diff --git a/.github/workflows/test-invoke-conda.yml b/.github/workflows/test-invoke-conda.yml
index 5812b63fcc..f37512d2bb 100644
--- a/.github/workflows/test-invoke-conda.yml
+++ b/.github/workflows/test-invoke-conda.yml
@@ -86,14 +86,14 @@ jobs:
         if: ${{ github.ref != 'refs/heads/main' && github.ref != 'refs/heads/development' }}
         run: echo "TEST_PROMPTS=tests/validate_pr_prompt.txt" >> $GITHUB_ENV
 
-      - name: run preload_models.py
+      - name: run configure_invokeai.py
         id: run-preload-models
         run: |
           if [ "${HAVE_SECRETS}" == true ] ; then
             mkdir -p ~/.huggingface
             echo -n '${{ secrets.HUGGINGFACE_TOKEN }}' > ~/.huggingface/token
           fi
-          python scripts/preload_models.py \
+          python scripts/configure_invokeai.py \
             --no-interactive --yes \
             --full-precision  # can't use fp16 weights without a GPU
 
diff --git a/backend/invoke_ai_web_server.py b/backend/invoke_ai_web_server.py
index d525cf87f8..a66e12a1a1 100644
--- a/backend/invoke_ai_web_server.py
+++ b/backend/invoke_ai_web_server.py
@@ -1,33 +1,31 @@
-import eventlet
+import base64
 import glob
+import io
+import json
+import math
+import mimetypes
 import os
 import shutil
-import mimetypes
 import traceback
-import math
-import io
-import base64
-import os
-import json
+from threading import Event
+from uuid import uuid4
 
-from werkzeug.utils import secure_filename
+import eventlet
+from PIL import Image
+from PIL.Image import Image as ImageType
 from flask import Flask, redirect, send_from_directory, request, make_response
 from flask_socketio import SocketIO
-from PIL import Image, ImageOps
-from PIL.Image import Image as ImageType
-from uuid import uuid4
-from threading import Event
+from werkzeug.utils import secure_filename
 
-from ldm.invoke.args import Args, APP_ID, APP_VERSION, calculate_init_img_hash
-from ldm.invoke.generator.diffusers_pipeline import PipelineIntermediateState
-from ldm.invoke.pngwriter import PngWriter, retrieve_metadata
-from ldm.invoke.prompt_parser import split_weighted_subprompts
-from ldm.invoke.generator.inpaint import infill_methods
-
-from backend.modules.parameters import parameters_to_command
 from backend.modules.get_canvas_generation_mode import (
     get_canvas_generation_mode,
 )
+from backend.modules.parameters import parameters_to_command
+from ldm.invoke.args import Args, APP_ID, APP_VERSION, calculate_init_img_hash
+from ldm.invoke.generator.diffusers_pipeline import PipelineIntermediateState
+from ldm.invoke.generator.inpaint import infill_methods
+from ldm.invoke.pngwriter import PngWriter, retrieve_metadata
+from ldm.invoke.prompt_parser import split_weighted_subprompts
 
 # Loading Arguments
 opt = Args()
@@ -251,7 +249,7 @@ class InvokeAIWebServer:
                 return candidate
         assert "Frontend files cannot be found. Cannot continue"
 
-                                        
+
     def setup_app(self):
         self.result_url = "outputs/"
         self.init_image_url = "outputs/init-images/"
@@ -776,10 +774,10 @@ class InvokeAIWebServer:
                 ).convert("RGBA")
 
                 """
-                The outpaint image and mask are pre-cropped by the UI, so the bounding box we pass 
+                The outpaint image and mask are pre-cropped by the UI, so the bounding box we pass
                 to the generator should be:
                     {
-                        "x": 0, 
+                        "x": 0,
                         "y": 0,
                         "width": original_bounding_box["width"],
                         "height": original_bounding_box["height"]
@@ -799,7 +797,7 @@ class InvokeAIWebServer:
                 )
 
