Merge branch 'AUTOMATIC1111:master' into master

modules/api/api.py

@@ -1,11 +1,14 @@
+import time
 import uvicorn
 from gradio.processing_utils import encode_pil_to_base64, decode_base64_to_file, decode_base64_to_image
-from fastapi import APIRouter, HTTPException
+from fastapi import APIRouter, Depends, HTTPException
 import modules.shared as shared
+from modules import devices
 from modules.api.models import *
 from modules.processing import StableDiffusionProcessingTxt2Img, StableDiffusionProcessingImg2Img, process_images
 from modules.sd_samplers import all_samplers
-from modules.extras import run_extras
+from modules.extras import run_extras, run_pnginfo
+
 
 def upscaler_to_index(name: str):
     try:
@@ -13,8 +16,10 @@ def upscaler_to_index(name: str):
     except:
         raise HTTPException(status_code=400, detail=f"Invalid upscaler, needs to be on of these: {' , '.join([x.name for x in sd_upscalers])}")
 
+
 sampler_to_index = lambda name: next(filter(lambda row: name.lower() == row[1].name.lower(), enumerate(all_samplers)), None)
 
+
 def setUpscalers(req: dict):
     reqDict = vars(req)
     reqDict['extras_upscaler_1'] = upscaler_to_index(req.upscaler_1)
@@ -23,6 +28,7 @@ def setUpscalers(req: dict):
     reqDict.pop('upscaler_2')
     return reqDict
 
+
 class Api:
     def __init__(self, app, queue_lock):
         self.router = APIRouter()
@@ -32,15 +38,17 @@ class Api:
         self.app.add_api_route("/sdapi/v1/img2img", self.img2imgapi, methods=["POST"], response_model=ImageToImageResponse)
         self.app.add_api_route("/sdapi/v1/extra-single-image", self.extras_single_image_api, methods=["POST"], response_model=ExtrasSingleImageResponse)
         self.app.add_api_route("/sdapi/v1/extra-batch-images", self.extras_batch_images_api, methods=["POST"], response_model=ExtrasBatchImagesResponse)
+        self.app.add_api_route("/sdapi/v1/png-info", self.pnginfoapi, methods=["POST"], response_model=PNGInfoResponse)
+        self.app.add_api_route("/sdapi/v1/progress", self.progressapi, methods=["GET"], response_model=ProgressResponse)
 
     def text2imgapi(self, txt2imgreq: StableDiffusionTxt2ImgProcessingAPI):
         sampler_index = sampler_to_index(txt2imgreq.sampler_index)
-        
+
         if sampler_index is None:
-            raise HTTPException(status_code=404, detail="Sampler not found") 
-        
+            raise HTTPException(status_code=404, detail="Sampler not found")
+
         populate = txt2imgreq.copy(update={ # Override __init__ params
-            "sd_model": shared.sd_model, 
+            "sd_model": shared.sd_model,
             "sampler_index": sampler_index[0],
             "do_not_save_samples": True,
             "do_not_save_grid": True
@@ -48,34 +56,39 @@ class Api:
         )
         p = StableDiffusionProcessingTxt2Img(**vars(populate))
         # Override object param
+
+        shared.state.begin()
+
         with self.queue_lock:
             processed = process_images(p)
-        
+
+        shared.state.end()
+
         b64images = list(map(encode_pil_to_base64, processed.images))
-        
+
         return TextToImageResponse(images=b64images, parameters=vars(txt2imgreq), info=processed.js())
 
     def img2imgapi(self, img2imgreq: StableDiffusionImg2ImgProcessingAPI):
         sampler_index = sampler_to_index(img2imgreq.sampler_index)
-        
+
         if sampler_index is None:
-            raise HTTPException(status_code=404, detail="Sampler not found") 
+            raise HTTPException(status_code=404, detail="Sampler not found")
 
 
         init_images = img2imgreq.init_images
         if init_images is None:
-            raise HTTPException(status_code=404, detail="Init image not found") 
+            raise HTTPException(status_code=404, detail="Init image not found")
 
         mask = img2imgreq.mask
         if mask:
             mask = decode_base64_to_image(mask)
 
