images history improvement

modules/devices.py

@@ -36,6 +36,7 @@ errors.run(enable_tf32, "Enabling TF32")
 
 device = device_gfpgan = device_bsrgan = device_esrgan = device_scunet = device_codeformer = get_optimal_device()
 dtype = torch.float16
+dtype_vae = torch.float16
 
 def randn(seed, shape):
     # Pytorch currently doesn't handle setting randomness correctly when the metal backend is used.
@@ -59,9 +60,12 @@ def randn_without_seed(shape):
     return torch.randn(shape, device=device)
 
 
-def autocast():
+def autocast(disable=False):
     from modules import shared
 
+    if disable:
+        return contextlib.nullcontext()
+
     if dtype == torch.float32 or shared.cmd_opts.precision == "full":
         return contextlib.nullcontext()
 

modules/hypernetwork.py

@@ -1,98 +0,0 @@
-import glob
-import os
-import sys
-import traceback
-
-import torch
-
-from ldm.util import default
-from modules import devices, shared
-import torch
-from torch import einsum
-from einops import rearrange, repeat
-
-
-class HypernetworkModule(torch.nn.Module):
-    def __init__(self, dim, state_dict):
-        super().__init__()
-
-        self.linear1 = torch.nn.Linear(dim, dim * 2)
-        self.linear2 = torch.nn.Linear(dim * 2, dim)
-
-        self.load_state_dict(state_dict, strict=True)
-        self.to(devices.device)
-
-    def forward(self, x):
-        return x + (self.linear2(self.linear1(x)))
-
-
-class Hypernetwork:
-    filename = None
-    name = None
-
-    def __init__(self, filename):
-        self.filename = filename
-        self.name = os.path.splitext(os.path.basename(filename))[0]
-        self.layers = {}
-
-        state_dict = torch.load(filename, map_location='cpu')
-        for size, sd in state_dict.items():
-            self.layers[size] = (HypernetworkModule(size, sd[0]), HypernetworkModule(size, sd[1]))
-
-
-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
-    return res
-
-
-def load_hypernetwork(filename):
-    path = shared.hypernetworks.get(filename, None)
-    if path is not None:
-        print(f"Loading hypernetwork {filename}")
-        try:
-            shared.loaded_hypernetwork = Hypernetwork(path)
-        except Exception:
-            print(f"Error loading hypernetwork {path}", file=sys.stderr)
-            print(traceback.format_exc(), file=sys.stderr)
-    else:
-        if shared.loaded_hypernetwork is not None:
-            print(f"Unloading hypernetwork")
-
-        shared.loaded_hypernetwork = None
-
-
-def attention_CrossAttention_forward(self, x, context=None, mask=None):
-    h = self.heads
-
-    q = self.to_q(x)
-    context = default(context, x)
-
-    hypernetwork = shared.loaded_hypernetwork
-    hypernetwork_layers = (hypernetwork.layers if hypernetwork is not None else {}).get(context.shape[2], None)
-
-    if hypernetwork_layers is not None:
-        k = self.to_k(hypernetwork_layers[0](context))
-        v = self.to_v(hypernetwork_layers[1](context))
-    else:
-        k = self.to_k(context)
-        v = self.to_v(context)
-
-    q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> (b h) n d', h=h), (q, k, v))
-
-    sim = einsum('b i d, b j d -> b i j', q, k) * self.scale
-
-    if mask is not None:
-        mask = rearrange(mask, 'b ... -> b (...)')
-        max_neg_value = -torch.finfo(sim.dtype).max
-        mask = repeat(mask, 'b j -> (b h) () j', h=h)
-        sim.masked_fill_(~mask, max_neg_value)
-
-    # attention, what we cannot get enough of
-    attn = sim.softmax(dim=-1)
-
-    out = einsum('b i j, b j d -> b i d', attn, v)
-    out = rearrange(out, '(b h) n d -> b n (h d)', h=h)
-    return self.to_out(out)

