Merge remote-tracking branch 'upstream' into feat-change-img-count

CHANGES.md

@@ -17,6 +17,9 @@
 - **Major rewrite of the code** - We've switched to using diffusers under-the-hood, which allows us to release new features faster, and focus on making the UI and installer even easier to use.
 
 ### Detailed changelog
+* 3.0.9c - 6 Feb 2025 - (Internal code change) Remove hardcoded references to `torch.cuda`, and replace with torchruntime's device utilities.
+* 3.0.9b - 28 Jan 2025 - Fix a bug affecting older versions of Easy Diffusion, which tried to upgrade to an incompatible version of PyTorch.
+* 3.0.9b - 4 Jan 2025 - Replace the use of WMIC (deprecated) with a powershell call.
 * 3.0.9 - 28 May 2024 - Slider for controlling the strength of controlnets.
 * 3.0.8 - 27 May 2024 - SDXL ControlNets for Img2Img and Inpainting.
 * 3.0.7 - 11 Dec 2023 - Setting to enable/disable VAE tiling (in the Image Settings panel). Sometimes VAE tiling reduces the quality of the image, so this setting will help control that.

README.md

@@ -3,7 +3,7 @@
 
 Does not require technical knowledge, does not require pre-installed software. 1-click install, powerful features, friendly community.
 
-️‍🔥🎉 **New!** Support for SDXL, ControlNet, multiple LoRA files, embeddings (and a lot more) have been added!
+️‍🔥🎉 **New!** Support for Flux has been added in the beta branch (v3.5 engine)!
 
 [Installation guide](#installation) | [Troubleshooting guide](https://github.com/easydiffusion/easydiffusion/wiki/Troubleshooting) | [User guide](https://github.com/easydiffusion/easydiffusion/wiki) | <sub>[![Discord Server](https://img.shields.io/discord/1014774730907209781?label=Discord)](https://discord.com/invite/u9yhsFmEkB)</sub> <sup>(for support queries, and development discussions)</sup>
 
@@ -21,15 +21,15 @@ Click the download button for your operating system:
 </p>
 
 **Hardware requirements:**
-- **Windows:** NVIDIA graphics card¹ (minimum 2 GB RAM), or run on your CPU.
+- **Windows:** NVIDIA¹ or AMD graphics card (minimum 2 GB RAM), or run on your CPU.
 - **Linux:** NVIDIA¹ or AMD² graphics card (minimum 2 GB RAM), or run on your CPU.
-- **Mac:** M1 or M2, or run on your CPU.
+- **Mac:** M1/M2/M3/M4 or AMD graphics card (Intel Mac), or run on your CPU.
 - Minimum 8 GB of system RAM.
 - Atleast 25 GB of space on the hard disk.
 
 ¹) [CUDA Compute capability](https://en.wikipedia.org/wiki/CUDA#GPUs_supported) level of 3.7 or higher required.
 
-²) ROCm 5.2 support required.
+²) ROCm 5.2 (or newer) support required.
 
 The installer will take care of whatever is needed. If you face any problems, you can join the friendly [Discord community](https://discord.com/invite/u9yhsFmEkB) and ask for assistance.
 

scripts/Developer Console.cmd

@@ -4,7 +4,7 @@ echo "Opening Stable Diffusion UI - Developer Console.." & echo.
 
 cd /d %~dp0
 
-set PATH=C:\Windows\System32;%PATH%
+set PATH=C:\Windows\System32;C:\Windows\System32\WindowsPowerShell\v1.0;%PATH%
 
 @rem set legacy and new installer's PATH, if they exist
 if exist "installer" set PATH=%cd%\installer;%cd%\installer\Library\bin;%cd%\installer\Scripts;%cd%\installer\Library\usr\bin;%PATH%
@@ -26,6 +26,7 @@ call conda --version
 echo.
 echo COMSPEC=%COMSPEC%
 echo.
+powershell -Command "(Get-WmiObject Win32_VideoController | Select-Object Name, AdapterRAM, DriverDate, DriverVersion)"
 