                 """
-                Apply the mask to the init image, creating a "mask" image with 
+                Apply the mask to the init image, creating a "mask" image with
                 transparency where inpainting should occur. This is the kind of
                 mask that prompt2image() needs.
                 """
diff --git a/environments-and-requirements/environment-lin-cuda.yml b/environments-and-requirements/environment-lin-cuda.yml
index 83ff8b5bb0..4397f95371 100644
--- a/environments-and-requirements/environment-lin-cuda.yml
+++ b/environments-and-requirements/environment-lin-cuda.yml
@@ -40,15 +40,6 @@ dependencies:
     - torch-fidelity==0.3.0
     - torchmetrics==0.7.0
     - transformers==4.21.3
-    - diffusers~=0.7
-    - torchmetrics==0.7.0
-    - flask==2.1.3
-    - flask_socketio==5.3.0
-    - flask_cors==3.0.10
-    - dependency_injector==4.40.0
-    - eventlet
-    - getpass_asterisk
-    - kornia==0.6.0
     - git+https://github.com/openai/CLIP.git@main#egg=clip
     - git+https://github.com/Birch-san/k-diffusion.git@mps#egg=k-diffusion
     - git+https://github.com/invoke-ai/clipseg.git@relaxed-python-requirement#egg=clipseg
diff --git a/ldm/generate.py b/ldm/generate.py
index 04cdaea89d..375b72b362 100644
--- a/ldm/generate.py
+++ b/ldm/generate.py
@@ -236,7 +236,7 @@ class Generate:
             except Exception:
                 print('** An error was encountered while installing the safety checker:')
                 print(traceback.format_exc())
-                
+
     def prompt2png(self, prompt, outdir, **kwargs):
         """
         Takes a prompt and an output directory, writes out the requested number
@@ -330,7 +330,7 @@ class Generate:
             infill_method = infill_methods[0], # The infill method to use
             force_outpaint: bool = False,
             enable_image_debugging = False,
-            
+
             **args,
     ):   # eat up additional cruft
         """
@@ -373,7 +373,7 @@ class Generate:
             def process_image(image,seed):
                 image.save(f{'images/seed.png'})
 
-        The code used to save images to a directory can be found in ldm/invoke/pngwriter.py. 
+        The code used to save images to a directory can be found in ldm/invoke/pngwriter.py.
         It contains code to create the requested output directory, select a unique informative
         name for each image, and write the prompt into the PNG metadata.
         """
@@ -593,7 +593,7 @@ class Generate:
         seed = opt.seed or args.seed
         if seed is None or seed < 0:
             seed = random.randrange(0, np.iinfo(np.uint32).max)
-        
+
         prompt = opt.prompt or args.prompt or ''
         print(f'>> using seed {seed} and prompt "{prompt}" for {image_path}')
 
@@ -645,7 +645,7 @@ class Generate:
 
             opt.seed = seed
             opt.prompt = prompt
-            
+
             if len(extend_instructions) > 0:
                 restorer = Outcrop(image,self,)
                 return restorer.process (
@@ -687,7 +687,7 @@ class Generate:
                 image_callback = callback,
                 prefix         = prefix
             )
-                
+
         elif tool is None:
             print(f'* please provide at least one postprocessing option, such as -G or -U')
             return None
@@ -710,13 +710,13 @@ class Generate:
 
         if embiggen is not None:
             return self._make_embiggen()
-            
+
         if inpainting_model_in_use:
             return self._make_omnibus()
 
         if ((init_image is not None) and (mask_image is not None)) or force_outpaint:
             return self._make_inpaint()
-        
+
         if init_image is not None:
             return self._make_img2img()
 
@@ -747,7 +747,7 @@ class Generate:
         if self._has_transparency(image):
             self._transparency_check_and_warning(image, mask, force_outpaint)
             init_mask = self._create_init_mask(image, width, height, fit=fit)
-            
+
         if (image.width * image.height) > (self.width * self.height) and self.size_matters:
             print(">> This input is larger than your defaults. If you run out of memory, please use a smaller image.")
             self.size_matters = False
@@ -763,7 +763,7 @@ class Generate:
 
         if invert_mask:
             init_mask = ImageOps.invert(init_mask)
-            
+
         return init_image,init_mask
 