modules/hypernetwork/hypernetwork.py

@@ -0,0 +1,283 @@
+import datetime
+import glob
+import html
+import os
+import sys
+import traceback
+import tqdm
+
+import torch
+
+from ldm.util import default
+from modules import devices, shared, processing, sd_models
+import torch
+from torch import einsum
+from einops import rearrange, repeat
+import modules.textual_inversion.dataset
+
+
+class HypernetworkModule(torch.nn.Module):
+    def __init__(self, dim, state_dict=None):
+        super().__init__()
+
+        self.linear1 = torch.nn.Linear(dim, dim * 2)
+        self.linear2 = torch.nn.Linear(dim * 2, dim)
+
+        if state_dict is not None:
+            self.load_state_dict(state_dict, strict=True)
+        else:
+
+            self.linear1.weight.data.normal_(mean=0.0, std=0.01)
+            self.linear1.bias.data.zero_()
+            self.linear2.weight.data.normal_(mean=0.0, std=0.01)
+            self.linear2.bias.data.zero_()
+
+        self.to(devices.device)
+
+    def forward(self, x):
+        return x + (self.linear2(self.linear1(x)))
+
+
+class Hypernetwork:
+    filename = None
+    name = None
+
+    def __init__(self, name=None):
+        self.filename = None
+        self.name = name
+        self.layers = {}
+        self.step = 0
+        self.sd_checkpoint = None
+        self.sd_checkpoint_name = None
+
+        for size in [320, 640, 768, 1280]:
+            self.layers[size] = (HypernetworkModule(size), HypernetworkModule(size))
+
+    def weights(self):
+        res = []
+
+        for k, layers in self.layers.items():
+            for layer in layers:
+                layer.train()
+                res += [layer.linear1.weight, layer.linear1.bias, layer.linear2.weight, layer.linear2.bias]
+
+        return res
+
+    def save(self, filename):
+        state_dict = {}
+
+        for k, v in self.layers.items():
+            state_dict[k] = (v[0].state_dict(), v[1].state_dict())
+
+        state_dict['step'] = self.step
+        state_dict['name'] = self.name
+        state_dict['sd_checkpoint'] = self.sd_checkpoint
+        state_dict['sd_checkpoint_name'] = self.sd_checkpoint_name
+
+        torch.save(state_dict, filename)
+
+    def load(self, filename):
+        self.filename = filename
+        if self.name is None:
+            self.name = os.path.splitext(os.path.basename(filename))[0]
+
+        state_dict = torch.load(filename, map_location='cpu')
+
+        for size, sd in state_dict.items():
+            if type(size) == int:
+                self.layers[size] = (HypernetworkModule(size, sd[0]), HypernetworkModule(size, sd[1]))
+
+        self.name = state_dict.get('name', self.name)
+        self.step = state_dict.get('step', 0)
+        self.sd_checkpoint = state_dict.get('sd_checkpoint', None)
+        self.sd_checkpoint_name = state_dict.get('sd_checkpoint_name', None)
+
+
+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
+    return res
+
+
+def load_hypernetwork(filename):
+    path = shared.hypernetworks.get(filename, None)
+    if path is not None:
+        print(f"Loading hypernetwork {filename}")
+        try:
+            shared.loaded_hypernetwork = Hypernetwork()
+            shared.loaded_hypernetwork.load(path)
+
+        except Exception:
+            print(f"Error loading hypernetwork {path}", file=sys.stderr)
+            print(traceback.format_exc(), file=sys.stderr)
+    else:
+        if shared.loaded_hypernetwork is not None:
+            print(f"Unloading hypernetwork")
+
+        shared.loaded_hypernetwork = None
+
+
+def apply_hypernetwork(hypernetwork, context, layer=None):
+    hypernetwork_layers = (hypernetwork.layers if hypernetwork is not None else {}).get(context.shape[2], None)
+
+    if hypernetwork_layers is None:
+        return context, context
+
+    if layer is not None:
+        layer.hyper_k = hypernetwork_layers[0]
+        layer.hyper_v = hypernetwork_layers[1]
+
+    context_k = hypernetwork_layers[0](context)
+    context_v = hypernetwork_layers[1](context)
+    return context_k, context_v
+
+
+def attention_CrossAttention_forward(self, x, context=None, mask=None):
+    h = self.heads
+
+    q = self.to_q(x)
+    context = default(context, x)
+
+    context_k, context_v = apply_hypernetwork(shared.loaded_hypernetwork, context, self)
+    k = self.to_k(context_k)
+    v = self.to_v(context_v)
+
+    q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> (b h) n d', h=h), (q, k, v))
+
+    sim = einsum('b i d, b j d -> b i j', q, k) * self.scale
+
+    if mask is not None:
+        mask = rearrange(mask, 'b ... -> b (...)')
+        max_neg_value = -torch.finfo(sim.dtype).max
+        mask = repeat(mask, 'b j -> (b h) () j', h=h)
+        sim.masked_fill_(~mask, max_neg_value)
+
+    # attention, what we cannot get enough of
+    attn = sim.softmax(dim=-1)
+
+    out = einsum('b i j, b j d -> b i d', attn, v)
+    out = rearrange(out, '(b h) n d -> b n (h d)', h=h)
+    return self.to_out(out)
+
+
+def train_hypernetwork(hypernetwork_name, learn_rate, data_root, log_directory, steps, create_image_every, save_hypernetwork_every, template_file, preview_image_prompt):
+    assert hypernetwork_name, 'embedding not selected'
+
+    path = shared.hypernetworks.get(hypernetwork_name, None)
+    shared.loaded_hypernetwork = Hypernetwork()
+    shared.loaded_hypernetwork.load(path)
+
+    shared.state.textinfo = "Initializing hypernetwork training..."
+    shared.state.job_count = steps
+
+    filename = os.path.join(shared.cmd_opts.hypernetwork_dir, f'{hypernetwork_name}.pt')
+
+    log_directory = os.path.join(log_directory, datetime.datetime.now().strftime("%Y-%m-%d"), hypernetwork_name)
+
+    if save_hypernetwork_every > 0:
+        hypernetwork_dir = os.path.join(log_directory, "hypernetworks")
+        os.makedirs(hypernetwork_dir, exist_ok=True)
+    else:
+        hypernetwork_dir = None
+
+    if create_image_every > 0:
+        images_dir = os.path.join(log_directory, "images")
+        os.makedirs(images_dir, exist_ok=True)
+    else:
+        images_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=512, height=512, repeats=1, placeholder_token=hypernetwork_name, model=shared.sd_model, device=devices.device, template_file=template_file)
+
+    hypernetwork = shared.loaded_hypernetwork
+    weights = hypernetwork.weights()
+    for weight in weights:
+        weight.requires_grad = True
+
+    optimizer = torch.optim.AdamW(weights, lr=learn_rate)
+
+    losses = torch.zeros((32,))
+
+    last_saved_file = "<none>"
+    last_saved_image = "<none>"
+
+    ititial_step = hypernetwork.step or 0
+    if ititial_step > steps:
+        return hypernetwork, filename
+
+    pbar = tqdm.tqdm(enumerate(ds), total=steps - ititial_step)
+    for i, (x, text) in pbar:
+        hypernetwork.step = i + ititial_step
+
+        if hypernetwork.step > steps:
+            break
+
+        if shared.state.interrupted:
+            break
+
+        with torch.autocast("cuda"):
+            c = cond_model([text])
+
+            x = x.to(devices.device)
+            loss = shared.sd_model(x.unsqueeze(0), c)[0]
+            del x
+
+            losses[hypernetwork.step % losses.shape[0]] = loss.item()
+
+            optimizer.zero_grad()
+            loss.backward()
+            optimizer.step()
+
+        pbar.set_description(f"loss: {losses.mean():.7f}")
+
+        if hypernetwork.step > 0 and hypernetwork_dir is not None and hypernetwork.step % save_hypernetwork_every == 0:
+            last_saved_file = os.path.join(hypernetwork_dir, f'{hypernetwork_name}-{hypernetwork.step}.pt')
+            hypernetwork.save(last_saved_file)
+
+        if hypernetwork.step > 0 and images_dir is not None and hypernetwork.step % create_image_every == 0:
+            last_saved_image = os.path.join(images_dir, f'{hypernetwork_name}-{hypernetwork.step}.png')
+
+            preview_text = text if preview_image_prompt == "" else preview_image_prompt
+
+            p = processing.StableDiffusionProcessingTxt2Img(
+                sd_model=shared.sd_model,
+                prompt=preview_text,
+                steps=20,
+                do_not_save_grid=True,
+                do_not_save_samples=True,
+            )
+
+            processed = processing.process_images(p)
+            image = processed.images[0]
+
+            shared.state.current_image = image
+            image.save(last_saved_image)
+
+            last_saved_image += f", prompt: {preview_text}"
+
+        shared.state.job_no = hypernetwork.step
+
+        shared.state.textinfo = f"""
+<p>
+Loss: {losses.mean():.7f}<br/>
+Step: {hypernetwork.step}<br/>
+Last prompt: {html.escape(text)}<br/>
+Last saved embedding: {html.escape(last_saved_file)}<br/>
+Last saved image: {html.escape(last_saved_image)}<br/>
+</p>
+"""
+
+    checkpoint = sd_models.select_checkpoint()
+
+    hypernetwork.sd_checkpoint = checkpoint.hash
+    hypernetwork.sd_checkpoint_name = checkpoint.model_name
+    hypernetwork.save(filename)
+
+    return hypernetwork, filename
+
+

modules/hypernetwork/ui.py

@@ -0,0 +1,43 @@
+import html
+import os
+
+import gradio as gr
+
+import modules.textual_inversion.textual_inversion
+import modules.textual_inversion.preprocess
+from modules import sd_hijack, shared
+from modules.hypernetwork import hypernetwork
+
+
+def create_hypernetwork(name):
+    fn = os.path.join(shared.cmd_opts.hypernetwork_dir, f"{name}.pt")
+    assert not os.path.exists(fn), f"file {fn} already exists"
+
+    hypernet = modules.hypernetwork.hypernetwork.Hypernetwork(name=name)
+    hypernet.save(fn)
+
+    shared.reload_hypernetworks()
+
+    return gr.Dropdown.update(choices=sorted([x for x in shared.hypernetworks.keys()])), f"Created: {fn}", ""
+
+
+def train_hypernetwork(*args):
+
+    initial_hypernetwork = shared.loaded_hypernetwork
+
+    try:
+        sd_hijack.undo_optimizations()
+
+        hypernetwork, filename = modules.hypernetwork.hypernetwork.train_hypernetwork(*args)
+
+        res = f"""
+Training {'interrupted' if shared.state.interrupted else 'finished'} at {hypernetwork.step} steps.
+Hypernetwork saved to {html.escape(filename)}
+"""
+        return res, ""
+    except Exception:
+        raise
+    finally:
+        shared.loaded_hypernetwork = initial_hypernetwork
+        sd_hijack.apply_optimizations()
+

