InstantID Stylize Portraits in 1 click ComfyUI 

Thank you

the version of the precompiled insightface wheels are in the video description those are the ones I used for this on windows portable

I recommend a used 3090 is about the price of the new 4070 having more vram is preferred even if 4070 super is faster for certain things vram is more important

500$ for used 3090 here in our place. It really is vram hungry for this to use efficiently

Did you installed the onnxrunyime-gpu? I think you are right it needs optimizations in my machine it uses about 16GB or more but I think it will work on the 4070 when it gets optimized.

Wow hope it works did you watch the last part were it still the nodes fail despite all being correctly installed Insightface = !Insightface message and it turns out that it is just the workflow that needs to be recreate it. Please watch that is about the end of the video and I am sure that issue is tripping people up. It actually happens similar to other custom nodes because the workflows can't update automatically when there is a change in the definitions inside the ui program of the nodes if you know what I mean.

We need to wait for a more optimized version of this At the end of the video I mention that it hogged the full of my 24GB VRAM but I am running many other apps including screen recording but I still think it went over 16GB. Photomaker can be run on 8 to12GB and is better in my opinion, this one is nice but if you can use more pics I like best Photomaker

@@impactframes Yeah. I think it needs about 13GB to complete. My workflow would get 100% through the next to last node, pause for a few seconds, and then throw a memory error before giving a final image. Frustrating

@@97BuckeyeGuy It won't be long to get a optimization there are many eyes on it as It's trending on github right now.

I saw a trick on another video. Try this: 1. Open Nvidia Control Panel 2. Manage 3D settings 3.CUDA - System Fallback Policy (select Prefer Sysmem Fallback option) This should prevent the error.

@@Thomas_Leo oh nice that could work mine those this automatically but I forgot it might be different depending on the GPU, thanks for sharing.

the version of the precompiled insightface wheels are in the video description those are the ones I used for this on windows portable

It looks like amodel mismatch error. I think instant I'd only works with sdxl models and not with 1.5 but is hard to say just from the message.

It prompts that the corresponding file cannot be found in the file directory: diffusion_pytorch_model.bin. The file directory file should be: diffusion_pytorch_mode.safetensors

Sorry I think this needs CUDA to work not totally sure because it runs on OSX servers but I think you can have Nvidia GPU on MAC servers so almost sure it is not working on M1 but I don't have mac sorry

damn, thanks for the feedback!@@impactframes

Reply not sure on the implementation from the video but it is possible on the ipadapter i'stant I'd I haven't check the latest updates on this one but at the time there was no masks. But I will check later

Yes hope someone figure out the optimizations needed for it. Thanks

Looks like the configs.json is not on the same folder look in the min 12:46 of this video I download it and place it on the checkpoints> controlnet> together with the actual model file so there are two files there configs.json and the model itself

Looks like the configs.json is not on the same folder look in the min 12:46 of this video I download it and place it on the checkpoints> controlnet> together with the actual model file so there are two files there configs.json and the model itself

That is silly the way the node works but... whatever I guess. @@impactframes

@@KINGLIFERISM yes is so complex

So you enter the environment install all the requirements.txt and now does not start? So you get to see the link to open 8188?

I like Photomaker better because it has more styles also looks more Realistic.

Looks like OOM I think this thing needs about 12GB to run apparently there is now a native version of this node that might help but sorry beyond the troubleshoot video I made I don't know how can you solve it.

Is this my workflows which I linked in the video? I Think you might have to recreate the workflow as I explain towards the end of this video mainly the insight face loader and the I'd gen get conflicted sometimes is around the 20min mark

