
+        share=args.share_enabled,
+        debug=True,
+        ssr_mode=args.ssr,
+        allowed_paths=[allowed_path],
+    )
 ## END MOD

constants.py

@@ -9,6 +9,8 @@ from stablepy import (
     IP_ADAPTERS_SDXL,
 )
 
+IS_ZERO_GPU = bool(os.getenv("SPACES_ZERO_GPU"))
+
 # - **Download Models**
 DOWNLOAD_MODEL = "https://huggingface.co/TechnoByte/MilkyWonderland/resolve/main/milkyWonderland_v40.safetensors"
 
@@ -23,12 +25,12 @@ LOAD_DIFFUSERS_FORMAT_MODEL = [
     'Laxhar/noobai-XL-1.1',
     'Laxhar/noobai-XL-Vpred-1.0',
     'black-forest-labs/FLUX.1-dev',
+    'black-forest-labs/FLUX.1-Krea-dev',
     'John6666/blue-pencil-flux1-v021-fp8-flux',
     'John6666/wai-ani-flux-v10forfp8-fp8-flux',
     'John6666/xe-anime-flux-v04-fp8-flux',
     'John6666/lyh-anime-flux-v2a1-fp8-flux',
     'John6666/carnival-unchained-v10-fp8-flux',
-    'John6666/iniverse-mix-xl-sfwnsfw-fluxdfp16nsfwv11-fp8-flux',
     'Freepik/flux.1-lite-8B-alpha',
     'shauray/FluxDev-HyperSD-merged',
     'mikeyandfriends/PixelWave_FLUX.1-dev_03',
@@ -37,23 +39,19 @@ LOAD_DIFFUSERS_FORMAT_MODEL = [
     # 'ostris/OpenFLUX.1',
     'shuttleai/shuttle-3-diffusion',
     'Laxhar/noobai-XL-1.0',
-    'John6666/noobai-xl-nai-xl-epsilonpred10version-sdxl',
     'Laxhar/noobai-XL-0.77',
     'John6666/noobai-xl-nai-xl-epsilonpred075version-sdxl',
     'Laxhar/noobai-XL-0.6',
     'John6666/noobai-xl-nai-xl-epsilonpred05version-sdxl',
     'John6666/noobai-cyberfix-v10-sdxl',
     'John6666/noobaiiter-xl-vpred-v075-sdxl',
-    'John6666/ntr-mix-illustrious-xl-noob-xl-v40-sdxl',
-    'John6666/ntr-mix-illustrious-xl-noob-xl-ntrmix35-sdxl',
-    'John6666/ntr-mix-illustrious-xl-noob-xl-v777-sdxl',
-    'John6666/ntr-mix-illustrious-xl-noob-xl-v777forlora-sdxl',
+    'John6666/ripplemix-noob-vpred10-illustrious01-v14-sdxl',
+    'John6666/sigmaih-15-sdxl',
     'John6666/ntr-mix-illustrious-xl-noob-xl-xi-sdxl',
     'John6666/ntr-mix-illustrious-xl-noob-xl-xii-sdxl',
     'John6666/ntr-mix-illustrious-xl-noob-xl-xiii-sdxl',
     'John6666/mistoon-anime-v10illustrious-sdxl',
-    'John6666/hassaku-xl-illustrious-v10-sdxl',
-    'John6666/hassaku-xl-illustrious-v10style-sdxl',
+    'John6666/hassaku-xl-illustrious-v22-sdxl',
     'John6666/haruki-mix-illustrious-v10-sdxl',
     'John6666/noobreal-v10-sdxl',
     'John6666/complicated-noobai-merge-vprediction-sdxl',
@@ -64,6 +62,7 @@ LOAD_DIFFUSERS_FORMAT_MODEL = [
     'Laxhar/noobai-XL-Vpred-0.65',
     'Laxhar/noobai-XL-Vpred-0.6',
     'John6666/cat-tower-noobai-xl-checkpoint-v14vpred-sdxl',
+    'John6666/cat-tower-noobai-xl-checkpoint-v15vpred-sdxl',
     'John6666/noobai-xl-nai-xl-vpred05version-sdxl',
     'John6666/noobai-fusion2-vpred-itercomp-v1-sdxl',
     'John6666/noobai-xl-nai-xl-vpredtestversion-sdxl',
@@ -74,19 +73,34 @@ LOAD_DIFFUSERS_FORMAT_MODEL = [
     'John6666/obsession-illustriousxl-v21-sdxl',
     'John6666/obsession-illustriousxl-v30-sdxl',
     'John6666/obsession-illustriousxl-v31-sdxl',
+    'John6666/one-obsession-13-sdxl',
+    'John6666/one-obsession-14-24d-sdxl',
+    'John6666/one-obsession-15-noobai-sdxl',
+    'John6666/one-obsession-v16-noobai-sdxl',
+    'John6666/prefect-illustrious-xl-v3-sdxl',
     'John6666/wai-nsfw-illustrious-v70-sdxl',
+    'John6666/wai-nsfw-illustrious-sdxl-v140-sdxl',
     'John6666/illustrious-pony-mix-v3-sdxl',
-    'John6666/nova-anime-xl-illustriousv10-sdxl',
-    'John6666/nova-orange-xl-v30-sdxl',