-        
+
         populate = img2imgreq.copy(update={ # Override __init__ params
-            "sd_model": shared.sd_model, 
+            "sd_model": shared.sd_model,
             "sampler_index": sampler_index[0],
             "do_not_save_samples": True,
-            "do_not_save_grid": True, 
+            "do_not_save_grid": True,
             "mask": mask
             }
         )
@@ -87,16 +100,20 @@ class Api:
             imgs = [img] * p.batch_size
 
         p.init_images = imgs
-        # Override object param
+
+        shared.state.begin()
+
         with self.queue_lock:
             processed = process_images(p)
-        
+
+        shared.state.end()
+
         b64images = list(map(encode_pil_to_base64, processed.images))
 
         if (not img2imgreq.include_init_images):
             img2imgreq.init_images = None
             img2imgreq.mask = None
-       
+
         return ImageToImageResponse(images=b64images, parameters=vars(img2imgreq), info=processed.js())
 
     def extras_single_image_api(self, req: ExtrasSingleImageRequest):
@@ -124,9 +141,40 @@ class Api:
             result = run_extras(extras_mode=1, image="", input_dir="", output_dir="", **reqDict)
 
         return ExtrasBatchImagesResponse(images=list(map(encode_pil_to_base64, result[0])), html_info=result[1])
-    
-    def pnginfoapi(self):
-        raise NotImplementedError
+
+    def pnginfoapi(self, req: PNGInfoRequest):
+        if(not req.image.strip()):
+            return PNGInfoResponse(info="")
+
+        result = run_pnginfo(decode_base64_to_image(req.image.strip()))
+
+        return PNGInfoResponse(info=result[1])
+
+    def progressapi(self, req: ProgressRequest = Depends()):
+        # copy from check_progress_call of ui.py
+
+        if shared.state.job_count == 0:
+            return ProgressResponse(progress=0, eta_relative=0, state=shared.state.dict())
+
+        # avoid dividing zero
+        progress = 0.01
+
+        if shared.state.job_count > 0:
+            progress += shared.state.job_no / shared.state.job_count
+        if shared.state.sampling_steps > 0:
+            progress += 1 / shared.state.job_count * shared.state.sampling_step / shared.state.sampling_steps
+
+        time_since_start = time.time() - shared.state.time_start
+        eta = (time_since_start/progress)
+        eta_relative = eta-time_since_start
+
+        progress = min(progress, 1)
+
+        current_image = None
+        if shared.state.current_image and not req.skip_current_image:
+            current_image = encode_pil_to_base64(shared.state.current_image)
+
+        return ProgressResponse(progress=progress, eta_relative=eta_relative, state=shared.state.dict(), current_image=current_image)
 
     def launch(self, server_name, port):
         self.app.include_router(self.router)

modules/api/models.py

@@ -1,4 +1,5 @@
 import inspect
+from click import prompt
 from pydantic import BaseModel, Field, create_model
 from typing import Any, Optional
 from typing_extensions import Literal
@@ -51,17 +52,17 @@ class PydanticModelGenerator:
             # field_type = str if not overrides.get(k) else overrides[k]["type"]
             # print(k, v.annotation, v.default)
             field_type = v.annotation
-            
+
             return Optional[field_type]
-        
+
         def merge_class_params(class_):
             all_classes = list(filter(lambda x: x is not object, inspect.getmro(class_)))
             parameters = {}
             for classes in all_classes:
                 parameters = {**parameters, **inspect.signature(classes.__init__).parameters}
             return parameters
-            
-                
+
+
         self._model_name = model_name
         self._class_data = merge_class_params(class_instance)
         self._model_def = [
@@ -73,11 +74,11 @@ class PydanticModelGenerator:
             )
             for (k,v) in self._class_data.items() if k not in API_NOT_ALLOWED
         ]
-        
+
         for fields in additional_fields:
             self._model_def.append(ModelDef(
-                field=underscore(fields["key"]), 
-                field_alias=fields["key"], 
+                field=underscore(fields["key"]),
+                field_alias=fields["key"],
                 field_type=fields["type"],
                 field_value=fields["default"],
                 field_exclude=fields["exclude"] if "exclude" in fields else False))
@@ -94,15 +95,15 @@ class PydanticModelGenerator:
         DynamicModel.__config__.allow_population_by_field_name = True
         DynamicModel.__config__.allow_mutation = True
         return DynamicModel
-    
+
 StableDiffusionTxt2ImgProcessingAPI = PydanticModelGenerator(
-    "StableDiffusionProcessingTxt2Img", 
+    "StableDiffusionProcessingTxt2Img",
     StableDiffusionProcessingTxt2Img,
     [{"key": "sampler_index", "type": str, "default": "Euler"}]
 ).generate_model()
 