modules/processing.py

@@ -207,7 +207,7 @@ def create_random_tensors(shape, seeds, subseeds=None, subseed_strength=0.0, see
     # enables the generation of additional tensors with noise that the sampler will use during its processing.
     # Using those pre-generated tensors instead of simple torch.randn allows a batch with seeds [100, 101] to
     # produce the same images as with two batches [100], [101].
-    if p is not None and p.sampler is not None and len(seeds) > 1 and opts.enable_batch_seeds:
+    if p is not None and p.sampler is not None and (len(seeds) > 1 and opts.enable_batch_seeds or opts.eta_noise_seed_delta > 0):
         sampler_noises = [[] for _ in range(p.sampler.number_of_needed_noises(p))]
     else:
         sampler_noises = None
@@ -247,6 +247,9 @@ def create_random_tensors(shape, seeds, subseeds=None, subseed_strength=0.0, see
         if sampler_noises is not None:
             cnt = p.sampler.number_of_needed_noises(p)
 
+            if opts.eta_noise_seed_delta > 0:
+                torch.manual_seed(seed + opts.eta_noise_seed_delta)
+
             for j in range(cnt):
                 sampler_noises[j].append(devices.randn_without_seed(tuple(noise_shape)))
 
@@ -259,6 +262,13 @@ def create_random_tensors(shape, seeds, subseeds=None, subseed_strength=0.0, see
     return x
 
 
+def decode_first_stage(model, x):
+    with devices.autocast(disable=x.dtype == devices.dtype_vae):
+        x = model.decode_first_stage(x)
+
+    return x
+
+
 def get_fixed_seed(seed):
     if seed is None or seed == '' or seed == -1:
         return int(random.randrange(4294967294))
@@ -294,6 +304,7 @@ def create_infotext(p, all_prompts, all_seeds, all_subseeds, comments, iteration
         "Denoising strength": getattr(p, 'denoising_strength', None),
         "Eta": (None if p.sampler is None or p.sampler.eta == p.sampler.default_eta else p.sampler.eta),
         "Clip skip": None if clip_skip <= 1 else clip_skip,
+        "ENSD": None if opts.eta_noise_seed_delta == 0 else opts.eta_noise_seed_delta,
     }
 
     generation_params.update(p.extra_generation_params)
@@ -398,9 +409,8 @@ def process_images(p: StableDiffusionProcessing) -> Processed:
                 # use the image collected previously in sampler loop
                 samples_ddim = shared.state.current_latent
 
-            samples_ddim = samples_ddim.to(devices.dtype)
-
-            x_samples_ddim = p.sd_model.decode_first_stage(samples_ddim)
+            samples_ddim = samples_ddim.to(devices.dtype_vae)
+            x_samples_ddim = decode_first_stage(p.sd_model, samples_ddim)
             x_samples_ddim = torch.clamp((x_samples_ddim + 1.0) / 2.0, min=0.0, max=1.0)
 
             del samples_ddim
@@ -533,7 +543,7 @@ class StableDiffusionProcessingTxt2Img(StableDiffusionProcessing):
         if self.scale_latent:
             samples = torch.nn.functional.interpolate(samples, size=(self.height // opt_f, self.width // opt_f), mode="bilinear")
         else:
-            decoded_samples = self.sd_model.decode_first_stage(samples)
+            decoded_samples = decode_first_stage(self.sd_model, samples)
 
             if opts.upscaler_for_img2img is None or opts.upscaler_for_img2img == "None":
                 decoded_samples = torch.nn.functional.interpolate(decoded_samples, size=(self.height, self.width), mode="bilinear")

modules/sd_hijack.py

@@ -8,7 +8,7 @@ from torch import einsum
 from torch.nn.functional import silu
 
 import modules.textual_inversion.textual_inversion
-from modules import prompt_parser, devices, sd_hijack_optimizations, shared, hypernetwork
+from modules import prompt_parser, devices, sd_hijack_optimizations, shared
 from modules.shared import opts, device, cmd_opts
 
 import ldm.modules.attention
@@ -23,7 +23,7 @@ def apply_optimizations():
 
     ldm.modules.diffusionmodules.model.nonlinearity = silu
 
-    if cmd_opts.force_enable_xformers or (cmd_opts.xformers and shared.xformers_available and torch.version.cuda and torch.cuda.get_device_capability(shared.device) == (8, 6)):
+    if cmd_opts.force_enable_xformers or (cmd_opts.xformers and shared.xformers_available and torch.version.cuda and (6, 0) <= torch.cuda.get_device_capability(shared.device) <= (8, 6)):
         print("Applying xformers cross attention optimization.")
         ldm.modules.attention.CrossAttention.forward = sd_hijack_optimizations.xformers_attention_forward
         ldm.modules.diffusionmodules.model.AttnBlock.forward = sd_hijack_optimizations.xformers_attnblock_forward
@@ -37,16 +37,15 @@ def apply_optimizations():
 
 
 def undo_optimizations():
+    from modules.hypernetwork import hypernetwork
+
     ldm.modules.attention.CrossAttention.forward = hypernetwork.attention_CrossAttention_forward
     ldm.modules.diffusionmodules.model.nonlinearity = diffusionmodules_model_nonlinearity
     ldm.modules.diffusionmodules.model.AttnBlock.forward = diffusionmodules_model_AttnBlock_forward
 
 
 def get_target_prompt_token_count(token_count):
-    if token_count < 75:
-        return 75
-
-    return math.ceil(token_count / 10) * 10
+    return math.ceil(max(token_count, 1) / 75) * 75
 
 
 class StableDiffusionModelHijack:
@@ -110,6 +109,8 @@ class FrozenCLIPEmbedderWithCustomWords(torch.nn.Module):
         self.tokenizer = wrapped.tokenizer
         self.token_mults = {}
 
+        self.comma_token = [v for k, v in self.tokenizer.get_vocab().items() if k == ',</w>'][0]
+
         tokens_with_parens = [(k, v) for k, v in self.tokenizer.get_vocab().items() if '(' in k or ')' in k or '[' in k or ']' in k]
         for text, ident in tokens_with_parens:
             mult = 1.0
@@ -127,7 +128,6 @@ class FrozenCLIPEmbedderWithCustomWords(torch.nn.Module):
                 self.token_mults[ident] = mult
 
     def tokenize_line(self, line, used_custom_terms, hijack_comments):
-        id_start = self.wrapped.tokenizer.bos_token_id
         id_end = self.wrapped.tokenizer.eos_token_id
 
         if opts.enable_emphasis:
@@ -140,6 +140,7 @@ class FrozenCLIPEmbedderWithCustomWords(torch.nn.Module):
         fixes = []
         remade_tokens = []
         multipliers = []
+        last_comma = -1
 
         for tokens, (text, weight) in zip(tokenized, parsed):
             i = 0
@@ -148,13 +149,33 @@ class FrozenCLIPEmbedderWithCustomWords(torch.nn.Module):
 
                 embedding, embedding_length_in_tokens = self.hijack.embedding_db.find_embedding_at_position(tokens, i)
 