+    'John6666/nova-anime-xl-il-v90-sdxl',
+    'John6666/nova-anime-xl-il-v110-sdxl',
+    'John6666/nova-orange-xl-re-v10-sdxl',
+    'John6666/nova-orange-xl-v110-sdxl',
+    'John6666/nova-orange-xl-re-v20-sdxl',
+    'John6666/nova-unreal-xl-v60-sdxl',
+    'John6666/nova-unreal-xl-v70-sdxl',
+    'John6666/nova-unreal-xl-v80-sdxl',
+    'John6666/nova-cartoon-xl-v40-sdxl',
     'John6666/silvermoon-mix03-illustrious-v10-sdxl',
     'eienmojiki/Anything-XL',
     'eienmojiki/Starry-XL-v5.2',
+    'votepurchase/plantMilkModelSuite_walnut',
     'John6666/meinaxl-v2-sdxl',
     'Eugeoter/artiwaifu-diffusion-2.0',
     'comin/IterComp',
-    'John6666/epicrealism-xl-vxiabeast-sdxl',
-    'John6666/epicrealism-xl-v10kiss2-sdxl',
     'John6666/epicrealism-xl-v8kiss-sdxl',
+    'John6666/epicrealism-xl-v10kiss2-sdxl',
+    'John6666/epicrealism-xl-vxiabeast-sdxl',
+    'John6666/epicrealism-xl-vxvii-crystal-clear-realism-sdxl',
     'misri/zavychromaxl_v80',
     'SG161222/RealVisXL_V4.0',
     'SG161222/RealVisXL_V5.0',
@@ -102,8 +116,10 @@ LOAD_DIFFUSERS_FORMAT_MODEL = [
     'John6666/ebara-mfcg-pony-mix-v12-sdxl',
     'John6666/t-ponynai3-v51-sdxl',
     'John6666/t-ponynai3-v65-sdxl',
+    'John6666/t-ponynai3-v7-sdxl',
     'John6666/prefect-pony-xl-v3-sdxl',
     'John6666/prefect-pony-xl-v4-sdxl',
+    'John6666/prefect-pony-xl-v50-sdxl',
     'John6666/mala-anime-mix-nsfw-pony-xl-v5-sdxl',
     'John6666/wai-ani-nsfw-ponyxl-v10-sdxl',
     'John6666/wai-real-mix-v11-sdxl',
@@ -111,13 +127,14 @@ LOAD_DIFFUSERS_FORMAT_MODEL = [
     'John6666/wai-c-v6-sdxl',
     'John6666/iniverse-mix-xl-sfwnsfw-pony-guofeng-v43-sdxl',
     'John6666/sifw-annihilation-xl-v2-sdxl',
+    'John6666/sifw-annihilation-xl-v305illustrious-beta-sdxl',
     'John6666/photo-realistic-pony-v5-sdxl',
     'John6666/pony-realism-v21main-sdxl',
     'John6666/pony-realism-v22main-sdxl',
-    'John6666/cyberrealistic-pony-v63-sdxl',
-    'John6666/cyberrealistic-pony-v64-sdxl',
+    'John6666/pony-realism-v23-ultra-sdxl',
     'John6666/cyberrealistic-pony-v65-sdxl',
     'John6666/cyberrealistic-pony-v7-sdxl',
+    'John6666/cyberrealistic-pony-v127-alternative-sdxl',
     'GraydientPlatformAPI/realcartoon-pony-diffusion',
     'John6666/nova-anime-xl-pony-v5-sdxl',
     'John6666/autismmix-sdxl-autismmix-pony-sdxl',
@@ -127,13 +144,15 @@ LOAD_DIFFUSERS_FORMAT_MODEL = [
     'John6666/duchaiten-pony-real-v11fix-sdxl',
     'John6666/duchaiten-pony-real-v20-sdxl',
     'John6666/duchaiten-pony-xl-no-score-v70-sdxl',
-    'Spestly/OdysseyXL-3.0',
-    'Spestly/OdysseyXL-4.0',
     'KBlueLeaf/Kohaku-XL-Zeta',
     'cagliostrolab/animagine-xl-3.1',
+    'cagliostrolab/animagine-xl-4.0',
     'yodayo-ai/kivotos-xl-2.0',
     'yodayo-ai/holodayo-xl-2.1',
     'yodayo-ai/clandestine-xl-1.0',
+    'https://huggingface.co/chemwolf/Karmix-XL-v0/resolve/main/Karmix-XL-v0.safetensors?download=true',
+    'https://civitai.com/api/download/models/128713?type=Model&format=SafeTensor&size=pruned&fp=fp16',
+    'https://civitai.com/models/30240?modelVersionId=125771',
     'digiplay/majicMIX_sombre_v2',
     'digiplay/majicMIX_realistic_v6',
     'digiplay/majicMIX_realistic_v7',
@@ -159,9 +178,9 @@ LOAD_DIFFUSERS_FORMAT_MODEL = [
     'GraydientPlatformAPI/realcartoon3d-17',
     'GraydientPlatformAPI/realcartoon-pixar11',
     'GraydientPlatformAPI/realcartoon-real17',
-    'nitrosocke/Ghibli-Diffusion',
 ]
 
