update requirements

CHANGELOG.md

@@ -1,6 +1,6 @@
 # Change Log for SD.Next
 
-## Update for 2024-05-17
+## Update for 2024-05-32
 
 - **Features**:
   - **ModernUI** preview of the new [ModernUI](https://github.com/BinaryQuantumSoul/sdnext-modernui)  
@@ -59,12 +59,12 @@
     *Note*: if there is a `JSON` file with the same name as the model it will be used as Unet config, otherwise Unet config from currently loaded model will be used  
   - [PixArt-Σ](https://pixart-alpha.github.io/PixArt-sigma-project/)
     pixart-Σ is a high end diffusion Transformer model (DiT) with a T5 encoder/decoder capable of directly generating images at 4K resolution  
-    to use, simply select from *networks -> models -> PixArt-Σ*  
+    to use, simply select from *networks -> models -> reference -> PixArt-Σ*  
     *note*: this is a very large model at ~22GB  
     set parameters: *sampler: Default*  
   - [SDXS](https://github.com/IDKiro/sdxs)
     sdxs is an extremely fast 1-step generation consistency model that also uses TAESD as quick VAE out-of-the-box  
-    to use, simply select from *networks -> models -> SDXS*  
+    to use, simply select from *networks -> models -> reference -> SDXS*  
     set parameters: *sampler: CMSI, steps: 1, cfg_scale: 0.0*
   - [Hyper-SD](https://huggingface.co/ByteDance/Hyper-SD)  
     sd15 and sdxl 1-step, 2-step, 4-step and 8-step optimized models using lora  

TODO.md

@@ -12,7 +12,7 @@ Requires `diffusers-0.28.0.dev0`:
 
 ## Dev Release Notes
 
-New [SD.Next](https://github.com/vladmandic/automatic) release has been baking in `dev` for a longer than usual, but changes are massive - over 300 commits...
+New [SD.Next](https://github.com/vladmandic/automatic) release has been baking in `dev` for a longer than usual, but changes are massive - over 320 commits...
 
 Starting with the new UI - yup, this version ships with a *preview* of the new [ModernUI](https://github.com/BinaryQuantumSoul/sdnext-modernui)  
 For details on how to enable and use it, see [Home](https://github.com/BinaryQuantumSoul/sdnext-modernui) and [WiKi](https://github.com/vladmandic/automatic/wiki/Themes)  
@@ -43,7 +43,7 @@ Thanks to @BinaryQuantumSoul for his hard work on this project!
 ### New models
 
 While still waiting for *Stable Diffusion 3.0*, there have been some significant models released in the meantime:
-- [PixArt-Σ](https://pixart-alpha.github.io/PixArt-sigma-project/), high end diffusion transformer model (DiT) capable of directly generating images at 4K resolution  
+- [PixArt-Σ](https://pixart-alpha.github.io/PixArt-sigma-project/), high end diffusion transformer model (*DiT*) capable of directly generating images at 4K resolution  
 - [SDXS](https://github.com/IDKiro/sdxs), extremely fast 1-step generation consistency model  
 - [Hyper-SD](https://huggingface.co/ByteDance/Hyper-SD), 1-step, 2-step, 4-step and 8-step optimized models  
 
@@ -52,7 +52,7 @@ While still waiting for *Stable Diffusion 3.0*, there have been some significant
 Given huge number of changes with *+3443/-3342* commits diff (at the time of fork detach) over the past year,  
 a completely different backend/engine and a change of focus, it is time to give credit to original [author](https://github.com/auTOMATIC1111),  and move on!  
 
-And best place to post questions is on our [Discord](https://discord.gg/VjvR2tabEX) server which now has over 2k active members!
+And best place to post questions is on our [Discord](https://discord.gg/VjvR2tabEX) server which now has **over 2k active members**!
 
 For more details see: [Changelog](https://github.com/vladmandic/automatic/blob/dev/CHANGELOG.md) | [ReadMe](https://github.com/vladmandic/automatic) | [Wiki](https://github.com/vladmandic/automatic/wiki) | [Discord](https://discord.gg/VjvR2tabEX)
 
@@ -69,6 +69,10 @@ For more details see: [Changelog](https://github.com/vladmandic/automatic/blob/d
 - diffusers public callbacks  
 - include reference styles
 - lora: sc lora, dora, etc
+- [SDXL Flash Mini](https://huggingface.co/sd-community/sdxl-flash-mini)  
+  SDXL type that weighs less, consumes less video memory, and the quality has not dropped much  
+  to use, simply select from *networks -> models -> reference -> SDXL Flash Mini*  
+  recommended parameters: steps: 6-9, cfg scale: 2.5-3.5, sampler: DPM++ SDE  
 