 @rem activate the legacy environment (if present) and set PYTHONPATH
 if exist "installer_files\env" (

scripts/Start Stable Diffusion UI.cmd

@@ -3,7 +3,7 @@
 cd /d %~dp0
 echo Install dir: %~dp0
 
-set PATH=C:\Windows\System32;%PATH%
+set PATH=C:\Windows\System32;C:\Windows\System32\WindowsPowerShell\v1.0;%PATH%
 set PYTHONHOME=
 
 if exist "on_sd_start.bat" (
@@ -39,7 +39,7 @@ call where conda
 call conda --version
 echo .
 echo COMSPEC=%COMSPEC%
-wmic path win32_VideoController get name,AdapterRAM,DriverDate,DriverVersion
+powershell -Command "(Get-WmiObject Win32_VideoController | Select-Object Name, AdapterRAM, DriverDate, DriverVersion)"
 
 @rem Download the rest of the installer and UI
 call scripts\on_env_start.bat

scripts/on_sd_start.bat

@@ -70,6 +70,8 @@ echo PYTHONPATH=%PYTHONPATH%
 call where python
 call python --version
 
+call python -m pip install -q torchruntime
+
 call python scripts\check_modules.py --launch-uvicorn
 pause
 exit /b

scripts/on_sd_start.sh

@@ -46,6 +46,8 @@ fi
 if [ -e "src" ]; then mv src src-old; fi
 if [ -e "ldm" ]; then mv ldm ldm-old; fi
 
+python -m pip install -q torchruntime
+
 cd ..
 # Download the required packages
 python scripts/check_modules.py --launch-uvicorn

ui/easydiffusion/app.py

@@ -54,8 +54,7 @@ OUTPUT_DIRNAME = "Stable Diffusion UI"  # in the user's home folder
 PRESERVE_CONFIG_VARS = ["FORCE_FULL_PRECISION"]
 TASK_TTL = 15 * 60  # Discard last session's task timeout
 APP_CONFIG_DEFAULTS = {
-    # auto: selects the cuda device with the most free memory, cuda: use the currently active cuda device.
-    "render_devices": "auto",  # valid entries: 'auto', 'cpu' or 'cuda:N' (where N is a GPU index)
+    "render_devices": "auto",
     "update_branch": "main",
     "ui": {
         "open_browser_on_start": True,

ui/easydiffusion/device_manager.py

@@ -6,6 +6,15 @@ import traceback
 import torch
 from easydiffusion.utils import log
 
+from torchruntime.utils import (
+    get_installed_torch_platform,
+    get_device,
+    get_device_count,
+    get_device_name,
+    SUPPORTED_BACKENDS,
+)
+from sdkit.utils import mem_get_info, is_cpu_device, has_half_precision_bug
+
 """
 Set `FORCE_FULL_PRECISION` in the environment variables, or in `config.bat`/`config.sh` to set full precision (i.e. float32).
 Otherwise the models will load at half-precision (i.e. float16).
@@ -22,33 +31,15 @@ mem_free_threshold = 0
 
 def get_device_delta(render_devices, active_devices):
     """
-    render_devices: 'cpu', or 'auto', or 'mps' or ['cuda:N'...]
-    active_devices: ['cpu', 'mps', 'cuda:N'...]
+    render_devices: 'auto' or backends listed in `torchruntime.utils.SUPPORTED_BACKENDS`
+    active_devices: [backends listed in `torchruntime.utils.SUPPORTED_BACKENDS`]
     """
 
-    if render_devices in ("cpu", "auto", "mps"):
-        render_devices = [render_devices]
-    elif render_devices is not None:
-        if isinstance(render_devices, str):
-            render_devices = [render_devices]
-        if isinstance(render_devices, list) and len(render_devices) > 0:
-            render_devices = list(filter(lambda x: x.startswith("cuda:") or x == "mps", render_devices))
-            if len(render_devices) == 0:
-                raise Exception(
-                    'Invalid render_devices value in config.json. Valid: {"render_devices": ["cuda:0", "cuda:1"...]}, or {"render_devices": "cpu"} or {"render_devices": "mps"} or {"render_devices": "auto"}'
-                )
+    render_devices = render_devices or "auto"
+    render_devices = [render_devices] if isinstance(render_devices, str) else render_devices
 