+
 DIFFUSERS_FORMAT_LORAS = [
     "nerijs/animation2k-flux",
     "XLabs-AI/flux-RealismLora",
@@ -183,8 +202,11 @@ DIRECTORY_VAES = 'vaes'
 DIRECTORY_EMBEDS = 'embedings'
 DIRECTORY_UPSCALERS = 'upscalers'
 
-CACHE_HF = "/home/user/.cache/huggingface/hub/"
 STORAGE_ROOT = "/home/user/"
+CACHE_HF_ROOT = os.path.expanduser(os.getenv("HF_HOME")) if os.getenv("HF_HOME") else os.path.expanduser("~/cache/huggingface")
+CACHE_HF = os.path.join(CACHE_HF_ROOT, "hub")
+if IS_ZERO_GPU:
+    os.environ["HF_HOME"] = CACHE_HF
 
 TASK_STABLEPY = {
     'txt2img': 'txt2img',
@@ -226,6 +248,7 @@ UPSCALER_DICT_GUI = {
     # "realesr-general-x4v3": "https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesr-general-x4v3.pth",
     # "realesr-general-wdn-x4v3": "https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesr-general-wdn-x4v3.pth",
     "4x-UltraSharp": "https://huggingface.co/Shandypur/ESRGAN-4x-UltraSharp/resolve/main/4x-UltraSharp.pth",
+    "Real-ESRGAN-Anime-finetuning": "https://huggingface.co/danhtran2mind/Real-ESRGAN-Anime-finetuning/resolve/main/Real-ESRGAN-Anime-finetuning.pth",
     "4x_foolhardy_Remacri": "https://huggingface.co/FacehugmanIII/4x_foolhardy_Remacri/resolve/main/4x_foolhardy_Remacri.pth",
     "Remacri4xExtraSmoother": "https://huggingface.co/hollowstrawberry/upscalers-backup/resolve/main/ESRGAN/Remacri%204x%20ExtraSmoother.pth",
     "AnimeSharp4x": "https://huggingface.co/hollowstrawberry/upscalers-backup/resolve/main/ESRGAN/AnimeSharp%204x.pth",
@@ -359,9 +382,11 @@ SUBTITLE_GUI = (
     " to perform different tasks in image generation."
 )
 