 StableDiffusionImg2ImgProcessingAPI = PydanticModelGenerator(
-    "StableDiffusionProcessingImg2Img", 
+    "StableDiffusionProcessingImg2Img",
     StableDiffusionProcessingImg2Img,
     [{"key": "sampler_index", "type": str, "default": "Euler"}, {"key": "init_images", "type": list, "default": None}, {"key": "denoising_strength", "type": float, "default": 0.75}, {"key": "mask", "type": str, "default": None}, {"key": "include_init_images", "type": bool, "default": False, "exclude" : True}]
 ).generate_model()
@@ -148,4 +149,19 @@ class ExtrasBatchImagesRequest(ExtrasBaseRequest):
     imageList: list[FileData] = Field(title="Images", description="List of images to work on. Must be Base64 strings")
 
 class ExtrasBatchImagesResponse(ExtraBaseResponse):
-    images: list[str] = Field(title="Images", description="The generated images in base64 format.")
+    images: list[str] = Field(title="Images", description="The generated images in base64 format.")
+
+class PNGInfoRequest(BaseModel):
+    image: str = Field(title="Image", description="The base64 encoded PNG image")
+
+class PNGInfoResponse(BaseModel):
+    info: str = Field(title="Image info", description="A string with all the info the image had")
+
+class ProgressRequest(BaseModel):
+    skip_current_image: bool = Field(default=False, title="Skip current image", description="Skip current image serialization")
+
+class ProgressResponse(BaseModel):
+    progress: float = Field(title="Progress", description="The progress with a range of 0 to 1")
+    eta_relative: float = Field(title="ETA in secs")
+    state: dict = Field(title="State", description="The current state snapshot")
+    current_image: str = Field(default=None, title="Current image", description="The current image in base64 format. opts.show_progress_every_n_steps is required for this to work.")

modules/generation_parameters_copypaste.py

@@ -66,6 +66,7 @@ def integrate_settings_paste_fields(component_dict):
 
     settings_map = {
         'sd_hypernetwork': 'Hypernet',
+        'sd_hypernetwork_strength': 'Hypernet strength',
         'CLIP_stop_at_last_layers': 'Clip skip',
         'sd_model_checkpoint': 'Model hash',
     }

modules/hypernetworks/hypernetwork.py

@@ -209,13 +209,16 @@ def list_hypernetworks(path):
     res = {}
     for filename in glob.iglob(os.path.join(path, '**/*.pt'), recursive=True):
         name = os.path.splitext(os.path.basename(filename))[0]
-        res[name] = filename
+        # Prevent a hypothetical "None.pt" from being listed.
+        if name != "None":
+            res[name] = filename
     return res
 
 
 def load_hypernetwork(filename):
     path = shared.hypernetworks.get(filename, None)
-    if path is not None:
+    # Prevent any file named "None.pt" from being loaded.
+    if path is not None and filename != "None":
         print(f"Loading hypernetwork {filename}")
         try:
             shared.loaded_hypernetwork = Hypernetwork()
@@ -332,7 +335,9 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
     # images allows training previews to have infotext. Importing it at the top causes a circular import problem.
     from modules import images
 
-    assert hypernetwork_name, 'hypernetwork not selected'
+    save_hypernetwork_every = save_hypernetwork_every or 0
+    create_image_every = create_image_every or 0
+    textual_inversion.validate_train_inputs(hypernetwork_name, learn_rate, batch_size, data_root, template_file, steps, save_hypernetwork_every, create_image_every, log_directory, name="hypernetwork")
 
     path = shared.hypernetworks.get(hypernetwork_name, None)
     shared.loaded_hypernetwork = Hypernetwork()
@@ -358,39 +363,44 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
     else:
         images_dir = None
 