-            render_devices = list(filter(lambda x: is_device_compatible(x), render_devices))
-            if len(render_devices) == 0:
-                raise Exception(
-                    "Sorry, none of the render_devices configured in config.json are compatible with Stable Diffusion"
-                )
-        else:
-            raise Exception(
-                'Invalid render_devices value in config.json. Valid: {"render_devices": ["cuda:0", "cuda:1"...]}, or {"render_devices": "cpu"} or {"render_devices": "auto"}'
-            )
-    else:
-        render_devices = ["auto"]
+    # check for backend support
+    validate_render_devices(render_devices)
 
     if "auto" in render_devices:
         render_devices = auto_pick_devices(active_devices)
@@ -64,47 +55,39 @@ def get_device_delta(render_devices, active_devices):
     return devices_to_start, devices_to_stop
 
 
-def is_mps_available():
-    return (
-        platform.system() == "Darwin"
-        and hasattr(torch.backends, "mps")
-        and torch.backends.mps.is_available()
-        and torch.backends.mps.is_built()
-    )
+def validate_render_devices(render_devices):
+    supported_backends = ("auto",) + SUPPORTED_BACKENDS
+    unsupported_render_devices = [d for d in render_devices if not d.lower().startswith(supported_backends)]
 
-
-def is_cuda_available():
-    return torch.cuda.is_available()
+    if unsupported_render_devices:
+        raise ValueError(
+            f"Invalid render devices in config: {unsupported_render_devices}. Valid render devices: {supported_backends}"
+        )
 
 
 def auto_pick_devices(currently_active_devices):
     global mem_free_threshold
 
-    if is_mps_available():
-        return ["mps"]
+    torch_platform_name = get_installed_torch_platform()[0]
 
-    if not is_cuda_available():
-        return ["cpu"]
-
-    device_count = torch.cuda.device_count()
-    if device_count == 1:
-        return ["cuda:0"] if is_device_compatible("cuda:0") else ["cpu"]
+    if is_cpu_device(torch_platform_name):
+        return [torch_platform_name]
 
+    device_count = get_device_count()
     log.debug("Autoselecting GPU. Using most free memory.")
     devices = []
-    for device in range(device_count):
-        device = f"cuda:{device}"
-        if not is_device_compatible(device):
-            continue
+    for device_id in range(device_count):
+        device_id = f"{torch_platform_name}:{device_id}" if device_count > 1 else torch_platform_name
+        device = get_device(device_id)
 
-        mem_free, mem_total = torch.cuda.mem_get_info(device)
+        mem_free, mem_total = mem_get_info(device)
         mem_free /= float(10**9)
         mem_total /= float(10**9)
-        device_name = torch.cuda.get_device_name(device)
+        device_name = get_device_name(device)
         log.debug(
-            f"{device} detected: {device_name} - Memory (free/total): {round(mem_free, 2)}Gb / {round(mem_total, 2)}Gb"
+            f"{device_id} detected: {device_name} - Memory (free/total): {round(mem_free, 2)}Gb / {round(mem_total, 2)}Gb"
         )
-        devices.append({"device": device, "device_name": device_name, "mem_free": mem_free})
+        devices.append({"device": device_id, "device_name": device_name, "mem_free": mem_free})
 
     devices.sort(key=lambda x: x["mem_free"], reverse=True)
     max_mem_free = devices[0]["mem_free"]
@@ -117,69 +100,45 @@ def auto_pick_devices(currently_active_devices):
     #    always be very low (since their VRAM contains the model).
     #    These already-running devices probably aren't terrible, since they were picked in the past.
     #    Worst case, the user can restart the program and that'll get rid of them.
-    devices = list(
-        filter(
-            (lambda x: x["mem_free"] > mem_free_threshold or x["device"] in currently_active_devices),
-            devices,
-        )
-    )
-    devices = list(map(lambda x: x["device"], devices))
+    devices = [
+        x["device"] for x in devices if x["mem_free"] >= mem_free_threshold or x["device"] in currently_active_devices
+    ]
     return devices
 