+msg_zero = "" if not IS_ZERO_GPU else "- The current space runs on a ZERO GPU which is assigned for approximately 60 seconds; Therefore, if you submit expensive tasks, the operation may be canceled upon reaching the maximum allowed time with 'GPU TASK ABORTED'."
+
 HELP_GUI = (
-    """### Help:
-    - The current space runs on a ZERO GPU which is assigned for approximately 60 seconds; Therefore, if you submit expensive tasks, the operation may be canceled upon reaching the maximum allowed time with 'GPU TASK ABORTED'.
+    f"""### Help:
+    {msg_zero}
     - Distorted or strange images often result from high prompt weights, so it's best to use low weights and scales, and consider using Classic variants like 'Classic-original'.
     - For better results with Pony Diffusion, try using sampler DPM++ 1s or DPM2 with Compel or Classic prompt weights.
     """
@@ -485,7 +510,7 @@ EXAMPLES_GUI = [
         20,
         4.0,
         -1,
-        "loras/Coloring_book_-_LineArt.safetensors",
+        ("loras/Coloring_book_-_LineArt.safetensors" if os.path.exists("loras/Coloring_book_-_LineArt.safetensors") else "None"),
         1.0,
         "DPM++ 2M SDE",
         1024,
@@ -580,6 +605,7 @@ EXAMPLES_GUI = [
 RESOURCES = (
     """### Resources
     - John6666's space has some great features you might find helpful [link](https://huggingface.co/spaces/John6666/DiffuseCraftMod).
-    - You can also try the image generator in Colab’s free tier, which provides free GPU [link](https://github.com/R3gm/SD_diffusers_interactive).
+    - Try the image generator in Colab’s free tier, which provides free GPU [link](https://github.com/R3gm/SD_diffusers_interactive).
+    - `DiffuseCraft` in Colab:[link](https://github.com/R3gm/DiffuseCraft?tab=readme-ov-file#diffusecraft).
     """
-)
+)

env.py

@@ -39,6 +39,7 @@ LOAD_DIFFUSERS_FORMAT_MODEL = [
     'Raelina/Raehoshi-illust-XL-4',
     'Raelina/Raehoshi-illust-XL-5',
     'Raelina/Raehoshi-illust-XL-5.1',
+    'Raelina/Raehoshi-illust-XL-6',
     'camenduru/FLUX.1-dev-diffusers',
     'black-forest-labs/FLUX.1-schnell',
     'sayakpaul/FLUX.1-merged',

