feat: add metadata handling

False alarm, worked as intended before. Sorry for the fuzz.
This reverts commit d16a54edd6.

fooocus_version.py

@@ -1 +1 @@
-version = '2.2.1'
+version = '2.3.1'

language/en.json

@@ -339,6 +339,8 @@
     "sgm_uniform": "sgm_uniform",
     "simple": "simple",
     "ddim_uniform": "ddim_uniform",
+    "VAE": "VAE",
+    "Default (model)": "Default (model)",
     "Forced Overwrite of Sampling Step": "Forced Overwrite of Sampling Step",
     "Set as -1 to disable. For developer debugging.": "Set as -1 to disable. For developer debugging.",
     "Forced Overwrite of Refiner Switch Step": "Forced Overwrite of Refiner Switch Step",
@@ -384,5 +386,6 @@
     "Metadata Scheme": "Metadata Scheme",
     "Image Prompt parameters are not included. Use png and a1111 for compatibility with Civitai.": "Image Prompt parameters are not included. Use png and a1111 for compatibility with Civitai.",
     "fooocus (json)": "fooocus (json)",
-    "a1111 (plain text)": "a1111 (plain text)"
+    "a1111 (plain text)": "a1111 (plain text)",
+    "Unsupported image type in input": "Unsupported image type in input"
 }

ldm_patched/modules/sd.py

@@ -427,12 +427,13 @@ def load_checkpoint(config_path=None, ckpt_path=None, output_vae=True, output_cl
 
     return (ldm_patched.modules.model_patcher.ModelPatcher(model, load_device=model_management.get_torch_device(), offload_device=offload_device), clip, vae)
 
-def load_checkpoint_guess_config(ckpt_path, output_vae=True, output_clip=True, output_clipvision=False, embedding_directory=None, output_model=True):
+def load_checkpoint_guess_config(ckpt_path, output_vae=True, output_clip=True, output_clipvision=False, embedding_directory=None, output_model=True, vae_filename_param=None):
     sd = ldm_patched.modules.utils.load_torch_file(ckpt_path)
     sd_keys = sd.keys()
     clip = None
     clipvision = None
     vae = None
+    vae_filename = None
     model = None
     model_patcher = None
     clip_target = None
@@ -462,8 +463,12 @@ def load_checkpoint_guess_config(ckpt_path, output_vae=True, output_clip=True, o
         model.load_model_weights(sd, "model.diffusion_model.")
 
     if output_vae:
-        vae_sd = ldm_patched.modules.utils.state_dict_prefix_replace(sd, {"first_stage_model.": ""}, filter_keys=True)
+        if vae_filename_param is None:
-        vae_sd = model_config.process_vae_state_dict(vae_sd)
+            vae_sd = ldm_patched.modules.utils.state_dict_prefix_replace(sd, {"first_stage_model.": ""}, filter_keys=True)
+            vae_sd = model_config.process_vae_state_dict(vae_sd)
+        else:
+            vae_sd = ldm_patched.modules.utils.load_torch_file(vae_filename_param)
+            vae_filename = vae_filename_param
         vae = VAE(sd=vae_sd)
 
     if output_clip:
@@ -485,7 +490,7 @@ def load_checkpoint_guess_config(ckpt_path, output_vae=True, output_clip=True, o
             print("loaded straight to GPU")
             model_management.load_model_gpu(model_patcher)
 
-    return (model_patcher, clip, vae, clipvision)
+    return model_patcher, clip, vae, vae_filename, clipvision
 
 
 def load_unet_state_dict(sd): #load unet in diffusers format

modules/async_worker.py

@@ -166,6 +166,7 @@ def worker():
         adaptive_cfg = args.pop()
         sampler_name = args.pop()
         scheduler_name = args.pop()
+        vae_name = args.pop()
         overwrite_step = args.pop()
         overwrite_switch = args.pop()
         overwrite_width = args.pop()
@@ -428,7 +429,7 @@ def worker():
             progressbar(async_task, 3, 'Loading models ...')
             pipeline.refresh_everything(refiner_model_name=refiner_model_name, base_model_name=base_model_name,
                                         loras=loras, base_model_additional_loras=base_model_additional_loras,
-                                        use_synthetic_refiner=use_synthetic_refiner)
+                                        use_synthetic_refiner=use_synthetic_refiner, vae_name=vae_name)
 
             progressbar(async_task, 3, 'Processing prompts ...')
             tasks = []
@@ -614,12 +615,12 @@ def worker():
 