     # lots o' repeated code here! Turn into a make_func()
@@ -822,7 +822,7 @@ class Generate:
         self.set_model(self.model_name)
 
     def set_model(self,model_name):
-        """ 
+        """
         Given the name of a model defined in models.yaml, will load and initialize it
         and return the model object. Previously-used models will be cached.
         """
@@ -834,7 +834,7 @@ class Generate:
         if not cache.valid_model(model_name):
             print(f'** "{model_name}" is not a known model name. Please check your models.yaml file')
             return self.model
-        
+
         cache.print_vram_usage()
 
         # have to get rid of all references to model in order
@@ -843,7 +843,7 @@ class Generate:
         self.sampler = None
         self.generators = {}
         gc.collect()
-        
+
         model_data = cache.get_model(model_name)
         if model_data is None:  # restore previous
             model_data = cache.get_model(self.model_name)
@@ -856,7 +856,7 @@ class Generate:
 
         # uncache generators so they pick up new models
         self.generators = {}
-        
+
         seed_everything(random.randrange(0, np.iinfo(np.uint32).max))
         if self.embedding_path is not None:
             self.model.embedding_manager.load(
@@ -905,7 +905,7 @@ class Generate:
                                 image_callback = None,
                                 prefix = None,
     ):
-            
+
         for r in image_list:
             image, seed = r
             try:
@@ -915,7 +915,7 @@ class Generate:
                             if self.gfpgan is None:
                                 print('>> GFPGAN not found. Face restoration is disabled.')
                             else:
-                              image = self.gfpgan.process(image, strength, seed)                              
+                              image = self.gfpgan.process(image, strength, seed)
                         if facetool == 'codeformer':
                             if self.codeformer is None:
                                 print('>> CodeFormer not found. Face restoration is disabled.')
diff --git a/ldm/invoke/generator/base.py b/ldm/invoke/generator/base.py
index 7325e1334e..da2d2900ca 100644
--- a/ldm/invoke/generator/base.py
+++ b/ldm/invoke/generator/base.py
@@ -9,9 +9,10 @@ import os.path as osp
 import random
 import traceback
 
+import cv2
 import numpy as np
 import torch
-from PIL import Image, ImageFilter
+from PIL import Image, ImageFilter, ImageChops
 from diffusers import DiffusionPipeline
 from einops import rearrange
 from pytorch_lightning import seed_everything
@@ -169,7 +170,7 @@ class Generator:
         # Blur the mask out (into init image) by specified amount
         if mask_blur_radius > 0:
             nm = np.asarray(pil_init_mask, dtype=np.uint8)
-            nmd = cv.erode(nm, kernel=np.ones((3,3), dtype=np.uint8), iterations=int(mask_blur_radius / 2))
+            nmd = cv2.erode(nm, kernel=np.ones((3,3), dtype=np.uint8), iterations=int(mask_blur_radius / 2))
             pmd = Image.fromarray(nmd, mode='L')
             blurred_init_mask = pmd.filter(ImageFilter.BoxBlur(mask_blur_radius))
         else:
@@ -181,8 +182,6 @@ class Generator:
         matched_result.paste(init_image, (0,0), mask = multiplied_blurred_init_mask)
         return matched_result
 
-
-
     def sample_to_lowres_estimated_image(self,samples):
         # origingally adapted from code by @erucipe and @keturn here:
         # https://discuss.huggingface.co/t/decoding-latents-to-rgb-without-upscaling/23204/7
diff --git a/ldm/invoke/model_cache.py b/ldm/invoke/model_cache.py
index 94eb8e9cea..4efa7aca0d 100644
--- a/ldm/invoke/model_cache.py
+++ b/ldm/invoke/model_cache.py
@@ -21,9 +21,6 @@ from typing import Union
 