+                if token == self.comma_token:
+                    last_comma = len(remade_tokens)
+                elif opts.comma_padding_backtrack != 0 and max(len(remade_tokens), 1) % 75 == 0 and last_comma != -1 and len(remade_tokens) - last_comma <= opts.comma_padding_backtrack:
+                    last_comma += 1
+                    reloc_tokens = remade_tokens[last_comma:]
+                    reloc_mults = multipliers[last_comma:]
+
+                    remade_tokens = remade_tokens[:last_comma]
+                    length = len(remade_tokens)
+                    
+                    rem = int(math.ceil(length / 75)) * 75 - length
+                    remade_tokens += [id_end] * rem + reloc_tokens
+                    multipliers = multipliers[:last_comma] + [1.0] * rem + reloc_mults
+                
                 if embedding is None:
                     remade_tokens.append(token)
                     multipliers.append(weight)
                     i += 1
                 else:
                     emb_len = int(embedding.vec.shape[0])
-                    fixes.append((len(remade_tokens), embedding))
+                    iteration = len(remade_tokens) // 75
+                    if (len(remade_tokens) + emb_len) // 75 != iteration:
+                        rem = (75 * (iteration + 1) - len(remade_tokens))
+                        remade_tokens += [id_end] * rem
+                        multipliers += [1.0] * rem
+                        iteration += 1
+                    fixes.append((iteration, (len(remade_tokens) % 75, embedding)))
                     remade_tokens += [0] * emb_len
                     multipliers += [weight] * emb_len
                     used_custom_terms.append((embedding.name, embedding.checksum()))
@@ -162,10 +183,10 @@ class FrozenCLIPEmbedderWithCustomWords(torch.nn.Module):
 
         token_count = len(remade_tokens)
         prompt_target_length = get_target_prompt_token_count(token_count)
-        tokens_to_add = prompt_target_length - len(remade_tokens) + 1
+        tokens_to_add = prompt_target_length - len(remade_tokens)
 
-        remade_tokens = [id_start] + remade_tokens + [id_end] * tokens_to_add
-        multipliers = [1.0] + multipliers + [1.0] * tokens_to_add
+        remade_tokens = remade_tokens + [id_end] * tokens_to_add
+        multipliers = multipliers + [1.0] * tokens_to_add
 
         return remade_tokens, fixes, multipliers, token_count
 
@@ -260,29 +281,55 @@ class FrozenCLIPEmbedderWithCustomWords(torch.nn.Module):
             hijack_fixes.append(fixes)
             batch_multipliers.append(multipliers)
         return batch_multipliers, remade_batch_tokens, used_custom_terms, hijack_comments, hijack_fixes, token_count
-
+    
     def forward(self, text):
-
-        if opts.use_old_emphasis_implementation:
+        use_old = opts.use_old_emphasis_implementation
+        if use_old:
             batch_multipliers, remade_batch_tokens, used_custom_terms, hijack_comments, hijack_fixes, token_count = self.process_text_old(text)
         else:
             batch_multipliers, remade_batch_tokens, used_custom_terms, hijack_comments, hijack_fixes, token_count = self.process_text(text)
 
-        self.hijack.fixes = hijack_fixes
         self.hijack.comments += hijack_comments
 
         if len(used_custom_terms) > 0:
             self.hijack.comments.append("Used embeddings: " + ", ".join([f'{word} [{checksum}]' for word, checksum in used_custom_terms]))
+        
+        if use_old:
+            self.hijack.fixes = hijack_fixes
+            return self.process_tokens(remade_batch_tokens, batch_multipliers)
+        
+        z = None
+        i = 0
+        while max(map(len, remade_batch_tokens)) != 0:
+            rem_tokens = [x[75:] for x in remade_batch_tokens]
+            rem_multipliers = [x[75:] for x in batch_multipliers]
 
-        target_token_count = get_target_prompt_token_count(token_count) + 2
+            self.hijack.fixes = []
+            for unfiltered in hijack_fixes:
+                fixes = []
+                for fix in unfiltered:
+                    if fix[0] == i:
+                        fixes.append(fix[1])
+                self.hijack.fixes.append(fixes)
+            
+            z1 = self.process_tokens([x[:75] for x in remade_batch_tokens], [x[:75] for x in batch_multipliers])
+            z = z1 if z is None else torch.cat((z, z1), axis=-2)
+            
+            remade_batch_tokens = rem_tokens
+            batch_multipliers = rem_multipliers
+            i += 1
+        
+        return z
+        
+    
+    def process_tokens(self, remade_batch_tokens, batch_multipliers):
+        if not opts.use_old_emphasis_implementation:
+            remade_batch_tokens = [[self.wrapped.tokenizer.bos_token_id] + x[:75] + [self.wrapped.tokenizer.eos_token_id] for x in remade_batch_tokens]
+            batch_multipliers = [[1.0] + x[:75] + [1.0] for x in batch_multipliers]
+        
+        tokens = torch.asarray(remade_batch_tokens).to(device)
+        outputs = self.wrapped.transformer(input_ids=tokens, output_hidden_states=-opts.CLIP_stop_at_last_layers)
 
-        position_ids_array = [min(x, 75) for x in range(target_token_count-1)] + [76]
-        position_ids = torch.asarray(position_ids_array, device=devices.device).expand((1, -1))
-
-        remade_batch_tokens_of_same_length = [x + [self.wrapped.tokenizer.eos_token_id] * (target_token_count - len(x)) for x in remade_batch_tokens]
-        tokens = torch.asarray(remade_batch_tokens_of_same_length).to(device)
-
-        outputs = self.wrapped.transformer(input_ids=tokens, position_ids=position_ids, output_hidden_states=-opts.CLIP_stop_at_last_layers)
         if opts.CLIP_stop_at_last_layers > 1:
             z = outputs.hidden_states[-opts.CLIP_stop_at_last_layers]
             z = self.wrapped.transformer.text_model.final_layer_norm(z)
@@ -290,7 +337,7 @@ class FrozenCLIPEmbedderWithCustomWords(torch.nn.Module):
             z = outputs.last_hidden_state
 
         # restoring original mean is likely not correct, but it seems to work well to prevent artifacts that happen otherwise
-        batch_multipliers_of_same_length = [x + [1.0] * (target_token_count - len(x)) for x in batch_multipliers]
+        batch_multipliers_of_same_length = [x + [1.0] * (75 - len(x)) for x in batch_multipliers]
         batch_multipliers = torch.asarray(batch_multipliers_of_same_length).to(device)
         original_mean = z.mean()
         z *= batch_multipliers.reshape(batch_multipliers.shape + (1,)).expand(z.shape)

modules/sd_hijack_optimizations.py

@@ -9,12 +9,12 @@ from ldm.util import default
 from einops import rearrange
 
 from modules import shared
+from modules.hypernetwork import hypernetwork
+
 
 if shared.cmd_opts.xformers or shared.cmd_opts.force_enable_xformers:
     try:
         import xformers.ops
-        import functorch
-        xformers._is_functorch_available = True
         shared.xformers_available = True
     except Exception:
         print("Cannot import xformers", file=sys.stderr)
@@ -28,16 +28,10 @@ def split_cross_attention_forward_v1(self, x, context=None, mask=None):
     q_in = self.to_q(x)
     context = default(context, x)
 