 
-def device_init(context, device):
-    """
-    This function assumes the 'device' has already been verified to be compatible.
-    `get_device_delta()` has already filtered out incompatible devices.
-    """
+def device_init(context, device_id):
+    context.device = device_id
 
-    validate_device_id(device, log_prefix="device_init")
-
-    if "cuda" not in device:
-        context.device = device
+    if is_cpu_device(context.torch_device):
         context.device_name = get_processor_name()
         context.half_precision = False
-        log.debug(f"Render device available as {context.device_name}")
-        return
+    else:
+        context.device_name = get_device_name(context.torch_device)
 
-    context.device_name = torch.cuda.get_device_name(device)
-    context.device = device
+        # Some graphics cards have bugs in their firmware that prevent image generation at half precision
+        if needs_to_force_full_precision(context.device_name):
+            log.warn(f"forcing full precision on this GPU, to avoid corrupted images. GPU: {context.device_name}")
+            context.half_precision = False
 
-    # Force full precision on 1660 and 1650 NVIDIA cards to avoid creating green images
-    if needs_to_force_full_precision(context):
-        log.warn(f"forcing full precision on this GPU, to avoid green images. GPU detected: {context.device_name}")
-        # Apply force_full_precision now before models are loaded.
-        context.half_precision = False
-
-    log.info(f'Setting {device} as active, with precision: {"half" if context.half_precision else "full"}')
-    torch.cuda.device(device)
+    log.info(f'Setting {device_id} as active, with precision: {"half" if context.half_precision else "full"}')
 
 
-def needs_to_force_full_precision(context):
+def needs_to_force_full_precision(device_name):
     if "FORCE_FULL_PRECISION" in os.environ:
         return True
 
-    device_name = context.device_name.lower()
-    return (
-        ("nvidia" in device_name or "geforce" in device_name or "quadro" in device_name)
-        and (
-            " 1660" in device_name
-            or " 1650" in device_name
-            or " 1630" in device_name
-            or " t400" in device_name
-            or " t550" in device_name
-            or " t600" in device_name
-            or " t1000" in device_name
-            or " t1200" in device_name
-            or " t2000" in device_name
-        )
-    ) or ("tesla k40m" in device_name)
+    return has_half_precision_bug(device_name.lower())
 
 
 def get_max_vram_usage_level(device):
-    if "cuda" in device:
-        _, mem_total = torch.cuda.mem_get_info(device)
-    else:
+    "Expects a torch.device as the argument"
+
+    if is_cpu_device(device):
+        return "high"
+
+    _, mem_total = mem_get_info(device)
+
+    if mem_total < 0.001:  # probably a torch platform without a mem_get_info() implementation
         return "high"
 
     mem_total /= float(10**9)
@@ -191,51 +150,6 @@ def get_max_vram_usage_level(device):
     return "high"
 