 import torch
 import transformers
-import textwrap
-import contextlib
-from typing import Union
 from omegaconf import OmegaConf
 from omegaconf.errors import ConfigAttributeError
 from picklescan.scanner import scan_file_path
@@ -99,7 +96,7 @@ class ModelCache(object):
                 assert self.current_model,'** FATAL: no current model to restore to'
                 print(f'** restoring {self.current_model}')
                 self.get_model(self.current_model)
-                return None
+                return
 
         self.current_model = model_name
         self._push_newest_model(model_name)
@@ -219,7 +216,7 @@ class ModelCache(object):
         if model_format == 'ckpt':
             weights = mconfig.weights
             print(f'>> Loading {model_name} from {weights}')
-            model, width, height, model_hash = self._load_ckpt_model(mconfig)
+            model, width, height, model_hash = self._load_ckpt_model(model_name, mconfig)
         elif model_format == 'diffusers':
             model, width, height, model_hash = self._load_diffusers_model(mconfig)
         else:
@@ -237,10 +234,10 @@ class ModelCache(object):
             )
         return model, width, height, model_hash
 
-    def _load_ckpt_model(self, mconfig):
+    def _load_ckpt_model(self, model_name, mconfig):
         config = mconfig.config
         weights = mconfig.weights
-        vae = mconfig.get('vae', None)
+        vae = mconfig.get('vae')
         width = mconfig.width
         height = mconfig.height
 
@@ -249,10 +246,22 @@ class ModelCache(object):
         if not os.path.isabs(weights):
             weights = os.path.normpath(os.path.join(Globals.root,weights))
         # scan model
-        self._scan_model(model_name, weights)
+        self.scan_model(model_name, weights)
 
-        c = OmegaConf.load(config)
-        with open(weights, 'rb') as f:
+        print(f'>> Loading {model_name} from {weights}')
+
+        # for usage statistics
+        if self._has_cuda():
+            torch.cuda.reset_peak_memory_stats()
+            torch.cuda.empty_cache()
+
+        tic = time.time()
+
+        # this does the work
+        if not os.path.isabs(config):
+            config = os.path.join(Globals.root,config)
+        omega_config = OmegaConf.load(config)
+        with open(weights,'rb') as f:
             weight_bytes = f.read()
         model_hash = self._cached_sha256(weights, weight_bytes)
         sd = torch.load(io.BytesIO(weight_bytes), map_location='cpu')
@@ -289,6 +298,18 @@ class ModelCache(object):
             if isinstance(module, (torch.nn.Conv2d, torch.nn.ConvTranspose2d)):
                 module._orig_padding_mode = module.padding_mode
 
+        # usage statistics
+        toc = time.time()
+        print(f'>> Model loaded in', '%4.2fs' % (toc - tic))
+
+        if self._has_cuda():
+            print(
+                '>> Max VRAM used to load the model:',
+                '%4.2fG' % (torch.cuda.max_memory_allocated() / 1e9),
+                '\n>> Current VRAM usage:'
+                '%4.2fG' % (torch.cuda.memory_allocated() / 1e9),
+            )
+
         return model, width, height, model_hash
 
     def _load_diffusers_model(self, mconfig):
@@ -308,6 +329,8 @@ class ModelCache(object):
 
         print(f'>> Loading diffusers model from {name_or_path}')
 
+        # TODO: scan weights maybe?
+
         if self.precision == 'float16':
             print('   | Using faster float16 precision')
             pipeline_args.update(revision="fp16", torch_dtype=torch.float16)
@@ -342,7 +365,7 @@ class ModelCache(object):
         else:
             raise ValueError("Model config must specify either repo_name or path.")
 
-    def offload_model(self, model_name:str):
+    def offload_model(self, model_name:str) -> None:
         '''
         Offload the indicated model to CPU. Will call
         _make_cache_room() to free space if needed.
diff --git a/ldm/models/diffusion/cross_attention_control.py b/ldm/models/diffusion/cross_attention_control.py
index 08145b1e76..ec7c3c215c 100644
--- a/ldm/models/diffusion/cross_attention_control.py
+++ b/ldm/models/diffusion/cross_attention_control.py
@@ -248,33 +248,33 @@ def inject_attention_function(unet, context: Context):
 