                 H, W, C = inpaint_image.shape
                 if 'left' in outpaint_selections:
-                    inpaint_image = np.pad(inpaint_image, [[0, 0], [int(H * 0.3), 0], [0, 0]], mode='edge')
+                    inpaint_image = np.pad(inpaint_image, [[0, 0], [int(W * 0.3), 0], [0, 0]], mode='edge')
-                    inpaint_mask = np.pad(inpaint_mask, [[0, 0], [int(H * 0.3), 0]], mode='constant',
+                    inpaint_mask = np.pad(inpaint_mask, [[0, 0], [int(W * 0.3), 0]], mode='constant',
                                           constant_values=255)
                 if 'right' in outpaint_selections:
-                    inpaint_image = np.pad(inpaint_image, [[0, 0], [0, int(H * 0.3)], [0, 0]], mode='edge')
+                    inpaint_image = np.pad(inpaint_image, [[0, 0], [0, int(W * 0.3)], [0, 0]], mode='edge')
-                    inpaint_mask = np.pad(inpaint_mask, [[0, 0], [0, int(H * 0.3)]], mode='constant',
+                    inpaint_mask = np.pad(inpaint_mask, [[0, 0], [0, int(W * 0.3)]], mode='constant',
                                           constant_values=255)
 
                 inpaint_image = np.ascontiguousarray(inpaint_image.copy())
@@ -869,6 +870,7 @@ def worker():
 
                     d.append(('Sampler', 'sampler', sampler_name))
                     d.append(('Scheduler', 'scheduler', scheduler_name))
+                    d.append(('VAE', 'vae', vae_name))
                     d.append(('Seed', 'seed', str(task['task_seed'])))
 
                     if freeu_enabled:
@@ -883,7 +885,7 @@ def worker():
                         metadata_parser = modules.meta_parser.get_metadata_parser(metadata_scheme)
                         metadata_parser.set_data(task['log_positive_prompt'], task['positive'],
                                                  task['log_negative_prompt'], task['negative'],
-                                                 steps, base_model_name, refiner_model_name, loras)
+                                                 steps, base_model_name, refiner_model_name, loras, vae_name)
                     d.append(('Metadata Scheme', 'metadata_scheme', metadata_scheme.value if save_metadata_to_images else save_metadata_to_images))
                     d.append(('Version', 'version', 'Fooocus v' + fooocus_version.version))
                     img_paths.append(log(x, d, metadata_parser, output_format))

modules/config.py

@@ -124,14 +124,6 @@ def try_get_preset_content(preset):
             print(e)
     return {}
 
-
-try:
-    with open(os.path.abspath(f'./presets/default.json'), "r", encoding="utf-8") as json_file:
-        config_dict.update(json.load(json_file))
-except Exception as e:
-    print(f'Load default preset failed.')
-    print(e)
-
 available_presets = get_presets()
 preset = args_manager.args.preset
 config_dict.update(try_get_preset_content(preset))
@@ -197,6 +189,7 @@ paths_checkpoints = get_dir_or_set_default('path_checkpoints', ['../models/check
 paths_loras = get_dir_or_set_default('path_loras', ['../models/loras/'], True)
 path_embeddings = get_dir_or_set_default('path_embeddings', '../models/embeddings/')
 path_vae_approx = get_dir_or_set_default('path_vae_approx', '../models/vae_approx/')
+path_vae = get_dir_or_set_default('path_vae', '../models/vae/')
 path_upscale_models = get_dir_or_set_default('path_upscale_models', '../models/upscale_models/')
 path_inpaint = get_dir_or_set_default('path_inpaint', '../models/inpaint/')
 path_controlnet = get_dir_or_set_default('path_controlnet', '../models/controlnet/')
@@ -323,8 +316,12 @@ default_loras = get_config_item_or_set_default(
             1.0
         ]
     ],
-    validator=lambda x: isinstance(x, list) and all(len(y) == 3 and isinstance(y[0], bool) and isinstance(y[1], str) and isinstance(y[2], numbers.Number) for y in x)
+    validator=lambda x: isinstance(x, list) and all(
+        len(y) == 3 and isinstance(y[0], bool) and isinstance(y[1], str) and isinstance(y[2], numbers.Number)
+        or len(y) == 2 and isinstance(y[0], str) and isinstance(y[1], numbers.Number)
+        for y in x)
 )
+default_loras = [(y[0], y[1], y[2]) if len(y) == 3 else (True, y[0], y[1]) for y in default_loras]
 default_max_lora_number = get_config_item_or_set_default(
     key='default_max_lora_number',
     default_value=len(default_loras) if isinstance(default_loras, list) and len(default_loras) > 0 else 5,
@@ -350,6 +347,11 @@ default_scheduler = get_config_item_or_set_default(
     default_value='karras',
     validator=lambda x: x in modules.flags.scheduler_list
 )
+default_vae = get_config_item_or_set_default(
+    key='default_vae',
+    default_value=modules.flags.default_vae,
+    validator=lambda x: isinstance(x, str)
+)
 default_styles = get_config_item_or_set_default(
     key='default_styles',
     default_value=[
@@ -472,7 +474,7 @@ metadata_created_by = get_config_item_or_set_default(
 
 example_inpaint_prompts = [[x] for x in example_inpaint_prompts]
 
-config_dict["default_loras"] = default_loras = default_loras[:default_max_lora_number] + [['None', 1.0] for _ in range(default_max_lora_number - len(default_loras))]
+config_dict["default_loras"] = default_loras = default_loras[:default_max_lora_number] + [[True, 'None', 1.0] for _ in range(default_max_lora_number - len(default_loras))]
 