@@impactframes No, with your workflow this error appears: !!! Exception during processing !!! Traceback (most recent call last): File "C:\Users\Luna\Documents\ComfyUI_windows_portable\python_embeded\lib\site-packages\diffusers\configuration_utils.py", line 428, in load_config config_dict = cls._dict_from_json_file(config_file) File "C:\Users\Luna\Documents\ComfyUI_windows_portable\python_embeded\lib\site-packages\diffusers\configuration_utils.py", line 551, in _dict_from_json_file return json.loads(text) File "json\__init__.py", line 346, in loads File "json\decoder.py", line 337, in decode File "json\decoder.py", line 355, in raw_decode json.decoder.JSONDecodeError: Expecting value: line 1 column 1 (char 0) During handling of the above exception, another exception occurred: Traceback (most recent call last): File "C:\Users\Luna\Documents\ComfyUI_windows_portable\ComfyUI\execution.py", line 155, in recursive_execute output_data, output_ui = get_output_data(obj, input_data_all) File "C:\Users\Luna\Documents\ComfyUI_windows_portable\ComfyUI\execution.py", line 85, in get_output_data return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True) File "C:\Users\Luna\Documents\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-0246\utils.py", line 373, in new_func res_value = old_func(*final_args, **kwargs) File "C:\Users\Luna\Documents\ComfyUI_windows_portable\ComfyUI\execution.py", line 78, in map_node_over_list results.append(getattr(obj, func)(**slice_dict(input_data_all, i))) File "C:\Users\Luna\Documents\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-InstantID\InstantIDNode.py", line 96, in load_idcontrolnet controlnet = ControlNetModel.from_pretrained(controlnet_path, torch_dtype=torch.float16) File "C:\Users\Luna\Documents\ComfyUI_windows_portable\python_embeded\lib\site-packages\huggingface_hub\utils\_validators.py", line 118, in _inner_fn return fn(*args, **kwargs) File "C:\Users\Luna\Documents\ComfyUI_windows_portable\python_embeded\lib\site-packages\diffusers\models\modeling_utils.py", line 712, in from_pretrained config, unused_kwargs, commit_hash = cls.load_config( File "C:\Users\Luna\Documents\ComfyUI_windows_portable\python_embeded\lib\site-packages\huggingface_hub\utils\_validators.py", line 118, in _inner_fn return fn(*args, **kwargs) File "C:\Users\Luna\Documents\ComfyUI_windows_portable\python_embeded\lib\site-packages\diffusers\configuration_utils.py", line 432, in load_config raise EnvironmentError(f"It looks like the config file at '{config_file}' is not a valid JSON file.") OSError: It looks like the config file at 'C:\Users\Luna\Documents\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-InstantID\checkpoints\controlnet\config.json' is not a valid JSON file.

Yes but it won't take long this repo is too popular I think someone will make the optimizations needed soon and release them for all.

@@impactframes I hope so 👍

Yes is very 😎 one click one image and it can do all of this. It's amazing.

make sure you are using onnxruntime-gpu like I do on min 10:42 if not it might be better to wait for optimiz\ations on this implementation because the memory management might be slowing things down

@@impactframes I'm using onnxruntime-gp.Yes, it might be better to wait for optimizations. Does PhotoMake have the same problem?

I just try them and they are working

Di you make sure to do all the steps with the installation make sure to watch the whole video for example did you installed Insightface and onnxrunyime-gpu?

@@impactframes I solved the 'dtype' problem according to your method, reinstalled 'insightface-0.7.3-cp310-cp310-win_amd64.whl' and then upgraded diffusers from 0.21 to 0.25 and it was solved but torch.cuda.OutOfMemoryError appeared. : CUDA out of memory. So frustrating🤪

@@blender6665 somebody else commented here the Nvidia panel has a feature to fallback on system memory >nvdia control panel>manage 3D settings>Cuda system fallback policy>Prefer system fallback. See if that works but I don't know cos this thing hogs so much vram but some people said it works for them

That's why I made a video about Enfugue this is UI is more for people to do experiments with SD it allows for a grain control but Enfugue is more for User experience and is very powerful too.

Maybe you need 3.11 are you on ComfyUI windows portable? Go into your python embedded folder type CMD at the top and do python.exe python --version If you qre on sepqrqte environment just activate your environment .\Scripts\activate.bat and pip install -U insightface

Sorry It must be the memory management beetr wait for the optimizations but also make sure you have onnxruntime-gpu and you did all the steps I showed. The video is almost realtime with leosamhelloworld turboLCM model you can see how it changes the generations while I talk specially at the beggining of the video it took no more than 40 seconds for an image many came under 20you can read the stats in the main menu. when I tested it the last one took a long time because it had to load all the models and I was doing other stuff on my computer.

make sure you using onnxruntime-gpu like on min 10:42