     cross_attention_modules = [(name, module) for (name, module) in unet.named_modules()
                                    if type(module).__name__ == "CrossAttention"]
-        for identifier, module in cross_attention_modules:
-            module.identifier = identifier
-            try:
-                module.set_attention_slice_wrangler(attention_slice_wrangler)
-                module.set_slicing_strategy_getter(
-                    lambda module: context.get_slicing_strategy(identifier)
-                )
-            except AttributeError as e:
-                if is_attribute_error_about(e, 'set_attention_slice_wrangler'):
-                    warnings.warn(f"TODO: implement for {type(module)}")  # TODO
-                else:
-                    raise
+    for identifier, module in cross_attention_modules:
+        module.identifier = identifier
+        try:
+            module.set_attention_slice_wrangler(attention_slice_wrangler)
+            module.set_slicing_strategy_getter(
+                lambda module: context.get_slicing_strategy(identifier)
+            )
+        except AttributeError as e:
+            if is_attribute_error_about(e, 'set_attention_slice_wrangler'):
+                warnings.warn(f"TODO: implement for {type(module)}")  # TODO
+            else:
+                raise
 
 
 def remove_attention_function(unet):
     cross_attention_modules = [module for (_, module) in unet.named_modules()
                                    if type(module).__name__ == "CrossAttention"]
-        for module in cross_attention_modules:
-            try:
-                # clear wrangler callback
-                module.set_attention_slice_wrangler(None)
-                module.set_slicing_strategy_getter(None)
-            except AttributeError as e:
-                if is_attribute_error_about(e, 'set_attention_slice_wrangler'):
-                    warnings.warn(f"TODO: implement for {type(module)}")  # TODO
-                else:
-                    raise
+    for module in cross_attention_modules:
+        try:
+            # clear wrangler callback
+            module.set_attention_slice_wrangler(None)
+            module.set_slicing_strategy_getter(None)
+        except AttributeError as e:
+            if is_attribute_error_about(e, 'set_attention_slice_wrangler'):
+                warnings.warn(f"TODO: implement for {type(module)}")  # TODO
+            else:
+                raise
 
 
 def is_attribute_error_about(error: AttributeError, attribute: str):
diff --git a/scripts/configure_invokeai.py b/scripts/configure_invokeai.py
index fd593bd584..af3489d737 100644
--- a/scripts/configure_invokeai.py
+++ b/scripts/configure_invokeai.py
@@ -34,6 +34,12 @@ warnings.filterwarnings('ignore')
 import torch
 transformers.logging.set_verbosity_error()
 
+try:
+    from ldm.invoke.model_cache import ModelCache
+except ImportError:
+    sys.path.append('.')
+    from ldm.invoke.model_cache import ModelCache
+
 #--------------------------globals-----------------------
 Model_dir = 'models'
 Weights_dir = 'ldm/stable-diffusion-v1/'
@@ -267,6 +273,19 @@ def download_weight_datasets(models:dict, access_token:str):
     print(f'Successfully installed {keys}')
     return successful
 
+#---------------------------------------------
+def is_huggingface_authenticated():
+    # huggingface_hub 0.10 API isn't great for this, it could be OSError, ValueError,
+    # maybe other things, not all end-user-friendly.
+    # noinspection PyBroadException
+    try:
+        response = hf_whoami()
+        if response.get('id') is not None:
+            return True
+    except Exception:
+        pass
+    return False
+
 #---------------------------------------------
 def hf_download_with_resume(repo_id:str, model_dir:str, model_name:str, access_token:str=None)->bool:
     model_dest = os.path.join(model_dir, model_name)
@@ -749,6 +768,12 @@ def main():
                         action=argparse.BooleanOptionalAction,
                         default=True,
                         help='run in interactive mode (default)')
+    parser.add_argument('--full-precision',
+                        dest='full_precision',
+                        action=argparse.BooleanOptionalAction,
+                        type=bool,
+                        default=False,
+                        help='use 32-bit weights instead of faster 16-bit weights')
     parser.add_argument('--yes','-y',
                         dest='yes_to_all',
                         action='store_true',