 # mapping config to meta parameter 
 possible_preset_keys = {
@@ -489,6 +491,7 @@ possible_preset_keys = {
     "default_scheduler": "scheduler",
     "default_overwrite_step": "steps",
     "default_performance": "performance",
+    "default_image_number": "image_number",
     "default_prompt": "prompt",
     "default_prompt_negative": "negative_prompt",
     "default_styles": "styles",
@@ -538,25 +541,32 @@ with open(config_example_path, "w", encoding="utf-8") as json_file:
 
 model_filenames = []
 lora_filenames = []
+vae_filenames = []
 wildcard_filenames = []
 
 sdxl_lcm_lora = 'sdxl_lcm_lora.safetensors'
 sdxl_lightning_lora = 'sdxl_lightning_4step_lora.safetensors'
+loras_metadata_remove = [sdxl_lcm_lora, sdxl_lightning_lora]
 
 
 def get_model_filenames(folder_paths, extensions=None, name_filter=None):
     if extensions is None:
         extensions = ['.pth', '.ckpt', '.bin', '.safetensors', '.fooocus.patch']
     files = []
+
+    if not isinstance(folder_paths, list):
+        folder_paths = [folder_paths]
     for folder in folder_paths:
         files += get_files_from_folder(folder, extensions, name_filter)
+
     return files
 
 
 def update_files():
-    global model_filenames, lora_filenames, wildcard_filenames, available_presets
+    global model_filenames, lora_filenames, vae_filenames, wildcard_filenames, available_presets
     model_filenames = get_model_filenames(paths_checkpoints)
     lora_filenames = get_model_filenames(paths_loras)
+    vae_filenames = get_model_filenames(path_vae)
     wildcard_filenames = get_files_from_folder(path_wildcards, ['.txt'])
     available_presets = get_presets()
     return

modules/core.py

@@ -35,12 +35,13 @@ opModelSamplingDiscrete = ModelSamplingDiscrete()
 
 
 class StableDiffusionModel:
-    def __init__(self, unet=None, vae=None, clip=None, clip_vision=None, filename=None):
+    def __init__(self, unet=None, vae=None, clip=None, clip_vision=None, filename=None, vae_filename=None):
         self.unet = unet
         self.vae = vae
         self.clip = clip
         self.clip_vision = clip_vision
         self.filename = filename
+        self.vae_filename = vae_filename
         self.unet_with_lora = unet
         self.clip_with_lora = clip
         self.visited_loras = ''
@@ -142,9 +143,10 @@ def apply_controlnet(positive, negative, control_net, image, strength, start_per
 
 @torch.no_grad()
 @torch.inference_mode()
-def load_model(ckpt_filename):
+def load_model(ckpt_filename, vae_filename=None):
-    unet, clip, vae, clip_vision = load_checkpoint_guess_config(ckpt_filename, embedding_directory=path_embeddings)
+    unet, clip, vae, vae_filename, clip_vision = load_checkpoint_guess_config(ckpt_filename, embedding_directory=path_embeddings,
-    return StableDiffusionModel(unet=unet, clip=clip, vae=vae, clip_vision=clip_vision, filename=ckpt_filename)
+                                                                vae_filename_param=vae_filename)
+    return StableDiffusionModel(unet=unet, clip=clip, vae=vae, clip_vision=clip_vision, filename=ckpt_filename, vae_filename=vae_filename)
 
 
 @torch.no_grad()

modules/default_pipeline.py

@@ -3,6 +3,7 @@ import os
 import torch
 import modules.patch
 import modules.config
+import modules.flags
 import ldm_patched.modules.model_management
 import ldm_patched.modules.latent_formats
 import modules.inpaint_worker
@@ -58,17 +59,21 @@ def assert_model_integrity():
 
 @torch.no_grad()
 @torch.inference_mode()
-def refresh_base_model(name):
+def refresh_base_model(name, vae_name=None):
     global model_base
 
     filename = get_file_from_folder_list(name, modules.config.paths_checkpoints)
 
-    if model_base.filename == filename:
+    vae_filename = None
+    if vae_name is not None and vae_name != modules.flags.default_vae:
+        vae_filename = get_file_from_folder_list(vae_name, modules.config.path_vae)
+
+    if model_base.filename == filename and model_base.vae_filename == vae_filename:
         return
 
-    model_base = core.StableDiffusionModel()
+    model_base = core.load_model(filename, vae_filename)
-    model_base = core.load_model(filename)
     print(f'Base model loaded: {model_base.filename}')
+    print(f'VAE loaded: {model_base.vae_filename}')
     return
 