+    hypernetwork = shared.loaded_hypernetwork
+    checkpoint = sd_models.select_checkpoint()
+
+    ititial_step = hypernetwork.step or 0
+    if ititial_step >= steps:
+        shared.state.textinfo = f"Model has already been trained beyond specified max steps"
+        return hypernetwork, filename
+
+    scheduler = LearnRateScheduler(learn_rate, steps, ititial_step)
+    
+    # dataset loading may take a while, so input validations and early returns should be done before this
     shared.state.textinfo = f"Preparing dataset from {html.escape(data_root)}..."
     with torch.autocast("cuda"):
         ds = modules.textual_inversion.dataset.PersonalizedBase(data_root=data_root, width=training_width, height=training_height, repeats=shared.opts.training_image_repeats_per_epoch, placeholder_token=hypernetwork_name, model=shared.sd_model, device=devices.device, template_file=template_file, include_cond=True, batch_size=batch_size)
+
     if unload:
         shared.sd_model.cond_stage_model.to(devices.cpu)
         shared.sd_model.first_stage_model.to(devices.cpu)
 
-    hypernetwork = shared.loaded_hypernetwork
-    weights = hypernetwork.weights()
-    for weight in weights:
-        weight.requires_grad = True
-
     size = len(ds.indexes)
     loss_dict = defaultdict(lambda : deque(maxlen = 1024))
     losses = torch.zeros((size,))
     previous_mean_losses = [0]
     previous_mean_loss = 0
     print("Mean loss of {} elements".format(size))
-
-    last_saved_file = "<none>"
-    last_saved_image = "<none>"
-    forced_filename = "<none>"
-
-    ititial_step = hypernetwork.step or 0
-    if ititial_step > steps:
-        return hypernetwork, filename
-
-    scheduler = LearnRateScheduler(learn_rate, steps, ititial_step)
+    
+    weights = hypernetwork.weights()
+    for weight in weights:
+        weight.requires_grad = True
     # if optimizer == "AdamW": or else Adam / AdamW / SGD, etc...
     optimizer = torch.optim.AdamW(weights, lr=scheduler.learn_rate)
 
     steps_without_grad = 0
 
+    last_saved_file = "<none>"
+    last_saved_image = "<none>"
+    forced_filename = "<none>"
+
     pbar = tqdm.tqdm(enumerate(ds), total=steps - ititial_step)
     for i, entries in pbar:
         hypernetwork.step = i + ititial_step
@@ -443,9 +453,9 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
 
         if hypernetwork_dir is not None and steps_done % save_hypernetwork_every == 0:
             # Before saving, change name to match current checkpoint.
-            hypernetwork.name = f'{hypernetwork_name}-{steps_done}'
-            last_saved_file = os.path.join(hypernetwork_dir, f'{hypernetwork.name}.pt')
-            hypernetwork.save(last_saved_file)
+            hypernetwork_name_every = f'{hypernetwork_name}-{steps_done}'
+            last_saved_file = os.path.join(hypernetwork_dir, f'{hypernetwork_name_every}.pt')
+            save_hypernetwork(hypernetwork, checkpoint, hypernetwork_name, last_saved_file)
 
         textual_inversion.write_loss(log_directory, "hypernetwork_loss.csv", hypernetwork.step, len(ds), {
             "loss": f"{previous_mean_loss:.7f}",
@@ -506,13 +516,23 @@ Last saved image: {html.escape(last_saved_image)}<br/>
 """
         
     report_statistics(loss_dict)
-    checkpoint = sd_models.select_checkpoint()
 
-    hypernetwork.sd_checkpoint = checkpoint.hash
-    hypernetwork.sd_checkpoint_name = checkpoint.model_name
-    # Before saving for the last time, change name back to the base name (as opposed to the save_hypernetwork_every step-suffixed naming convention).
-    hypernetwork.name = hypernetwork_name
-    filename = os.path.join(shared.cmd_opts.hypernetwork_dir, f'{hypernetwork.name}.pt')
-    hypernetwork.save(filename)
+    filename = os.path.join(shared.cmd_opts.hypernetwork_dir, f'{hypernetwork_name}.pt')
+    save_hypernetwork(hypernetwork, checkpoint, hypernetwork_name, filename)
 