 ### Missing
 

html/reference.json

@@ -50,6 +50,13 @@
     "desc": "SDXS: Real-Time One-Step Latent Diffusion Models with Image Conditions",
     "extras": "width: 512, height: 512, sampler: CMSI, steps: 1, cfg_scale: 0.0"
   },
+  "SDXL Flash Mini": {
+    "path": "SDXL-Flash_Mini.safetensors@https://huggingface.co/sd-community/sdxl-flash-mini/resolve/main/SDXL-Flash_Mini.safetensors?download=true",
+    "preview": "SDXL-Flash_Mini.jpg",
+    "desc": "Introducing the new fast model SDXL Flash (Mini), we learned that all fast XL models work fast, but the quality decreases, and we also made a fast model, but it is not as fast as LCM, Turbo, Lightning and Hyper, but the quality is higher.",
+    "extras": "width: 2048, height: 1024, sampler: DEIS, steps: 40, cfg_scale: 6.0",
+    "experimental": true
+  },
 
   "RunwayML SD 1.5": {
     "original": true,

modules/sd_cascade.py

@@ -48,7 +48,7 @@ def load_text_encoder(path):
 
 
 def load_prior(path, config_file="default"):
-    from diffusers import StableCascadeUNet
+    from diffusers.models.unets import StableCascadeUNet
     prior_text_encoder = None
 
     if config_file == "default":
@@ -71,7 +71,8 @@ def load_prior(path, config_file="default"):
 
 
 def load_cascade_combined(checkpoint_info, diffusers_load_config):
-    from diffusers import StableCascadeUNet, StableCascadeDecoderPipeline, StableCascadePriorPipeline, StableCascadeCombinedPipeline
+    from diffusers import StableCascadeDecoderPipeline, StableCascadePriorPipeline, StableCascadeCombinedPipeline
+    from diffusers.models.unets import StableCascadeUNet
     from modules.sd_unet import unet_dict
 
     diffusers_load_config.pop("vae", None)

modules/sd_models.py

@@ -789,11 +789,27 @@ def move_model(model, device=None, force=False):
     if getattr(model, 'has_accelerate', False) and not force:
         return
     try:
-        model.to(device)
+        try:
+            model.to(device)
+        except Exception as e0:
+            if 'Cannot copy out of meta tensor' in str(e0):
+                if hasattr(model, "components"):
+                    for _name, component in model.components.items():
+                        if hasattr(component, 'modules'):
+                            for module in component.modules():
+                                try:
+                                    module.to(device)
+                                except Exception as e2:
+                                    if 'Cannot copy out of meta tensor' in str(e2):
+                                        if os.environ.get('SD_MOVE_DEBUG', None):
+                                            shared.log.warning(f'Model move meta: module={module.__class__}')
+                                        module.to_empty(device=device)
+            else:
+                raise e0
         if hasattr(model, "prior_pipe"):
             model.prior_pipe.to(device)
-    except Exception as e:
-        shared.log.error(f'Model move: device={device} {e}')
+    except Exception as e1:
+        shared.log.error(f'Model move: device={device} {e1}')
     devices.torch_gc()
 
 

modules/ui_extra_networks_checkpoints.py

@@ -22,6 +22,12 @@ class ExtraNetworksPageCheckpoints(ui_extra_networks.ExtraNetworksPage):
                 url = v.get('alt', None) or v['path']
             else:
                 url = v['path']
+            experimental = v.get('experimental', False)
+            if experimental:
+                if shared.cmd_opts.experimental:
+                    shared.log.debug(f'Extra networks experimental: model="{k}"')
+                else:
+                    continue
             name = os.path.join(reference_dir, k)
             preview = v.get('preview', v['path'])
             yield {

requirements.txt

@@ -44,7 +44,7 @@ clip-interrogator==0.6.0
 antlr4-python3-runtime==4.9.3
 requests==2.31.0
 tqdm==4.66.4
-accelerate==0.30.0
+accelerate==0.30.1
 opencv-contrib-python-headless==4.9.0.80
 diffusers==0.27.2
 einops==0.4.1
@@ -57,7 +57,7 @@ pandas
 protobuf==3.20.3
 pytorch_lightning==1.9.4
 tokenizers==0.19.1
-transformers==4.40.2
+transformers==4.41.0
 urllib3==1.26.18
 Pillow==10.3.0
 timm==0.9.16