 
@@ -216,7 +221,7 @@ def prepare_text_encoder(async_call=True):
 @torch.no_grad()
 @torch.inference_mode()
 def refresh_everything(refiner_model_name, base_model_name, loras,
-                       base_model_additional_loras=None, use_synthetic_refiner=False):
+                       base_model_additional_loras=None, use_synthetic_refiner=False, vae_name=None):
     global final_unet, final_clip, final_vae, final_refiner_unet, final_refiner_vae, final_expansion
 
     final_unet = None
@@ -227,11 +232,11 @@ def refresh_everything(refiner_model_name, base_model_name, loras,
 
     if use_synthetic_refiner and refiner_model_name == 'None':
         print('Synthetic Refiner Activated')
-        refresh_base_model(base_model_name)
+        refresh_base_model(base_model_name, vae_name)
         synthesize_refiner_model()
     else:
         refresh_refiner_model(refiner_model_name)
-        refresh_base_model(base_model_name)
+        refresh_base_model(base_model_name, vae_name)
 
     refresh_loras(loras, base_model_additional_loras=base_model_additional_loras)
     assert_model_integrity()
@@ -254,7 +259,8 @@ def refresh_everything(refiner_model_name, base_model_name, loras,
 refresh_everything(
     refiner_model_name=modules.config.default_refiner_model_name,
     base_model_name=modules.config.default_base_model_name,
-    loras=get_enabled_loras(modules.config.default_loras)
+    loras=get_enabled_loras(modules.config.default_loras),
+    vae_name=modules.config.default_vae,
 )
 
 

modules/flags.py

@@ -53,6 +53,8 @@ SAMPLER_NAMES = KSAMPLER_NAMES + list(SAMPLER_EXTRA.keys())
 sampler_list = SAMPLER_NAMES
 scheduler_list = SCHEDULER_NAMES
 
+default_vae = 'Default (model)'
+
 refiner_swap_method = 'joint'
 
 cn_ip = "ImagePrompt"

modules/meta_parser.py

@@ -1,5 +1,4 @@
 import json
-import os
 import re
 from abc import ABC, abstractmethod
 from pathlib import Path
@@ -12,7 +11,7 @@ import modules.config
 import modules.sdxl_styles
 from modules.flags import MetadataScheme, Performance, Steps
 from modules.flags import SAMPLERS, CIVITAI_NO_KARRAS
-from modules.util import quote, unquote, extract_styles_from_prompt, is_json, get_file_from_folder_list, calculate_sha256
+from modules.util import quote, unquote, extract_styles_from_prompt, is_json, get_file_from_folder_list, sha256
 
 re_param_code = r'\s*(\w[\w \-/]+):\s*("(?:\\.|[^\\"])+"|[^,]*)(?:,|$)'
 re_param = re.compile(re_param_code)
@@ -27,8 +26,9 @@ def load_parameter_button_click(raw_metadata: dict | str, is_generating: bool):
         loaded_parameter_dict = json.loads(raw_metadata)
     assert isinstance(loaded_parameter_dict, dict)
 
-    results = [len(loaded_parameter_dict) > 0, 1]
+    results = [len(loaded_parameter_dict) > 0]
 
+    get_image_number('image_number', 'Image Number', loaded_parameter_dict, results)
     get_str('prompt', 'Prompt', loaded_parameter_dict, results)
     get_str('negative_prompt', 'Negative Prompt', loaded_parameter_dict, results)
     get_list('styles', 'Styles', loaded_parameter_dict, results)
@@ -46,6 +46,7 @@ def load_parameter_button_click(raw_metadata: dict | str, is_generating: bool):
     get_float('refiner_switch', 'Refiner Switch', loaded_parameter_dict, results)
     get_str('sampler', 'Sampler', loaded_parameter_dict, results)
     get_str('scheduler', 'Scheduler', loaded_parameter_dict, results)
+    get_str('vae', 'VAE', loaded_parameter_dict, results)
     get_seed('seed', 'Seed', loaded_parameter_dict, results)
 
     if is_generating:
@@ -92,13 +93,25 @@ def get_float(key: str, fallback: str | None, source_dict: dict, results: list,
         results.append(gr.update())
 
 
+def get_image_number(key: str, fallback: str | None, source_dict: dict, results: list, default=None):
+    try:
+        h = source_dict.get(key, source_dict.get(fallback, default))
+        assert h is not None
+        h = int(h)
+        h = min(h, modules.config.default_max_image_number)
+        results.append(h)
+    except:
+        results.append(1)
+
+
 def get_steps(key: str, fallback: str | None, source_dict: dict, results: list, default=None):
     try:
         h = source_dict.get(key, source_dict.get(fallback, default))
         assert h is not None
         h = int(h)
         # if not in steps or in steps and performance is not the same
-        if h not in iter(Steps) or Steps(h).name.casefold() != source_dict.get('performance', '').replace(' ', '_').casefold():
+        if h not in iter(Steps) or Steps(h).name.casefold() != source_dict.get('performance', '').replace(' ',
+                                                                                                          '_').casefold():
             results.append(h)
             return
         results.append(-1)
@@ -169,11 +182,20 @@ def get_freeu(key: str, fallback: str | None, source_dict: dict, results: list,
 