-    hypernetwork = shared.loaded_hypernetwork
-    hypernetwork_layers = (hypernetwork.layers if hypernetwork is not None else {}).get(context.shape[2], None)
-
-    if hypernetwork_layers is not None:
-        k_in = self.to_k(hypernetwork_layers[0](context))
-        v_in = self.to_v(hypernetwork_layers[1](context))
-    else:
-        k_in = self.to_k(context)
-        v_in = self.to_v(context)
-    del context, x
+    context_k, context_v = hypernetwork.apply_hypernetwork(shared.loaded_hypernetwork, context)
+    k_in = self.to_k(context_k)
+    v_in = self.to_v(context_v)
+    del context, context_k, context_v, x
 
     q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> (b h) n d', h=h), (q_in, k_in, v_in))
     del q_in, k_in, v_in
@@ -61,22 +55,16 @@ def split_cross_attention_forward_v1(self, x, context=None, mask=None):
     return self.to_out(r2)
 
 
-# taken from https://github.com/Doggettx/stable-diffusion
+# taken from https://github.com/Doggettx/stable-diffusion and modified
 def split_cross_attention_forward(self, x, context=None, mask=None):
     h = self.heads
 
     q_in = self.to_q(x)
     context = default(context, x)
 
-    hypernetwork = shared.loaded_hypernetwork
-    hypernetwork_layers = (hypernetwork.layers if hypernetwork is not None else {}).get(context.shape[2], None)
-
-    if hypernetwork_layers is not None:
-        k_in = self.to_k(hypernetwork_layers[0](context))
-        v_in = self.to_v(hypernetwork_layers[1](context))
-    else:
-        k_in = self.to_k(context)
-        v_in = self.to_v(context)
+    context_k, context_v = hypernetwork.apply_hypernetwork(shared.loaded_hypernetwork, context)
+    k_in = self.to_k(context_k)
+    v_in = self.to_v(context_v)
 
     k_in *= self.scale
 
@@ -132,14 +120,11 @@ def xformers_attention_forward(self, x, context=None, mask=None):
     h = self.heads
     q_in = self.to_q(x)
     context = default(context, x)
-    hypernetwork = shared.loaded_hypernetwork
-    hypernetwork_layers = (hypernetwork.layers if hypernetwork is not None else {}).get(context.shape[2], None)
-    if hypernetwork_layers is not None:
-        k_in = self.to_k(hypernetwork_layers[0](context))
-        v_in = self.to_v(hypernetwork_layers[1](context))
-    else:
-        k_in = self.to_k(context)
-        v_in = self.to_v(context)
+
+    context_k, context_v = hypernetwork.apply_hypernetwork(shared.loaded_hypernetwork, context)
+    k_in = self.to_k(context_k)
+    v_in = self.to_v(context_v)
+
     q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> b n h d', h=h), (q_in, k_in, v_in))
     del q_in, k_in, v_in
     out = xformers.ops.memory_efficient_attention(q, k, v, attn_bias=None)

modules/sd_models.py

@@ -149,8 +149,13 @@ def load_model_weights(model, checkpoint_info):
         model.half()
 
     devices.dtype = torch.float32 if shared.cmd_opts.no_half else torch.float16
+    devices.dtype_vae = torch.float32 if shared.cmd_opts.no_half or shared.cmd_opts.no_half_vae else torch.float16
 
     vae_file = os.path.splitext(checkpoint_file)[0] + ".vae.pt"
+
+    if not os.path.exists(vae_file) and shared.cmd_opts.vae_path is not None:
+        vae_file = shared.cmd_opts.vae_path
+
     if os.path.exists(vae_file):
         print(f"Loading VAE weights from: {vae_file}")
         vae_ckpt = torch.load(vae_file, map_location="cpu")
@@ -158,6 +163,8 @@ def load_model_weights(model, checkpoint_info):
 
         model.first_stage_model.load_state_dict(vae_dict)
 
+    model.first_stage_model.to(devices.dtype_vae)
+
     model.sd_model_hash = sd_model_hash
     model.sd_model_checkpoint = checkpoint_file
     model.sd_checkpoint_info = checkpoint_info

modules/sd_samplers.py

@@ -7,7 +7,7 @@ import inspect
 import k_diffusion.sampling
 import ldm.models.diffusion.ddim
 import ldm.models.diffusion.plms
-from modules import prompt_parser
+from modules import prompt_parser, devices, processing
 
 from modules.shared import opts, cmd_opts, state
 import modules.shared as shared
@@ -83,7 +83,7 @@ def setup_img2img_steps(p, steps=None):
 
 
 def sample_to_image(samples):
-    x_sample = shared.sd_model.decode_first_stage(samples[0:1].type(shared.sd_model.dtype))[0]
+    x_sample = processing.decode_first_stage(shared.sd_model, samples[0:1])[0]
     x_sample = torch.clamp((x_sample + 1.0) / 2.0, min=0.0, max=1.0)
     x_sample = 255. * np.moveaxis(x_sample.cpu().numpy(), 0, 2)
     x_sample = x_sample.astype(np.uint8)

modules/shared.py

@@ -13,7 +13,8 @@ import modules.memmon
 import modules.sd_models
 import modules.styles
 import modules.devices as devices
-from modules import sd_samplers, hypernetwork
+from modules import sd_samplers
+from modules.hypernetwork import hypernetwork
 from modules.paths import models_path, script_path, sd_path
 
 sd_model_file = os.path.join(script_path, 'model.ckpt')
@@ -25,9 +26,11 @@ parser.add_argument("--ckpt-dir", type=str, default=None, help="Path to director
 parser.add_argument("--gfpgan-dir", type=str, help="GFPGAN directory", default=('./src/gfpgan' if os.path.exists('./src/gfpgan') else './GFPGAN'))
 parser.add_argument("--gfpgan-model", type=str, help="GFPGAN model file name", default=None)
 parser.add_argument("--no-half", action='store_true', help="do not switch the model to 16-bit floats")
+parser.add_argument("--no-half-vae", action='store_true', help="do not switch the VAE model to 16-bit floats")
 parser.add_argument("--no-progressbar-hiding", action='store_true', help="do not hide progressbar in gradio UI (we hide it because it slows down ML if you have hardware acceleration in browser)")
 parser.add_argument("--max-batch-count", type=int, default=16, help="maximum batch count value for the UI")
 parser.add_argument("--embeddings-dir", type=str, default=os.path.join(script_path, 'embeddings'), help="embeddings directory for textual inversion (default: embeddings)")
+parser.add_argument("--hypernetwork-dir", type=str, default=os.path.join(models_path, 'hypernetworks'), help="hypernetwork directory")
 parser.add_argument("--allow-code", action='store_true', help="allow custom script execution from webui")
 parser.add_argument("--medvram", action='store_true', help="enable stable diffusion model optimizations for sacrificing a little speed for low VRM usage")
 parser.add_argument("--lowvram", action='store_true', help="enable stable diffusion model optimizations for sacrificing a lot of speed for very low VRM usage")
@@ -65,6 +68,7 @@ parser.add_argument("--autolaunch", action='store_true', help="open the webui UR
 parser.add_argument("--use-textbox-seed", action='store_true', help="use textbox for seeds in UI (no up/down, but possible to input long seeds)", default=False)
 parser.add_argument("--disable-console-progressbars", action='store_true', help="do not output progressbars to console", default=False)
 parser.add_argument("--enable-console-prompts", action='store_true', help="print prompts to console when generating with txt2img and img2img", default=False)
+parser.add_argument('--vae-path', type=str, help='Path to Variational Autoencoders model', default=None)
 parser.add_argument("--disable-safe-unpickle", action='store_true', help="disable checking pytorch models for malicious code", default=False)
 
