itch.io

I don't have exact numbers to give, but for me invokeai3 takes more than ten seconds before each invoke starts processing, even with just one LORA. In ComfyUI it seems to be instant with 10 LORAs. My CPU is weak, so that might be part of the reason, but there might be some optimisations to be done, because I don't have any problems with ComfyUI. Maybe the problem is that invokeai doesn't keep LORA's in memory and therefore loads them every time, whereas ComfyUI does. I have no idea and I'm not an expert, just a guess.

Yes, it would be great to have a build that has everything, that stands out from the alternatives. I'm looking forward to that.

Canny is the only controlnet model that sort of works (except with inpainting). The other controlnet models

give errors in both the UI and the console. Something about controlnet processors is missing. Here is an example of an attempt to use OpenPose:

[InvokeAI]::ERROR --> Traceback (most recent call last):

  File "C:\Users\user\invokeai3_standalone\env\lib\site-packages\invokeai\app\services\processor.py", line 70, in __process

    outputs = invocation.invoke(

  File "C:\Users\user\invokeai3_standalone\env\lib\site-packages\invokeai\app\invocations\controlnet_image_processors.py", line 221, in invoke

    processed_image = self.run_processor(raw_image)

  File "C:\Users\user\invokeai3_standalone\env\lib\site-packages\invokeai\app\invocations\controlnet_image_processors.py", line 385, in run_processor

    openpose_processor = OpenposeDetector.from_pretrained(

  File "C:\Users\user\invokeai3_standalone\env\lib\site-packages\controlnet_aux\open_pose\__init__.py", line 103, in from_pretrained

    body_model_path = hf_hub_download(pretrained_model_or_path, filename, cache_dir=cache_dir)

  File "C:\Users\user\invokeai3_standalone\env\lib\site-packages\huggingface_hub\utils\_validators.py", line 118, in _inner_fn

    return fn(*args, **kwargs)

  File "C:\Users\user\invokeai3_standalone\env\lib\site-packages\huggingface_hub\file_download.py", line 1291, in hf_hub_download

    raise LocalEntryNotFoundError(

huggingface_hub.utils._errors.LocalEntryNotFoundError: Connection error, and we cannot find the requested files in the disk cache. Please try again or make sure your Internet connection is on.

LORA models load very slowly before each invoke, the more you have the longer it takes, but that's probably not the fault of this standalone, just a warning for people using them. Maybe a bug, because A1111 and ComfyUI load them quickly.

Controlnet models are missing from the invokeai3 standalone. This is unfortunate as they are one of the main features. Portable standalone should not need internet for extra downloads. Please include them so that it can be used in an offline environment. Thanks for your work.