 
-def validate_device_id(device, log_prefix=""):
-    def is_valid():
-        if not isinstance(device, str):
-            return False
-        if device == "cpu" or device == "mps":
-            return True
-        if not device.startswith("cuda:") or not device[5:].isnumeric():
-            return False
-        return True
-
-    if not is_valid():
-        raise EnvironmentError(
-            f"{log_prefix}: device id should be 'cpu', 'mps', or 'cuda:N' (where N is an integer index for the GPU). Got: {device}"
-        )
-
-
-def is_device_compatible(device):
-    """
-    Returns True/False, and prints any compatibility errors
-    """
-    # static variable "history".
-    is_device_compatible.history = getattr(is_device_compatible, "history", {})
-    try:
-        validate_device_id(device, log_prefix="is_device_compatible")
-    except:
-        log.error(str(e))
-        return False
-
-    if device in ("cpu", "mps"):
-        return True
-    # Memory check
-    try:
-        _, mem_total = torch.cuda.mem_get_info(device)
-        mem_total /= float(10**9)
-        if mem_total < 1.9:
-            if is_device_compatible.history.get(device) == None:
-                log.warn(f"GPU {device} with less than 2 GB of VRAM is not compatible with Stable Diffusion")
-                is_device_compatible.history[device] = 1
-            return False
-    except RuntimeError as e:
-        log.error(str(e))
-        return False
-    return True
-
-
 def get_processor_name():
     try:
         import subprocess
@@ -243,7 +157,8 @@ def get_processor_name():
         if platform.system() == "Windows":
             return platform.processor()
         elif platform.system() == "Darwin":
-            os.environ["PATH"] = os.environ["PATH"] + os.pathsep + "/usr/sbin"
+            if "/usr/sbin" not in os.environ["PATH"].split(os.pathsep):
+                os.environ["PATH"] = os.environ["PATH"] + os.pathsep + "/usr/sbin"
             command = "sysctl -n machdep.cpu.brand_string"
             return subprocess.check_output(command, shell=True).decode().strip()
         elif platform.system() == "Linux":

ui/easydiffusion/model_manager.py

@@ -37,7 +37,7 @@ MODEL_EXTENSIONS = {
 }
 DEFAULT_MODELS = {
     "stable-diffusion": [
-        {"file_name": "sd-v1-5.safetensors", "model_id": "1.5-pruned-emaonly-fp16"},
+        {"file_name": "sd-v1-4.ckpt", "model_id": "1.4"},
     ],
     "gfpgan": [
         {"file_name": "GFPGANv1.4.pth", "model_id": "1.4"},
@@ -76,7 +76,7 @@ def load_default_models(context: Context):
                 scan_model=context.model_paths[model_type] != None
                 and not context.model_paths[model_type].endswith(".safetensors"),
             )
-            if model_type in context.model_load_errors:
+            if hasattr(context, "model_load_errors") and model_type in context.model_load_errors:
                 del context.model_load_errors[model_type]
         except Exception as e:
             log.error(f"[red]Error while loading {model_type} model: {context.model_paths[model_type]}[/red]")
@@ -88,6 +88,8 @@ def load_default_models(context: Context):
                 log.exception(e)
             del context.model_paths[model_type]
 
+            if not hasattr(context, "model_load_errors"):
+                context.model_load_errors = {}
             context.model_load_errors[model_type] = str(e)  # storing the entire Exception can lead to memory leaks
 
 
@@ -179,11 +181,13 @@ def reload_models_if_necessary(context: Context, models_data: ModelsData, models
         extra_params = models_data.model_params.get(model_type, {})
         try:
             action_fn(context, model_type, scan_model=False, **extra_params)  # we've scanned them already
-            if model_type in context.model_load_errors:
+            if hasattr(context, "model_load_errors") and model_type in context.model_load_errors:
                 del context.model_load_errors[model_type]
         except Exception as e:
             log.exception(e)
             if action_fn == load_model:
+                if not hasattr(context, "model_load_errors"):
+                    context.model_load_errors = {}
                 context.model_load_errors[model_type] = str(e)  # storing the entire Exception can lead to memory leaks
 
 
@@ -207,7 +211,7 @@ def resolve_model_paths(models_data: ModelsData):
 
 def fail_if_models_did_not_load(context: Context):
     for model_type in KNOWN_MODEL_TYPES:
-        if model_type in context.model_load_errors:
+        if hasattr(context, "model_load_errors") and model_type in context.model_load_errors:
             e = context.model_load_errors[model_type]
             raise Exception(f"Could not load the {model_type} model! Reason: " + e)
 

ui/easydiffusion/server.py

@@ -196,11 +196,13 @@ def set_app_config_internal(req: SetAppConfigRequest):
 
 
 def update_render_devices_in_config(config, render_devices):
-    if render_devices not in ("cpu", "auto") and not render_devices.startswith("cuda:"):
-        raise HTTPException(status_code=400, detail=f"Invalid render device requested: {render_devices}")
+    from easydiffusion.device_manager import validate_render_devices
 