 
@@ -80,10 +84,17 @@ parallel_processing_allowed = not cmd_opts.lowvram and not cmd_opts.medvram
 xformers_available = False
 config_filename = cmd_opts.ui_settings_file
 
-hypernetworks = hypernetwork.list_hypernetworks(os.path.join(models_path, 'hypernetworks'))
+hypernetworks = hypernetwork.list_hypernetworks(cmd_opts.hypernetwork_dir)
 loaded_hypernetwork = None
 
 
+def reload_hypernetworks():
+    global hypernetworks
+
+    hypernetworks = hypernetwork.list_hypernetworks(cmd_opts.hypernetwork_dir)
+    hypernetwork.load_hypernetwork(opts.sd_hypernetwork)
+
+
 class State:
     skipped = False
     interrupted = False
@@ -171,6 +182,7 @@ options_templates.update(options_section(('saving-images', "Saving images/grids"
 
     "use_original_name_batch": OptionInfo(False, "Use original name for output filename during batch process in extras tab"),
     "save_selected_only": OptionInfo(True, "When using 'Save' button, only save a single selected image"),
+    "do_not_add_watermark": OptionInfo(False, "Do not add watermark to images"),
 }))
 
 options_templates.update(options_section(('saving-paths', "Paths for saving"), {
@@ -224,6 +236,7 @@ options_templates.update(options_section(('sd', "Stable Diffusion"), {
     "enable_emphasis": OptionInfo(True, "Emphasis: use (text) to make model pay more attention to text and [text] to make it pay less attention"),
     "use_old_emphasis_implementation": OptionInfo(False, "Use old emphasis implementation. Can be useful to reproduce old seeds."),
     "enable_batch_seeds": OptionInfo(True, "Make K-diffusion samplers produce same images in a batch as when making a single image"),
+    "comma_padding_backtrack": OptionInfo(20, "Increase coherency by padding from the last comma within n tokens when using more than 75 tokens", gr.Slider, {"minimum": 0, "maximum": 74, "step": 1 }),
     "filter_nsfw": OptionInfo(False, "Filter NSFW content"),
     'CLIP_stop_at_last_layers': OptionInfo(1, "Stop At last layers of CLIP model", gr.Slider, {"minimum": 1, "maximum": 12, "step": 1}),
     "random_artist_categories": OptionInfo([], "Allowed categories for random artists selection when using the Roll button", gr.CheckboxGroup, {"choices": artist_db.categories()}),
@@ -236,6 +249,7 @@ options_templates.update(options_section(('interrogate', "Interrogate Options"),
     "interrogate_clip_min_length": OptionInfo(24, "Interrogate: minimum description length (excluding artists, etc..)", gr.Slider, {"minimum": 1, "maximum": 128, "step": 1}),
     "interrogate_clip_max_length": OptionInfo(48, "Interrogate: maximum description length", gr.Slider, {"minimum": 1, "maximum": 256, "step": 1}),
     "interrogate_clip_dict_limit": OptionInfo(1500, "Interrogate: maximum number of lines in text file (0 = No limit)"),
+    "interrogate_deepbooru_score_threshold": OptionInfo(0.5, "Interrogate: deepbooru score threshold", gr.Slider, {"minimum": 0, "maximum": 1, "step": 0.01}),
 }))
 
 options_templates.update(options_section(('ui', "User interface"), {
@@ -259,6 +273,7 @@ options_templates.update(options_section(('sampler-params', "Sampler parameters"
     's_churn': OptionInfo(0.0, "sigma churn", gr.Slider, {"minimum": 0.0, "maximum": 1.0, "step": 0.01}),
     's_tmin':  OptionInfo(0.0, "sigma tmin",  gr.Slider, {"minimum": 0.0, "maximum": 1.0, "step": 0.01}),
     's_noise': OptionInfo(1.0, "sigma noise", gr.Slider, {"minimum": 0.0, "maximum": 1.0, "step": 0.01}),
+    'eta_noise_seed_delta': OptionInfo(0, "Eta noise seed delta", gr.Number, {"precision": 0}),
 }))
 
 

modules/swinir_model.py

@@ -10,6 +10,7 @@ from tqdm import tqdm
 from modules import modelloader
 from modules.shared import cmd_opts, opts, device
 from modules.swinir_model_arch import SwinIR as net
+from modules.swinir_model_arch_v2 import Swin2SR as net2
 from modules.upscaler import Upscaler, UpscalerData
 
 precision_scope = (
@@ -57,22 +58,42 @@ class UpscalerSwinIR(Upscaler):
             filename = path
         if filename is None or not os.path.exists(filename):
             return None
-        model = net(
+        if filename.endswith(".v2.pth"):
+            model = net2(
             upscale=scale,
             in_chans=3,
             img_size=64,
             window_size=8,
             img_range=1.0,
-            depths=[6, 6, 6, 6, 6, 6, 6, 6, 6],
-            embed_dim=240,
-            num_heads=[8, 8, 8, 8, 8, 8, 8, 8, 8],
+            depths=[6, 6, 6, 6, 6, 6],
+            embed_dim=180,
+            num_heads=[6, 6, 6, 6, 6, 6],
             mlp_ratio=2,
             upsampler="nearest+conv",
-            resi_connection="3conv",
-        )
+            resi_connection="1conv",
+            )
+            params = None
+        else:
+            model = net(
+                upscale=scale,
+                in_chans=3,
+                img_size=64,
+                window_size=8,
+                img_range=1.0,
+                depths=[6, 6, 6, 6, 6, 6, 6, 6, 6],
+                embed_dim=240,
+                num_heads=[8, 8, 8, 8, 8, 8, 8, 8, 8],
+                mlp_ratio=2,
+                upsampler="nearest+conv",
+                resi_connection="3conv",
+            )
+            params = "params_ema"
 
         pretrained_model = torch.load(filename)
-        model.load_state_dict(pretrained_model["params_ema"], strict=True)
+        if params is not None:
+            model.load_state_dict(pretrained_model[params], strict=True)
+        else:
+            model.load_state_dict(pretrained_model, strict=True)
         if not cmd_opts.no_half:
             model = model.half()
         return model

modules/textual_inversion/dataset.py

@@ -15,11 +15,10 @@ re_tag = re.compile(r"[a-zA-Z][_\w\d()]+")
 
 
 class PersonalizedBase(Dataset):
-    def __init__(self, data_root, size=None, repeats=100, flip_p=0.5, placeholder_token="*", width=512, height=512, model=None, device=None, template_file=None):
+    def __init__(self, data_root, width, height, repeats, flip_p=0.5, placeholder_token="*", model=None, device=None, template_file=None):
 
         self.placeholder_token = placeholder_token
 
-        self.size = size
         self.width = width
         self.height = height
         self.flip = transforms.RandomHorizontalFlip(p=flip_p)

modules/textual_inversion/preprocess.py

@@ -7,8 +7,9 @@ import tqdm
 from modules import shared, images
 
 
-def preprocess(process_src, process_dst, process_flip, process_split, process_caption):
-    size = 512
+def preprocess(process_src, process_dst, process_width, process_height, process_flip, process_split, process_caption):
+    width = process_width
+    height = process_height
     src = os.path.abspath(process_src)
     dst = os.path.abspath(process_dst)
 