     return hypernetwork, filename
+
+def save_hypernetwork(hypernetwork, checkpoint, hypernetwork_name, filename):
+    old_hypernetwork_name = hypernetwork.name
+    old_sd_checkpoint = hypernetwork.sd_checkpoint if hasattr(hypernetwork, "sd_checkpoint") else None
+    old_sd_checkpoint_name = hypernetwork.sd_checkpoint_name if hasattr(hypernetwork, "sd_checkpoint_name") else None
+    try:
+        hypernetwork.sd_checkpoint = checkpoint.hash
+        hypernetwork.sd_checkpoint_name = checkpoint.model_name
+        hypernetwork.name = hypernetwork_name
+        hypernetwork.save(filename)
+    except:
+        hypernetwork.sd_checkpoint = old_sd_checkpoint
+        hypernetwork.sd_checkpoint_name = old_sd_checkpoint_name
+        hypernetwork.name = old_hypernetwork_name
+        raise

modules/processing.py

@@ -388,6 +388,7 @@ def create_infotext(p, all_prompts, all_seeds, all_subseeds, comments, iteration
         "Model hash": getattr(p, 'sd_model_hash', None if not opts.add_model_hash_to_info or not shared.sd_model.sd_model_hash else shared.sd_model.sd_model_hash),
         "Model": (None if not opts.add_model_name_to_info or not shared.sd_model.sd_checkpoint_info.model_name else shared.sd_model.sd_checkpoint_info.model_name.replace(',', '').replace(':', '')),
         "Hypernet": (None if shared.loaded_hypernetwork is None else shared.loaded_hypernetwork.name),
+        "Hypernet strength": (None if shared.loaded_hypernetwork is None or shared.opts.sd_hypernetwork_strength >= 1 else shared.opts.sd_hypernetwork_strength),
         "Batch size": (None if p.batch_size < 2 else p.batch_size),
         "Batch pos": (None if p.batch_size < 2 else position_in_batch),
         "Variation seed": (None if p.subseed_strength == 0 else all_subseeds[index]),
@@ -470,7 +471,7 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed:
         model_hijack.embedding_db.load_textual_inversion_embeddings()
 
     if p.scripts is not None:
-        p.scripts.run_alwayson_scripts(p)
+        p.scripts.process(p)
 
     infotexts = []
     output_images = []
@@ -493,7 +494,7 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed:
             seeds = p.all_seeds[n * p.batch_size:(n + 1) * p.batch_size]
             subseeds = p.all_subseeds[n * p.batch_size:(n + 1) * p.batch_size]
 
-            if (len(prompts) == 0):
+            if len(prompts) == 0:
                 break
 
             with devices.autocast():
@@ -582,7 +583,13 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed:
                 images.save_image(grid, p.outpath_grids, "grid", p.all_seeds[0], p.all_prompts[0], opts.grid_format, info=infotext(), short_filename=not opts.grid_extended_filename, p=p, grid=True)
 
     devices.torch_gc()
-    return Processed(p, output_images, p.all_seeds[0], infotext() + "".join(["\n\n" + x for x in comments]), subseed=p.all_subseeds[0], all_prompts=p.all_prompts, all_seeds=p.all_seeds, all_subseeds=p.all_subseeds, index_of_first_image=index_of_first_image, infotexts=infotexts)
+
+    res = Processed(p, output_images, p.all_seeds[0], infotext() + "".join(["\n\n" + x for x in comments]), subseed=p.all_subseeds[0], all_prompts=p.all_prompts, all_seeds=p.all_seeds, all_subseeds=p.all_subseeds, index_of_first_image=index_of_first_image, infotexts=infotexts)
+
+    if p.scripts is not None:
+        p.scripts.postprocess(p, res)
+
+    return res
 
 
 class StableDiffusionProcessingTxt2Img(StableDiffusionProcessing):
@@ -681,6 +688,8 @@ class StableDiffusionProcessingTxt2Img(StableDiffusionProcessing):
 
         noise = create_random_tensors(samples.shape[1:], seeds=seeds, subseeds=subseeds, subseed_strength=subseed_strength, seed_resize_from_h=self.seed_resize_from_h, seed_resize_from_w=self.seed_resize_from_w, p=self)
 