-    if render_devices.startswith("cuda:"):
+    try:
         render_devices = render_devices.split(",")
+        validate_render_devices(render_devices)
+    except ValueError as e:
+        raise HTTPException(status_code=400, detail=str(e))
 
     config["render_devices"] = render_devices
 

ui/easydiffusion/task_manager.py

@@ -21,6 +21,9 @@ from easydiffusion.tasks import Task
 from easydiffusion.utils import log
 from sdkit.utils import gc
 
+from torchruntime.utils import get_device_count, get_device, get_device_name, get_installed_torch_platform
+from sdkit.utils import is_cpu_device, mem_get_info
+
 THREAD_NAME_PREFIX = ""
 ERR_LOCK_FAILED = " failed to acquire lock within timeout."
 LOCK_TIMEOUT = 15  # Maximum locking time in seconds before failing a task.
@@ -329,34 +332,33 @@ def get_devices():
         "active": {},
     }
 
-    def get_device_info(device):
-        if device in ("cpu", "mps"):
+    def get_device_info(device_id):
+        if is_cpu_device(device_id):
             return {"name": device_manager.get_processor_name()}
 
-        mem_free, mem_total = torch.cuda.mem_get_info(device)
+        device = get_device(device_id)
+
+        mem_free, mem_total = mem_get_info(device)
         mem_free /= float(10**9)
         mem_total /= float(10**9)
 
         return {
-            "name": torch.cuda.get_device_name(device),
+            "name": get_device_name(device),
             "mem_free": mem_free,
             "mem_total": mem_total,
             "max_vram_usage_level": device_manager.get_max_vram_usage_level(device),
         }
 
     # list the compatible devices
-    cuda_count = torch.cuda.device_count()
-    for device in range(cuda_count):
-        device = f"cuda:{device}"
-        if not device_manager.is_device_compatible(device):
-            continue
+    torch_platform_name = get_installed_torch_platform()[0]
+    device_count = get_device_count()
+    for device_id in range(device_count):
+        device_id = f"{torch_platform_name}:{device_id}" if device_count > 1 else torch_platform_name
 
-        devices["all"].update({device: get_device_info(device)})
+        devices["all"].update({device_id: get_device_info(device_id)})
 
-    if device_manager.is_mps_available():
-        devices["all"].update({"mps": get_device_info("mps")})
-
-    devices["all"].update({"cpu": get_device_info("cpu")})
+    if torch_platform_name != "cpu":
+        devices["all"].update({"cpu": get_device_info("cpu")})
 
     # list the activated devices
     if not manager_lock.acquire(blocking=True, timeout=LOCK_TIMEOUT):
@@ -368,8 +370,8 @@ def get_devices():
             weak_data = weak_thread_data.get(rthread)
             if not weak_data or not "device" in weak_data or not "device_name" in weak_data:
                 continue
-            device = weak_data["device"]
-            devices["active"].update({device: get_device_info(device)})
+            device_id = weak_data["device"]
+            devices["active"].update({device_id: get_device_info(device_id)})
     finally:
         manager_lock.release()
 