 def get_lora(key: str, fallback: str | None, source_dict: dict, results: list):
     try:
-        n, w = source_dict.get(key, source_dict.get(fallback)).split(' : ')
+        split_data = source_dict.get(key, source_dict.get(fallback)).split(' : ')
-        w = float(w)
+        enabled = True
-        results.append(True)
+        name = split_data[0]
-        results.append(n)
+        weight = split_data[1]
-        results.append(w)
+
+        if len(split_data) == 3:
+            enabled = split_data[0] == 'True'
+            name = split_data[1]
+            weight = split_data[2]
+
+        weight = float(weight)
+        results.append(enabled)
+        results.append(name)
+        results.append(weight)
     except:
         results.append(True)
         results.append('None')
@@ -183,7 +205,8 @@ def get_lora(key: str, fallback: str | None, source_dict: dict, results: list):
 def get_sha256(filepath):
     global hash_cache
     if filepath not in hash_cache:
-        hash_cache[filepath] = calculate_sha256(filepath)
+        # is_safetensors = os.path.splitext(filepath)[1].lower() == '.safetensors'
+        hash_cache[filepath] = sha256(filepath)
 
     return hash_cache[filepath]
 
@@ -210,7 +233,8 @@ def parse_meta_from_preset(preset_content):
                 height = height[:height.index(" ")]
             preset_prepared[meta_key] = (width, height)
         else:
-            preset_prepared[meta_key] = items[settings_key] if settings_key in items and items[settings_key] is not None else getattr(modules.config, settings_key)
+            preset_prepared[meta_key] = items[settings_key] if settings_key in items and items[
+                settings_key] is not None else getattr(modules.config, settings_key)
 
         if settings_key == "default_styles" or settings_key == "default_aspect_ratio":
             preset_prepared[meta_key] = str(preset_prepared[meta_key])
@@ -230,6 +254,7 @@ class MetadataParser(ABC):
         self.refiner_model_name: str = ''
         self.refiner_model_hash: str = ''
         self.loras: list = []
+        self.vae_name: str = ''
 
     @abstractmethod
     def get_scheme(self) -> MetadataScheme:
@@ -244,7 +269,7 @@ class MetadataParser(ABC):
         raise NotImplementedError
 
     def set_data(self, raw_prompt, full_prompt, raw_negative_prompt, full_negative_prompt, steps, base_model_name,
-                 refiner_model_name, loras):
+                 refiner_model_name, loras, vae_name):
         self.raw_prompt = raw_prompt
         self.full_prompt = full_prompt
         self.raw_negative_prompt = raw_negative_prompt
@@ -266,6 +291,13 @@ class MetadataParser(ABC):
                 lora_path = get_file_from_folder_list(lora_name, modules.config.paths_loras)
                 lora_hash = get_sha256(lora_path)
                 self.loras.append((Path(lora_name).stem, lora_weight, lora_hash))
+        self.vae_name = Path(vae_name).stem
+
+    @staticmethod
+    def remove_special_loras(lora_filenames):
+        for lora_to_remove in modules.config.loras_metadata_remove:
+            if lora_to_remove in lora_filenames:
+                lora_filenames.remove(lora_to_remove)
 
 
 class A1111MetadataParser(MetadataParser):
@@ -281,6 +313,7 @@ class A1111MetadataParser(MetadataParser):
         'steps': 'Steps',
         'sampler': 'Sampler',
         'scheduler': 'Scheduler',
+        'vae': 'VAE',
         'guidance_scale': 'CFG scale',
         'seed': 'Seed',
         'resolution': 'Size',
@@ -368,20 +401,26 @@ class A1111MetadataParser(MetadataParser):
                     data['sampler'] = k
                     break
 
-        for key in ['base_model', 'refiner_model']:
+        for key in ['base_model', 'refiner_model', 'vae']:
             if key in data:
-                for filename in modules.config.model_filenames:
+                if key == 'vae':
-                    path = Path(filename)
+                    self.add_extension_to_filename(data, modules.config.vae_filenames, 'vae')
-                    if data[key] == path.stem:
+                else:
-                        data[key] = filename
+                    self.add_extension_to_filename(data, modules.config.model_filenames, key)
-                        break
 
-        if 'lora_hashes' in data:
+        lora_data = ''
+        if 'lora_weights' in data and data['lora_weights'] != '':
+            lora_data = data['lora_weights']
+        elif 'lora_hashes' in data and data['lora_hashes'] != '' and data['lora_hashes'].split(', ')[0].count(':') == 2:
+            lora_data = data['lora_hashes']
+
+        if lora_data != '':
             lora_filenames = modules.config.lora_filenames.copy()
-            if modules.config.sdxl_lcm_lora in lora_filenames:
+            self.remove_special_loras(lora_filenames)
-                lora_filenames.remove(modules.config.sdxl_lcm_lora)
+            for li, lora in enumerate(lora_data.split(', ')):
-            for li, lora in enumerate(data['lora_hashes'].split(', ')):
+                lora_split = lora.split(': ')
-                lora_name, lora_hash, lora_weight = lora.split(': ')
+                lora_name = lora_split[0]
+                lora_weight = lora_split[2] if len(lora_split) == 3 else lora_split[1]
                 for filename in lora_filenames:
                     path = Path(filename)
                     if lora_name == path.stem:
@@ -397,6 +436,7 @@ class A1111MetadataParser(MetadataParser):
 