modutils.py

@@ -1730,7 +1730,7 @@ EXAMPLES_GUI = [
         "votepurchase/ponyDiffusionV6XL",
     ],
     [
-        "1girl, oomuro sakurako, yuru yuri, official art, school uniform, anime artwork, anime style, vibrant, studio anime, highly detailed, masterpiece, best quality, very aesthetic, absurdres",
+        "1girl, oomuro sakurako, yuru yuri, official art, school uniform, anime artwork, anime style, studio anime, highly detailed, masterpiece, best quality, very aesthetic, absurdres",
         "photo, deformed, black and white, realism, disfigured, low contrast, lowres, (bad), text, error, fewer, extra, missing, worst quality, jpeg artifacts, low quality, watermark, unfinished, displeasing, oldest, early, chromatic aberration, signature, extra digits, artistic error, username, scan, [abstract]",
         1,
         40,
@@ -1743,17 +1743,17 @@ EXAMPLES_GUI = [
         "Raelina/Rae-Diffusion-XL-V2",
     ],
     [
-        "1girl, akaza akari, yuru yuri, official art, anime artwork, anime style, vibrant, studio anime, highly detailed, masterpiece, best quality, very aesthetic, absurdres",
-        "photo, deformed, black and white, realism, disfigured, low contrast, lowres, (bad), text, error, fewer, extra, missing, worst quality, jpeg artifacts, low quality, watermark, unfinished, displeasing, oldest, early, chromatic aberration, signature, extra digits, artistic error, username, scan, [abstract]",
+        "1girl, akaza akari, yuru yuri, official art, anime screencap, anime coloring, masterpiece, best quality, absurdres",
+        "bad quality, worst quality, poorly drawn, sketch, multiple views, bad anatomy, bad hands, missing fingers, extra fingers, extra digits, fewer digits, signature, watermark, username",
         1,
-        35,
-        7.0,
+        28,
+        5.5,
         True,
         -1,
         "Euler",
         1024,
         1024,
-        "Raelina/Raemu-XL-V5",
+        "Raelina/Raehoshi-illust-XL-6",
     ],
     [
         "yoshida yuuko, machikado mazoku, 1girl, solo, demon horns,horns, school uniform, long hair, open mouth, skirt, demon girl, ahoge, shiny, shiny hair, anime artwork",

packages.txt

@@ -1 +1 @@
-git-lfs aria2 -y ffmpeg
+git-lfs aria2 ffmpeg

requirements.txt

@@ -1,21 +1,23 @@
-stablepy==0.6.1
-diffusers<=0.32.0
-transformers==4.47.1
-torch==2.4.0
-numpy<2
-gdown
-opencv-python
-torchvision
-accelerate
-optimum[onnxruntime]
-dartrs
-huggingface_hub
-hf_transfer
-hf_xet
-translatepy
-timm
-rapidfuzz
-sentencepiece
-unidecode
-ultralytics>=8.3.47
+stablepy==0.6.2
+diffusers
+transformers<=4.48.3
+accelerate
+huggingface_hub
+hf_transfer
+hf_xet
+torch==2.5.1
+torchvision
+numpy<2
+gdown
+opencv-python
+optimum[onnxruntime]
+#dartrs
+git+https://github.com/John6666cat/dartrs
+translatepy
+timm
+rapidfuzz
+sentencepiece
+unidecode
+matplotlib-inline
+https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.4.11/flash_attn-2.8.3+cu124torch2.5-cp310-cp310-linux_x86_64.whl
 pydantic==2.10.6

tagger/tagger.py

@@ -504,7 +504,7 @@ def gen_prompt(rating: list[str], character: list[str], general: list[str]):
     return ", ".join(all_tags)
 
 
-@spaces.GPU(duration=30)
+@spaces.GPU(duration=10)
 def predict_tags(image: Image.Image, general_threshold: float = 0.3, character_threshold: float = 0.8):
     inputs = wd_processor.preprocess(image, return_tensors="pt")
 

tagger/v2.py

@@ -1,4 +1,4 @@
-import time
+import time, os
 import torch
 from typing import Callable
 from pathlib import Path
@@ -16,16 +16,15 @@ from dartrs.v2 import (
 from dartrs.dartrs import DartTokenizer
 from dartrs.utils import get_generation_config
 
-
 import gradio as gr
 from gradio.components import Component
 
-
 try:
     from output import UpsamplingOutput
 except:
     from .output import UpsamplingOutput
 
+HF_TOKEN = os.getenv("HF_TOKEN", None)
 
 V2_ALL_MODELS = {
     "dart-v2-moe-sft": {
@@ -43,8 +42,8 @@ V2_ALL_MODELS = {
 
 def prepare_models(model_config: dict):
     model_name = model_config["repo"]
-    tokenizer = DartTokenizer.from_pretrained(model_name)
-    model = model_config["class"].from_pretrained(model_name)
+    tokenizer = DartTokenizer.from_pretrained(model_name, auth_token=HF_TOKEN)
+    model = model_config["class"].from_pretrained(model_name, auth_token=HF_TOKEN)
 
     return {
         "tokenizer": tokenizer,