@@ -427,12 +429,6 @@ def start_render_thread(device):
 
 
 def stop_render_thread(device):
-    try:
-        device_manager.validate_device_id(device, log_prefix="stop_render_thread")
-    except:
-        log.error(traceback.format_exc())
-        return False
-
     if not manager_lock.acquire(blocking=True, timeout=LOCK_TIMEOUT):
         raise Exception("stop_render_thread" + ERR_LOCK_FAILED)
     log.info(f"Stopping Rendering Thread on device: {device}")

ui/easydiffusion/types.py

@@ -20,8 +20,8 @@ class GenerateImageRequest(BaseModel):
     control_image: Any = None
     control_alpha: Union[float, List[float]] = None
     prompt_strength: float = 0.8
-    preserve_init_image_color_profile = False
-    strict_mask_border = False
+    preserve_init_image_color_profile: bool = False
+    strict_mask_border: bool = False
 
     sampler_name: str = None  # "ddim", "plms", "heun", "euler", "euler_a", "dpm2", "dpm2_a", "lms"
     hypernetwork_strength: float = 0
@@ -100,7 +100,7 @@ class MergeRequest(BaseModel):
     model1: str = None
     ratio: float = None
     out_path: str = "mix"
-    use_fp16 = True
+    use_fp16: bool = True
 
 
 class Image:

ui/index.html

@@ -35,7 +35,7 @@
             <h1>
                 <img id="logo_img" src="/media/images/icon-512x512.png" >
                 Easy Diffusion
-                <small><span id="version">v3.0.9</span> <span id="updateBranchLabel"></span></small>
+                <small><span id="version">v3.0.9c</span> <span id="updateBranchLabel"></span></small>
             </h1>
         </div>
         <div id="server-status">

ui/media/js/auto-save.js

@@ -178,23 +178,6 @@ function loadSettings() {
             }
         })
         CURRENTLY_LOADING_SETTINGS = false
-    } else if (localStorage.length < 2) {
-        // localStorage is too short for OldSettings
-        // So this is likely the first time Easy Diffusion is running.
-        // Initialize vram_usage_level based on the available VRAM
-        function initGPUProfile(event) {
-            if (
-                "detail" in event &&
-                "active" in event.detail &&
-                "cuda:0" in event.detail.active &&
-                event.detail.active["cuda:0"].mem_total < 4.5
-            ) {
-                vramUsageLevelField.value = "low"
-                vramUsageLevelField.dispatchEvent(new Event("change"))
-            }
-            document.removeEventListener("system_info_update", initGPUProfile)
-        }
-        document.addEventListener("system_info_update", initGPUProfile)
     } else {
         CURRENTLY_LOADING_SETTINGS = true
         tryLoadOldSettings()

ui/media/js/main.js

@@ -611,6 +611,13 @@ function onUseAsInputClick(req, img) {
     initImagePreview.src = imgData
 
     maskSetting.checked = false
+
+    //Force the image settings size to match the input, as inpaint currently only works correctly
+    //if input image and generate sizes match.
+    addImageSizeOption(img.naturalWidth);
+    addImageSizeOption(img.naturalHeight);
+    widthField.value = img.naturalWidth;
+    heightField.value = img.naturalHeight;
 }
 
 function onUseForControlnetClick(req, img) {

ui/media/js/parameters.js

@@ -642,7 +642,7 @@ function setDeviceInfo(devices) {
 
     function ID_TO_TEXT(d) {
         let info = devices.all[d]
-        if ("mem_free" in info && "mem_total" in info) {
+        if ("mem_free" in info && "mem_total" in info && info["mem_total"] > 0) {
             return `${info.name} <small>(${d}) (${info.mem_free.toFixed(1)}Gb free / ${info.mem_total.toFixed(
                 1
             )} Gb total)</small>`