         sampler = data['sampler']
         scheduler = data['scheduler']
+
         if sampler in SAMPLERS and SAMPLERS[sampler] != '':
             sampler = SAMPLERS[sampler]
             if sampler not in CIVITAI_NO_KARRAS and scheduler == 'karras':
@@ -415,6 +455,7 @@ class A1111MetadataParser(MetadataParser):
 
             self.fooocus_to_a1111['performance']: data['performance'],
             self.fooocus_to_a1111['scheduler']: scheduler,
+            self.fooocus_to_a1111['vae']: Path(data['vae']).stem,
             # workaround for multiline prompts
             self.fooocus_to_a1111['raw_prompt']: self.raw_prompt,
             self.fooocus_to_a1111['raw_negative_prompt']: self.raw_negative_prompt,
@@ -430,16 +471,19 @@ class A1111MetadataParser(MetadataParser):
             if key in data:
                 generation_params[self.fooocus_to_a1111[key]] = data[key]
 
-        lora_hashes = []
+        if len(self.loras) > 0:
-        for index, (lora_name, lora_weight, lora_hash) in enumerate(self.loras):
+            lora_hashes = []
-            # workaround for Fooocus not knowing LoRA name in LoRA metadata
+            lora_weights = []
-            lora_hashes.append(f'{lora_name}: {lora_hash}: {lora_weight}')
+            for index, (lora_name, lora_weight, lora_hash) in enumerate(self.loras):
-        lora_hashes_string = ', '.join(lora_hashes)
+                # workaround for Fooocus not knowing LoRA name in LoRA metadata
+                lora_hashes.append(f'{lora_name}: {lora_hash}')
+                lora_weights.append(f'{lora_name}: {lora_weight}')
+            lora_hashes_string = ', '.join(lora_hashes)
+            lora_weights_string = ', '.join(lora_weights)
+            generation_params[self.fooocus_to_a1111['lora_hashes']] = lora_hashes_string
+            generation_params[self.fooocus_to_a1111['lora_weights']] = lora_weights_string
 
-        generation_params |= {
+        generation_params[self.fooocus_to_a1111['version']] = data['version']
-            self.fooocus_to_a1111['lora_hashes']: lora_hashes_string,
-            self.fooocus_to_a1111['version']: data['version']
-        }
 
         if modules.config.metadata_created_by != '':
             generation_params[self.fooocus_to_a1111['created_by']] = modules.config.metadata_created_by
@@ -452,6 +496,14 @@ class A1111MetadataParser(MetadataParser):
         negative_prompt_text = f"\nNegative prompt: {negative_prompt_resolved}" if negative_prompt_resolved else ""
         return f"{positive_prompt_resolved}{negative_prompt_text}\n{generation_params_text}".strip()
 
+    @staticmethod
+    def add_extension_to_filename(data, filenames, key):
+        for filename in filenames:
+            path = Path(filename)
+            if data[key] == path.stem:
+                data[key] = filename
+                break
+
 
 class FooocusMetadataParser(MetadataParser):
     def get_scheme(self) -> MetadataScheme:
@@ -460,9 +512,8 @@ class FooocusMetadataParser(MetadataParser):
     def parse_json(self, metadata: dict) -> dict:
         model_filenames = modules.config.model_filenames.copy()
         lora_filenames = modules.config.lora_filenames.copy()
-        if modules.config.sdxl_lcm_lora in lora_filenames:
+        vae_filenames = modules.config.vae_filenames.copy()
-            lora_filenames.remove(modules.config.sdxl_lcm_lora)
+        self.remove_special_loras(lora_filenames)
-
         for key, value in metadata.items():
             if value in ['', 'None']:
                 continue
@@ -470,6 +521,8 @@ class FooocusMetadataParser(MetadataParser):
                 metadata[key] = self.replace_value_with_filename(key, value, model_filenames)
             elif key.startswith('lora_combined_'):
                 metadata[key] = self.replace_value_with_filename(key, value, lora_filenames)
+            elif key == 'vae':
+                metadata[key] = self.replace_value_with_filename(key, value, vae_filenames)
             else:
                 continue
 
@@ -496,6 +549,7 @@ class FooocusMetadataParser(MetadataParser):
             res['refiner_model'] = self.refiner_model_name
             res['refiner_model_hash'] = self.refiner_model_hash
 
+        res['vae'] = self.vae_name
         res['loras'] = self.loras
 
         if modules.config.metadata_created_by != '':

modules/util.py

@@ -7,9 +7,9 @@ import math
 import os
 import cv2
 import json
+import hashlib
 
 from PIL import Image
-from hashlib import sha256
 
 import modules.sdxl_styles
 
@@ -182,16 +182,44 @@ def get_files_from_folder(folder_path, extensions=None, name_filter=None):
     return filenames
 