@@ -55,23 +56,23 @@ def preprocess(process_src, process_dst, process_flip, process_split, process_ca
         is_wide = ratio < 1 / 1.35
 
         if process_split and is_tall:
-            img = img.resize((size, size * img.height // img.width))
+            img = img.resize((width, height * img.height // img.width))
 
-            top = img.crop((0, 0, size, size))
+            top = img.crop((0, 0, width, height))
             save_pic(top, index)
 
-            bot = img.crop((0, img.height - size, size, img.height))
+            bot = img.crop((0, img.height - height, width, img.height))
             save_pic(bot, index)
         elif process_split and is_wide:
-            img = img.resize((size * img.width // img.height, size))
+            img = img.resize((width * img.width // img.height, height))
 
-            left = img.crop((0, 0, size, size))
+            left = img.crop((0, 0, width, height))
             save_pic(left, index)
 
-            right = img.crop((img.width - size, 0, img.width, size))
+            right = img.crop((img.width - width, 0, img.width, height))
             save_pic(right, index)
         else:
-            img = images.resize_image(1, img, size, size)
+            img = images.resize_image(1, img, width, height)
             save_pic(img, index)
 
         shared.state.nextjob()

modules/textual_inversion/textual_inversion.py

@@ -156,7 +156,7 @@ def create_embedding(name, num_vectors_per_token, init_text='*'):
     return fn
 
 
-def train_embedding(embedding_name, learn_rate, data_root, log_directory, steps, create_image_every, save_embedding_every, template_file):
+def train_embedding(embedding_name, learn_rate, data_root, log_directory, training_width, training_height, steps, num_repeats, create_image_every, save_embedding_every, template_file, preview_image_prompt):
     assert embedding_name, 'embedding not selected'
 
     shared.state.textinfo = "Initializing textual inversion training..."
@@ -182,7 +182,7 @@ def train_embedding(embedding_name, learn_rate, data_root, log_directory, steps,
 
     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, size=512, placeholder_token=embedding_name, model=shared.sd_model, device=devices.device, template_file=template_file)
+        ds = modules.textual_inversion.dataset.PersonalizedBase(data_root=data_root, width=training_width, height=training_height, repeats=num_repeats, placeholder_token=embedding_name, model=shared.sd_model, device=devices.device, template_file=template_file)
 
     hijack = sd_hijack.model_hijack
 
@@ -200,6 +200,9 @@ def train_embedding(embedding_name, learn_rate, data_root, log_directory, steps,
     if ititial_step > steps:
         return embedding, filename
 
+    tr_img_len = len([os.path.join(data_root, file_path) for file_path in os.listdir(data_root)])
+    epoch_len = (tr_img_len * num_repeats) + tr_img_len
+
     pbar = tqdm.tqdm(enumerate(ds), total=steps-ititial_step)
     for i, (x, text) in pbar:
         embedding.step = i + ititial_step
@@ -223,7 +226,10 @@ def train_embedding(embedding_name, learn_rate, data_root, log_directory, steps,
             loss.backward()
             optimizer.step()
 
-        pbar.set_description(f"loss: {losses.mean():.7f}")
+        epoch_num = embedding.step // epoch_len
+        epoch_step = embedding.step - (epoch_num * epoch_len) + 1
+
+        pbar.set_description(f"[Epoch {epoch_num}: {epoch_step}/{epoch_len}]loss: {losses.mean():.7f}")
 
         if embedding.step > 0 and embedding_dir is not None and embedding.step % save_embedding_every == 0:
             last_saved_file = os.path.join(embedding_dir, f'{embedding_name}-{embedding.step}.pt')
@@ -232,10 +238,14 @@ def train_embedding(embedding_name, learn_rate, data_root, log_directory, steps,
         if embedding.step > 0 and images_dir is not None and embedding.step % create_image_every == 0:
             last_saved_image = os.path.join(images_dir, f'{embedding_name}-{embedding.step}.png')
 
+            preview_text = text if preview_image_prompt == "" else preview_image_prompt
+
             p = processing.StableDiffusionProcessingTxt2Img(
                 sd_model=shared.sd_model,
-                prompt=text,
+                prompt=preview_text,
                 steps=20,
+                height=training_height,
+                width=training_width,
                 do_not_save_grid=True,
                 do_not_save_samples=True,
             )
@@ -246,7 +256,7 @@ def train_embedding(embedding_name, learn_rate, data_root, log_directory, steps,
             shared.state.current_image = image
             image.save(last_saved_image)
 
-            last_saved_image += f", prompt: {text}"
+            last_saved_image += f", prompt: {preview_text}"
 
         shared.state.job_no = embedding.step
 

modules/textual_inversion/ui.py

@@ -22,7 +22,6 @@ def preprocess(*args):
 
 
 def train_embedding(*args):
-
     try:
         sd_hijack.undo_optimizations()
 

modules/ui.py

@@ -39,6 +39,7 @@ import modules.generation_parameters_copypaste
 from modules import prompt_parser
 from modules.images import save_image
 import modules.textual_inversion.ui
+import modules.hypernetwork.ui
 import modules.images_history as img_his
 
 # this is a fix for Windows users. Without it, javascript files will be served with text/html content-type and the browser will not show any UI
@@ -312,7 +313,7 @@ def interrogate(image):
 
 
 def interrogate_deepbooru(image):
-    prompt = get_deepbooru_tags(image)
+    prompt = get_deepbooru_tags(image, opts.interrogate_deepbooru_score_threshold)
     return gr_show(True) if prompt is None else prompt
 