ui/plugins/ui/snow.plugin.js

@@ -0,0 +1,80 @@
+// christmas hack, courtesy: https://pajasevi.github.io/CSSnowflakes/
+
+;(function(){
+    "use strict";
+
+    function makeItSnow() {
+        const styleSheet = document.createElement("style")
+        styleSheet.textContent = `
+        /* customizable snowflake styling */
+        .snowflake {
+            color: #fff;
+            font-size: 1em;
+            font-family: Arial, sans-serif;
+            text-shadow: 0 0 5px #000;
+        }
+
+        .snowflake,.snowflake .inner{animation-iteration-count:infinite;animation-play-state:running}@keyframes snowflakes-fall{0%{transform:translateY(0)}100%{transform:translateY(110vh)}}@keyframes snowflakes-shake{0%,100%{transform:translateX(0)}50%{transform:translateX(80px)}}.snowflake{position:fixed;top:-10%;z-index:9999;-webkit-user-select:none;user-select:none;cursor:default;animation-name:snowflakes-shake;animation-duration:3s;animation-timing-function:ease-in-out}.snowflake .inner{animation-duration:10s;animation-name:snowflakes-fall;animation-timing-function:linear}.snowflake:nth-of-type(0){left:1%;animation-delay:0s}.snowflake:nth-of-type(0) .inner{animation-delay:0s}.snowflake:first-of-type{left:10%;animation-delay:1s}.snowflake:first-of-type .inner,.snowflake:nth-of-type(8) .inner{animation-delay:1s}.snowflake:nth-of-type(2){left:20%;animation-delay:.5s}.snowflake:nth-of-type(2) .inner,.snowflake:nth-of-type(6) .inner{animation-delay:6s}.snowflake:nth-of-type(3){left:30%;animation-delay:2s}.snowflake:nth-of-type(11) .inner,.snowflake:nth-of-type(3) .inner{animation-delay:4s}.snowflake:nth-of-type(4){left:40%;animation-delay:2s}.snowflake:nth-of-type(10) .inner,.snowflake:nth-of-type(4) .inner{animation-delay:2s}.snowflake:nth-of-type(5){left:50%;animation-delay:3s}.snowflake:nth-of-type(5) .inner{animation-delay:8s}.snowflake:nth-of-type(6){left:60%;animation-delay:2s}.snowflake:nth-of-type(7){left:70%;animation-delay:1s}.snowflake:nth-of-type(7) .inner{animation-delay:2.5s}.snowflake:nth-of-type(8){left:80%;animation-delay:0s}.snowflake:nth-of-type(9){left:90%;animation-delay:1.5s}.snowflake:nth-of-type(9) .inner{animation-delay:3s}.snowflake:nth-of-type(10){left:25%;animation-delay:0s}.snowflake:nth-of-type(11){left:65%;animation-delay:2.5s}
+        `
+        document.head.appendChild(styleSheet)
+
+        const snowflakes = document.createElement("div")
+        snowflakes.id = "snowflakes-container"
+        snowflakes.innerHTML = `
+            <div class="snowflakes" aria-hidden="true">
+                <div class="snowflake">
+                <div class="inner">❅</div>
+                </div>
+                <div class="snowflake">
+                <div class="inner">❅</div>
+                </div>
+                <div class="snowflake">
+                <div class="inner">❅</div>
+                </div>
+                <div class="snowflake">
+                <div class="inner">❅</div>
+                </div>
+                <div class="snowflake">
+                <div class="inner">❅</div>
+                </div>
+                <div class="snowflake">
+                <div class="inner">❅</div>
+                </div>
+                <div class="snowflake">
+                <div class="inner">❅</div>
+                </div>
+                <div class="snowflake">
+                <div class="inner">❅</div>
+                </div>
+                <div class="snowflake">
+                <div class="inner">❅</div>
+                </div>
+                <div class="snowflake">
+                <div class="inner">❅</div>
+                </div>
+                <div class="snowflake">
+                <div class="inner">❅</div>
+                </div>
+                <div class="snowflake">
+                <div class="inner">❅</div>
+                </div>
+            </div>`
+
+        document.body.appendChild(snowflakes)
+
+        const script = document.createElement("script")
+        script.innerHTML = `
+            $(document).ready(function() {
+                setTimeout(function() {
+                    $("#snowflakes-container").fadeOut("slow", function() {$(this).remove()})
+                }, 10 * 1000)
+            })
+        `
+        document.body.appendChild(script)
+    }
+
+    let date = new Date()
+    if (date.getMonth() === 11 && date.getDate() >= 12) {
+        makeItSnow()
+    }
+})()