 
-def calculate_sha256(filename, length=HASH_SHA256_LENGTH) -> str:
+def sha256(filename, use_addnet_hash=False, length=HASH_SHA256_LENGTH):
-    hash_sha256 = sha256()
+    print(f"Calculating sha256 for {filename}: ", end='')
+    if use_addnet_hash:
+        with open(filename, "rb") as file:
+            sha256_value = addnet_hash_safetensors(file)
+    else:
+        sha256_value = calculate_sha256(filename)
+    print(f"{sha256_value}")
+
+    return sha256_value[:length] if length is not None else sha256_value
+
+
+def addnet_hash_safetensors(b):
+    """kohya-ss hash for safetensors from https://github.com/kohya-ss/sd-scripts/blob/main/library/train_util.py"""
+    hash_sha256 = hashlib.sha256()
+    blksize = 1024 * 1024
+
+    b.seek(0)
+    header = b.read(8)
+    n = int.from_bytes(header, "little")
+
+    offset = n + 8
+    b.seek(offset)
+    for chunk in iter(lambda: b.read(blksize), b""):
+        hash_sha256.update(chunk)
+
+    return hash_sha256.hexdigest()
+
+
+def calculate_sha256(filename) -> str:
+    hash_sha256 = hashlib.sha256()
     blksize = 1024 * 1024
 
     with open(filename, "rb") as f:
         for chunk in iter(lambda: f.read(blksize), b""):
             hash_sha256.update(chunk)
 
-    res = hash_sha256.hexdigest()
+    return hash_sha256.hexdigest()
-    return res[:length] if length else res
 
 
 def quote(text):
@@ -343,6 +371,9 @@ def is_json(data: str) -> bool:
 
 
 def get_file_from_folder_list(name, folders):
+    if not isinstance(folders, list):
+        folders = [folders]
+
     for folder in folders:
         filename = os.path.abspath(os.path.realpath(os.path.join(folder, name)))
         if os.path.isfile(filename):

presets/anime.json

@@ -34,7 +34,7 @@
     "default_sampler": "dpmpp_2m_sde_gpu",
     "default_scheduler": "karras",
     "default_performance": "Speed",
-    "default_prompt": "1girl, ",
+    "default_prompt": "",
     "default_prompt_negative": "",
     "default_styles": [
         "Fooocus V2",

readme.md

@@ -84,6 +84,10 @@ The first time you launch the software, it will automatically download models:
 
 After Fooocus 2.1.60, you will also have `run_anime.bat` and `run_realistic.bat`. They are different model presets (and require different models, but they will be automatically downloaded). [Check here for more details](https://github.com/lllyasviel/Fooocus/discussions/679).
 
+After Fooocus 2.3.0 you can also switch presets directly in the browser. Keep in mind to add these arguments if you want to change the default behavior:
+* Use `--disable-preset-selection` to disable preset selection in the browser.
+* Use `--always-download-new-model` to download missing models on preset switch. Default is fallback to `previous_default_models` defined in the corresponding preset, also see terminal output.
+
 ![image](https://github.com/lllyasviel/Fooocus/assets/19834515/d386f817-4bd7-490c-ad89-c1e228c23447)
 
 If you already have these files, you can copy them to the above locations to speed up installation.
@@ -115,7 +119,7 @@ See also the common problems and troubleshoots [here](troubleshoot.md).
 
 ### Colab
 
-(Last tested - 2024 Mar 11)
+(Last tested - 2024 Mar 18 by [mashb1t](https://github.com/mashb1t))
 
 | Colab | Info
 | --- | --- |
@@ -123,11 +127,13 @@ See also the common problems and troubleshoots [here](troubleshoot.md).
 
 In Colab, you can modify the last line to `!python entry_with_update.py --share --always-high-vram` or `!python entry_with_update.py --share --always-high-vram --preset anime` or `!python entry_with_update.py --share --always-high-vram --preset realistic` for Fooocus Default/Anime/Realistic Edition.
 
+You can also change the preset in the UI. Please be aware that this may lead to timeouts after 60 seconds. If this is the case, please wait until the download has finished, change the preset to initial and back to the one you've selected or reload the page.
+
 Note that this Colab will disable refiner by default because Colab free's resources are relatively limited (and some "big" features like image prompt may cause free-tier Colab to disconnect). We make sure that basic text-to-image is always working on free-tier Colab.
 
-Using `--always-high-vram` shifts resource allocation from RAM to VRAM and achieves the overall best balance between performance, flexibility and stability on the default T4 instance.
+Using `--always-high-vram` shifts resource allocation from RAM to VRAM and achieves the overall best balance between performance, flexibility and stability on the default T4 instance. Please find more information [here](https://github.com/lllyasviel/Fooocus/pull/1710#issuecomment-1989185346).
 
-Thanks to [camenduru](https://github.com/camenduru)!
+Thanks to [camenduru](https://github.com/camenduru) for the template!
 