+        image_conditioning = self.txt2img_image_conditioning(x)
+
         # GC now before running the next img2img to prevent running out of memory
         x = None
         devices.torch_gc()

modules/scripts.py

@@ -64,7 +64,16 @@ class Script:
     def process(self, p, *args):
         """
         This function is called before processing begins for AlwaysVisible scripts.
-        scripts. You can modify the processing object (p) here, inject hooks, etc.
+        You can modify the processing object (p) here, inject hooks, etc.
+        args contains all values returned by components from ui()
+        """
+
+        pass
+
+    def postprocess(self, p, processed, *args):
+        """
+        This function is called after processing ends for AlwaysVisible scripts.
+        args contains all values returned by components from ui()
         """
 
         pass
@@ -289,13 +298,22 @@ class ScriptRunner:
 
         return processed
 
-    def run_alwayson_scripts(self, p):
+    def process(self, p):
         for script in self.alwayson_scripts:
             try:
                 script_args = p.script_args[script.args_from:script.args_to]
                 script.process(p, *script_args)
             except Exception:
-                print(f"Error running alwayson script: {script.filename}", file=sys.stderr)
+                print(f"Error running process: {script.filename}", file=sys.stderr)
+                print(traceback.format_exc(), file=sys.stderr)
+
+    def postprocess(self, p, processed):
+        for script in self.alwayson_scripts:
+            try:
+                script_args = p.script_args[script.args_from:script.args_to]
+                script.postprocess(p, processed, *script_args)
+            except Exception:
+                print(f"Error running postprocess: {script.filename}", file=sys.stderr)
                 print(traceback.format_exc(), file=sys.stderr)
 
     def reload_sources(self, cache):

modules/shared.py

@@ -144,9 +144,38 @@ class State:
         self.sampling_step = 0
         self.current_image_sampling_step = 0
 
-    def get_job_timestamp(self):
-        return datetime.datetime.now().strftime("%Y%m%d%H%M%S")  # shouldn't this return job_timestamp?
+    def dict(self):
+        obj = {
+            "skipped": self.skipped,
+            "interrupted": self.skipped,
+            "job": self.job,
+            "job_count": self.job_count,
+            "job_no": self.job_no,
+            "sampling_step": self.sampling_step,
+            "sampling_steps": self.sampling_steps,
+        }
 
+        return obj
+
+    def begin(self):
+        self.sampling_step = 0
+        self.job_count = -1
+        self.job_no = 0
+        self.job_timestamp = datetime.datetime.now().strftime("%Y%m%d%H%M%S")
+        self.current_latent = None
+        self.current_image = None
+        self.current_image_sampling_step = 0
+        self.skipped = False
+        self.interrupted = False
+        self.textinfo = None
+
+        devices.torch_gc()
+
+    def end(self):
+        self.job = ""
+        self.job_count = 0
+
+        devices.torch_gc()
 
 state = State()
 

modules/textual_inversion/dataset.py

@@ -42,6 +42,8 @@ class PersonalizedBase(Dataset):
         self.lines = lines
 
         assert data_root, 'dataset directory not specified'
+        assert os.path.isdir(data_root), "Dataset directory doesn't exist"
+        assert os.listdir(data_root), "Dataset directory is empty"
 
         cond_model = shared.sd_model.cond_stage_model
 

modules/textual_inversion/learn_schedule.py

@@ -4,30 +4,37 @@ import tqdm
 class LearnScheduleIterator:
     def __init__(self, learn_rate, max_steps, cur_step=0):
         """
-        specify learn_rate as "0.001:100, 0.00001:1000, 1e-5:10000" to have lr of 0.001 until step 100, 0.00001 until 1000, 1e-5:10000 until 10000
+        specify learn_rate as "0.001:100, 0.00001:1000, 1e-5:10000" to have lr of 0.001 until step 100, 0.00001 until 1000, and 1e-5 until 10000
         """
 
         pairs = learn_rate.split(',')
         self.rates = []
         self.it = 0
         self.maxit = 0
-        for i, pair in enumerate(pairs):
-            tmp = pair.split(':')
-            if len(tmp) == 2:
-                step = int(tmp[1])
-                if step > cur_step:
-                    self.rates.append((float(tmp[0]), min(step, max_steps)))
-                    self.maxit += 1
-                    if step > max_steps:
+        try:
+            for i, pair in enumerate(pairs):
+                if not pair.strip():
+                    continue
+                tmp = pair.split(':')
+                if len(tmp) == 2:
+                    step = int(tmp[1])
+                    if step > cur_step:
+                        self.rates.append((float(tmp[0]), min(step, max_steps)))
+                        self.maxit += 1
+                        if step > max_steps:
+                            return
+                    elif step == -1:
+                        self.rates.append((float(tmp[0]), max_steps))
+                        self.maxit += 1
                         return
-                elif step == -1:
+                else:
                     self.rates.append((float(tmp[0]), max_steps))
                     self.maxit += 1
                     return
-            else:
-                self.rates.append((float(tmp[0]), max_steps))
-                self.maxit += 1
-                return
+            assert self.rates
+        except (ValueError, AssertionError):
+            raise Exception('Invalid learning rate schedule. It should be a number or, for example, like "0.001:100, 0.00001:1000, 1e-5:10000" to have lr of 0.001 until step 100, 0.00001 until 1000, and 1e-5 until 10000.')
+
 