 
@@ -525,7 +526,7 @@ def create_ui(wrap_gradio_gpu_call):
                     denoising_strength = gr.Slider(minimum=0.0, maximum=1.0, step=0.01, label='Denoising strength', value=0.7)
 
                 with gr.Row():
-                    batch_count = gr.Slider(minimum=1, maximum=cmd_opts.max_batch_count, step=1, label='Batch count', value=1)
+                    batch_count = gr.Slider(minimum=1, step=1, label='Batch count', value=1)
                     batch_size = gr.Slider(minimum=1, maximum=8, step=1, label='Batch size', value=1)
 
                 cfg_scale = gr.Slider(minimum=1.0, maximum=30.0, step=0.5, label='CFG Scale', value=7.0)
@@ -712,7 +713,7 @@ def create_ui(wrap_gradio_gpu_call):
                     tiling = gr.Checkbox(label='Tiling', value=False)
 
                 with gr.Row():
-                    batch_count = gr.Slider(minimum=1, maximum=cmd_opts.max_batch_count, step=1, label='Batch count', value=1)
+                    batch_count = gr.Slider(minimum=1, step=1, label='Batch count', value=1)
                     batch_size = gr.Slider(minimum=1, maximum=8, step=1, label='Batch size', value=1)
 
                 with gr.Group():
@@ -964,7 +965,7 @@ def create_ui(wrap_gradio_gpu_call):
         
         extras_send_to_inpaint.click(
             fn=lambda x: image_from_url_text(x),
-            _js="extract_image_from_gallery_img2img",
+            _js="extract_image_from_gallery_inpaint",
             inputs=[result_images],
             outputs=[init_img_with_mask],
         )
@@ -1025,13 +1026,27 @@ def create_ui(wrap_gradio_gpu_call):
                             gr.HTML(value="")
 
                         with gr.Column():
-                            create_embedding = gr.Button(value="Create", variant='primary')
+                            create_embedding = gr.Button(value="Create embedding", variant='primary')
+
+                with gr.Group():
+                    gr.HTML(value="<p style='margin-bottom: 0.7em'>Create a new hypernetwork</p>")
+
+                    new_hypernetwork_name = gr.Textbox(label="Name")
+
+                    with gr.Row():
+                        with gr.Column(scale=3):
+                            gr.HTML(value="")
+
+                        with gr.Column():
+                            create_hypernetwork = gr.Button(value="Create hypernetwork", variant='primary')
 
                 with gr.Group():
                     gr.HTML(value="<p style='margin-bottom: 0.7em'>Preprocess images</p>")
 
                     process_src = gr.Textbox(label='Source directory')
                     process_dst = gr.Textbox(label='Destination directory')
+                    process_width = gr.Slider(minimum=64, maximum=2048, step=64, label="Width", value=512)
+                    process_height = gr.Slider(minimum=64, maximum=2048, step=64, label="Height", value=512)
 
                     with gr.Row():
                         process_flip = gr.Checkbox(label='Create flipped copies')
@@ -1046,24 +1061,25 @@ def create_ui(wrap_gradio_gpu_call):
                             run_preprocess = gr.Button(value="Preprocess", variant='primary')
 
                 with gr.Group():
-                    gr.HTML(value="<p style='margin-bottom: 0.7em'>Train an embedding; must specify a directory with a set of 512x512 images</p>")
+                    gr.HTML(value="<p style='margin-bottom: 0.7em'>Train an embedding; must specify a directory with a set of 1:1 ratio images</p>")
                     train_embedding_name = gr.Dropdown(label='Embedding', choices=sorted(sd_hijack.model_hijack.embedding_db.word_embeddings.keys()))
+                    train_hypernetwork_name = gr.Dropdown(label='Hypernetwork', choices=[x for x in shared.hypernetworks.keys()])
                     learn_rate = gr.Number(label='Learning rate', value=5.0e-03)
                     dataset_directory = gr.Textbox(label='Dataset directory', placeholder="Path to directory with input images")
                     log_directory = gr.Textbox(label='Log directory', placeholder="Path to directory where to write outputs", value="textual_inversion")
                     template_file = gr.Textbox(label='Prompt template file', value=os.path.join(script_path, "textual_inversion_templates", "style_filewords.txt"))
+                    training_width = gr.Slider(minimum=64, maximum=2048, step=64, label="Width", value=512)
+                    training_height = gr.Slider(minimum=64, maximum=2048, step=64, label="Height", value=512)
                     steps = gr.Number(label='Max steps', value=100000, precision=0)
+                    num_repeats = gr.Number(label='Number of repeats for a single input image per epoch', value=100, precision=0)
                     create_image_every = gr.Number(label='Save an image to log directory every N steps, 0 to disable', value=500, precision=0)
                     save_embedding_every = gr.Number(label='Save a copy of embedding to log directory every N steps, 0 to disable', value=500, precision=0)
+                    preview_image_prompt = gr.Textbox(label='Preview prompt', value="")
 
                     with gr.Row():
-                        with gr.Column(scale=2):
-                            gr.HTML(value="")
-
-                        with gr.Column():
-                            with gr.Row():
-                                interrupt_training = gr.Button(value="Interrupt")
-                                train_embedding = gr.Button(value="Train", variant='primary')
+                        interrupt_training = gr.Button(value="Interrupt")
+                        train_hypernetwork = gr.Button(value="Train Hypernetwork", variant='primary')
+                        train_embedding = gr.Button(value="Train Embedding", variant='primary')
 
             with gr.Column():
                 progressbar = gr.HTML(elem_id="ti_progressbar")
@@ -1089,12 +1105,26 @@ def create_ui(wrap_gradio_gpu_call):
             ]
         )
 
+        create_hypernetwork.click(
+            fn=modules.hypernetwork.ui.create_hypernetwork,
+            inputs=[
+                new_hypernetwork_name,
+            ],
+            outputs=[
+                train_hypernetwork_name,
+                ti_output,
+                ti_outcome,
+            ]
+        )
+
         run_preprocess.click(
             fn=wrap_gradio_gpu_call(modules.textual_inversion.ui.preprocess, extra_outputs=[gr.update()]),
             _js="start_training_textual_inversion",
             inputs=[
                 process_src,
                 process_dst,
+                process_width,
+                process_height,
                 process_flip,
                 process_split,
                 process_caption,
@@ -1113,10 +1143,34 @@ def create_ui(wrap_gradio_gpu_call):
                 learn_rate,
                 dataset_directory,
                 log_directory,
+                training_width,
+                training_height,
+                steps,
+                num_repeats,
+                create_image_every,
+                save_embedding_every,
+                template_file,
+                preview_image_prompt,
+            ],
+            outputs=[
+                ti_output,
+                ti_outcome,
+            ]
+        )
+
+        train_hypernetwork.click(
+            fn=wrap_gradio_gpu_call(modules.hypernetwork.ui.train_hypernetwork, extra_outputs=[gr.update()]),
+            _js="start_training_textual_inversion",
+            inputs=[
+                train_hypernetwork_name,
+                learn_rate,
+                dataset_directory,
+                log_directory,
                 steps,
                 create_image_every,
                 save_embedding_every,
                 template_file,
+                preview_image_prompt,
             ],
             outputs=[
                 ti_output,
@@ -1130,6 +1184,7 @@ def create_ui(wrap_gradio_gpu_call):
             outputs=[],
         )
 
+
     def create_setting_component(key):
         def fun():
             return opts.data[key] if key in opts.data else opts.data_labels[key].default
@@ -1335,7 +1390,7 @@ Requested path was: {f}
         
         with gr.Tabs() as tabs:
             for interface, label, ifid in interfaces:
-                with gr.TabItem(label, id=ifid):
+                with gr.TabItem(label, id=ifid, elem_id='tab_' + ifid):
                     interface.render()
         
         if os.path.exists(os.path.join(script_path, "notification.mp3")):