 ### Linux (Using Anaconda)
 

update_log.md

@@ -1,3 +1,21 @@
+# [2.3.1](https://github.com/lllyasviel/Fooocus/releases/tag/2.3.1)
+
+* Remove positive prompt from anime prefix to not reset prompt after switching presets
+* Fix image number being reset to 1 when switching preset, now doesn't reset anymore
+* Fix outpainting dimension calculation when extending left/right
+* Fix LoRA compatibility for LoRAs in a1111 metadata scheme
+
+# [2.3.0](https://github.com/lllyasviel/Fooocus/releases/tag/2.3.0)
+
+* Add performance "lightning" (based on [SDXL-Lightning 4 step LoRA](https://huggingface.co/ByteDance/SDXL-Lightning/blob/main/sdxl_lightning_4step_lora.safetensors))
+* Add preset selection to UI, disable with argument `--disable-preset-selection`. Use `--always-download-new-model` to download missing models on preset switch.
+* Improve face swap consistency by switching later in the process to (synthetic) refiner
+* Add temp path cleanup on startup
+* Add support for wildcard subdirectories
+* Add scrollable 2 column layout for styles for better structure
+* Improve Colab resource needs for T4 instances (default), positively tested with all image prompt features
+* Improve anime preset, now uses style `Fooocus Semi Realistic` instead of `Fooocus Negative` (less wet look images)
+
 # [2.2.1](https://github.com/lllyasviel/Fooocus/releases/tag/2.2.1)
 
 * Fix some small bugs (e.g. image grid, upscale fast 2x, LoRA weight width in Firefox)

webui.py

@@ -406,6 +406,8 @@ with shared.gradio_root:
                                                    value=modules.config.default_sampler)
                         scheduler_name = gr.Dropdown(label='Scheduler', choices=flags.scheduler_list,
                                                      value=modules.config.default_scheduler)
+                        vae_name = gr.Dropdown(label='VAE', choices=[modules.flags.default_vae] + modules.config.vae_filenames,
+                                                     value=modules.config.default_vae, show_label=True)
 
                         generate_image_grid = gr.Checkbox(label='Generate Image Grid for Each Batch',
                                                           info='(Experimental) This may cause performance problems on some computers and certain internet conditions.',
@@ -528,6 +530,7 @@ with shared.gradio_root:
                     modules.config.update_files()
                     results = [gr.update(choices=modules.config.model_filenames)]
                     results += [gr.update(choices=['None'] + modules.config.model_filenames)]
+                    results += [gr.update(choices=['None'] + modules.config.vae_filenames)]
                     if not args_manager.args.disable_preset_selection:
                         results += [gr.update(choices=modules.config.available_presets)]
                     for i in range(modules.config.default_max_lora_number):
@@ -535,7 +538,7 @@ with shared.gradio_root:
                                     gr.update(choices=['None'] + modules.config.lora_filenames), gr.update()]
                     return results
 
-                refresh_files_output = [base_model, refiner_model]
+                refresh_files_output = [base_model, refiner_model, vae_name]
                 if not args_manager.args.disable_preset_selection:
                     refresh_files_output += [preset_selection]
                 refresh_files.click(refresh_files_clicked, [], refresh_files_output + lora_ctrls,
@@ -547,8 +550,8 @@ with shared.gradio_root:
                              performance_selection, overwrite_step, overwrite_switch, aspect_ratios_selection,
                              overwrite_width, overwrite_height, guidance_scale, sharpness, adm_scaler_positive,
                              adm_scaler_negative, adm_scaler_end, refiner_swap_method, adaptive_cfg, base_model,
-                             refiner_model, refiner_switch, sampler_name, scheduler_name, seed_random, image_seed,
+                             refiner_model, refiner_switch, sampler_name, scheduler_name, vae_name, seed_random,
-                             generate_button, load_parameter_button] + freeu_ctrls + lora_ctrls
+                             image_seed, generate_button, load_parameter_button] + freeu_ctrls + lora_ctrls
 
         if not args_manager.args.disable_preset_selection:
             def preset_selection_change(preset, is_generating):
@@ -634,7 +637,7 @@ with shared.gradio_root:
         ctrls += [outpaint_selections, inpaint_input_image, inpaint_additional_prompt, inpaint_mask_image]
         ctrls += [disable_preview, disable_intermediate_results, disable_seed_increment]
         ctrls += [adm_scaler_positive, adm_scaler_negative, adm_scaler_end, adaptive_cfg]
-        ctrls += [sampler_name, scheduler_name]
+        ctrls += [sampler_name, scheduler_name, vae_name]
         ctrls += [overwrite_step, overwrite_switch, overwrite_width, overwrite_height, overwrite_vary_strength]
         ctrls += [overwrite_upscale_strength, mixing_image_prompt_and_vary_upscale, mixing_image_prompt_and_inpaint]
         ctrls += [debugging_cn_preprocessor, skipping_cn_preprocessor, canny_low_threshold, canny_high_threshold]