     def __iter__(self):
         return self

modules/textual_inversion/textual_inversion.py

@@ -119,7 +119,7 @@ class EmbeddingDatabase:
             vec = emb.detach().to(devices.device, dtype=torch.float32)
             embedding = Embedding(vec, name)
             embedding.step = data.get('step', None)
-            embedding.sd_checkpoint = data.get('hash', None)
+            embedding.sd_checkpoint = data.get('sd_checkpoint', None)
             embedding.sd_checkpoint_name = data.get('sd_checkpoint_name', None)
             self.register_embedding(embedding, shared.sd_model)
 
@@ -204,9 +204,30 @@ def write_loss(log_directory, filename, step, epoch_len, values):
             **values,
         })
 
+def validate_train_inputs(model_name, learn_rate, batch_size, data_root, template_file, steps, save_model_every, create_image_every, log_directory, name="embedding"):
+    assert model_name, f"{name} not selected"
+    assert learn_rate, "Learning rate is empty or 0"
+    assert isinstance(batch_size, int), "Batch size must be integer"
+    assert batch_size > 0, "Batch size must be positive"
+    assert data_root, "Dataset directory is empty"
+    assert os.path.isdir(data_root), "Dataset directory doesn't exist"
+    assert os.listdir(data_root), "Dataset directory is empty"
+    assert template_file, "Prompt template file is empty"
+    assert os.path.isfile(template_file), "Prompt template file doesn't exist"
+    assert steps, "Max steps is empty or 0"
+    assert isinstance(steps, int), "Max steps must be integer"
+    assert steps > 0 , "Max steps must be positive"
+    assert isinstance(save_model_every, int), "Save {name} must be integer"
+    assert save_model_every >= 0 , "Save {name} must be positive or 0"
+    assert isinstance(create_image_every, int), "Create image must be integer"
+    assert create_image_every >= 0 , "Create image must be positive or 0"
+    if save_model_every or create_image_every:
+        assert log_directory, "Log directory is empty"
 
 def train_embedding(embedding_name, learn_rate, batch_size, data_root, log_directory, training_width, training_height, steps, create_image_every, save_embedding_every, template_file, save_image_with_stored_embedding, preview_from_txt2img, preview_prompt, preview_negative_prompt, preview_steps, preview_sampler_index, preview_cfg_scale, preview_seed, preview_width, preview_height):
-    assert embedding_name, 'embedding not selected'
+    save_embedding_every = save_embedding_every or 0
+    create_image_every = create_image_every or 0
+    validate_train_inputs(embedding_name, learn_rate, batch_size, data_root, template_file, steps, save_embedding_every, create_image_every, log_directory, name="embedding")
 
     shared.state.textinfo = "Initializing textual inversion training..."
     shared.state.job_count = steps
@@ -232,17 +253,28 @@ def train_embedding(embedding_name, learn_rate, batch_size, data_root, log_direc
         os.makedirs(images_embeds_dir, exist_ok=True)
     else:
         images_embeds_dir = None
-        
-    cond_model = shared.sd_model.cond_stage_model
 
-    shared.state.textinfo = f"Preparing dataset from {html.escape(data_root)}..."
-    with torch.autocast("cuda"):
-        ds = modules.textual_inversion.dataset.PersonalizedBase(data_root=data_root, width=training_width, height=training_height, repeats=shared.opts.training_image_repeats_per_epoch, placeholder_token=embedding_name, model=shared.sd_model, device=devices.device, template_file=template_file, batch_size=batch_size)
+    cond_model = shared.sd_model.cond_stage_model
 
     hijack = sd_hijack.model_hijack
 
     embedding = hijack.embedding_db.word_embeddings[embedding_name]
+    checkpoint = sd_models.select_checkpoint()
+
+    ititial_step = embedding.step or 0
+    if ititial_step >= steps:
+        shared.state.textinfo = f"Model has already been trained beyond specified max steps"
+        return embedding, filename
+
+    scheduler = LearnRateScheduler(learn_rate, steps, ititial_step)
+
+    # dataset loading may take a while, so input validations and early returns should be done before this
+    shared.state.textinfo = f"Preparing dataset from {html.escape(data_root)}..."
+    with torch.autocast("cuda"):
+        ds = modules.textual_inversion.dataset.PersonalizedBase(data_root=data_root, width=training_width, height=training_height, repeats=shared.opts.training_image_repeats_per_epoch, placeholder_token=embedding_name, model=shared.sd_model, device=devices.device, template_file=template_file, batch_size=batch_size)
+
     embedding.vec.requires_grad = True
+    optimizer = torch.optim.AdamW([embedding.vec], lr=scheduler.learn_rate)
 
     losses = torch.zeros((32,))
 
@@ -251,13 +283,6 @@ def train_embedding(embedding_name, learn_rate, batch_size, data_root, log_direc
     forced_filename = "<none>"
     embedding_yet_to_be_embedded = False
 
-    ititial_step = embedding.step or 0
-    if ititial_step > steps:
-        return embedding, filename
-
-    scheduler = LearnRateScheduler(learn_rate, steps, ititial_step)
-    optimizer = torch.optim.AdamW([embedding.vec], lr=scheduler.learn_rate)
-
     pbar = tqdm.tqdm(enumerate(ds), total=steps-ititial_step)
     for i, entries in pbar:
         embedding.step = i + ititial_step
@@ -290,9 +315,9 @@ def train_embedding(embedding_name, learn_rate, batch_size, data_root, log_direc
 
         if embedding_dir is not None and steps_done % save_embedding_every == 0:
             # Before saving, change name to match current checkpoint.
-            embedding.name = f'{embedding_name}-{steps_done}'
-            last_saved_file = os.path.join(embedding_dir, f'{embedding.name}.pt')
-            embedding.save(last_saved_file)
+            embedding_name_every = f'{embedding_name}-{steps_done}'
+            last_saved_file = os.path.join(embedding_dir, f'{embedding_name_every}.pt')
+            save_embedding(embedding, checkpoint, embedding_name_every, last_saved_file, remove_cached_checksum=True)
             embedding_yet_to_be_embedded = True
 
         write_loss(log_directory, "textual_inversion_loss.csv", embedding.step, len(ds), {
@@ -373,14 +398,26 @@ Last saved image: {html.escape(last_saved_image)}<br/>
 </p>
 """
 
-    checkpoint = sd_models.select_checkpoint()
-
-    embedding.sd_checkpoint = checkpoint.hash
-    embedding.sd_checkpoint_name = checkpoint.model_name
-    embedding.cached_checksum = None
-    # Before saving for the last time, change name back to base name (as opposed to the save_embedding_every step-suffixed naming convention).
-    embedding.name = embedding_name
-    filename = os.path.join(shared.cmd_opts.embeddings_dir, f'{embedding.name}.pt')
-    embedding.save(filename)
+    filename = os.path.join(shared.cmd_opts.embeddings_dir, f'{embedding_name}.pt')
+    save_embedding(embedding, checkpoint, embedding_name, filename, remove_cached_checksum=True)
 
     return embedding, filename
+
+def save_embedding(embedding, checkpoint, embedding_name, filename, remove_cached_checksum=True):
+    old_embedding_name = embedding.name
+    old_sd_checkpoint = embedding.sd_checkpoint if hasattr(embedding, "sd_checkpoint") else None
+    old_sd_checkpoint_name = embedding.sd_checkpoint_name if hasattr(embedding, "sd_checkpoint_name") else None
+    old_cached_checksum = embedding.cached_checksum if hasattr(embedding, "cached_checksum") else None
+    try:
+        embedding.sd_checkpoint = checkpoint.hash
+        embedding.sd_checkpoint_name = checkpoint.model_name
+        if remove_cached_checksum:
+            embedding.cached_checksum = None
+        embedding.name = embedding_name
+        embedding.save(filename)
+    except:
+        embedding.sd_checkpoint = old_sd_checkpoint
+        embedding.sd_checkpoint_name = old_sd_checkpoint_name
+        embedding.name = old_embedding_name
+        embedding.cached_checksum = old_cached_checksum
+        raise