Thanks for sharing this. Good to see a simple clean SUPIR workflow out there. The layouts I've found elsewhere are terribly laid out. I'll be using this now. So glad I have 24gb VRAM

I am looking forward to trying this, but I'm getting the following error:

When loading the graph, the following node types were not found: mxreroute

That's after a complete update of Comfy and the missing nodes. An idea what's wrong?

This is absolutely amazing and really showcases flux and it's quality. Do you have a workflow that would be the same but just text to image without needing to load an image?

Thank you for releasing this at any rate, it's a great workflow.

Any idea what might cause the following error.

Error occurred when executing ReActorRestoreFace:

'NoneType' object has no attribute 'seek'. You can only torch.load from a file that is seekable. Please pre-load the data into a buffer like io.BytesIO and try to load from it instead.

I have this issue:When loading the graph, the following node types were not found:

ReActorRestoreFace

Nodes that have failed to load will show as red on the graph.

Can't fix it with manager. What can I do?

Here's the full traceback:

## Execute install/(de)activation script for 'D:\AI\PROGRAMS\TCHT\comfyui\ComfyU - Pastebin.com

I receive an error in both workflows, each having to do with ColorMatch. Any guidance is appreciated:

Error occurred when executing ColorMatch:

stack expects a non-empty TensorList

File "C:\Users\me\ComfyUI_windows_portable\ComfyUI\execution.py", line 152, in recursive_execute

output_data, output_ui = get_output_data(obj, input_data_all)

^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

File "C:\Users\me\ComfyUI_windows_portable\ComfyUI\execution.py", line 82, in get_output_data

return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True)

^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

File "C:\Users\me\ComfyUI_windows_portable\ComfyUI\execution.py", line 75, in map_node_over_list

results.append(getattr(obj, func)(**slice_dict(input_data_all, i)))

^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

File "C:\Users\me\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-KJNodes\nodes\image_nodes.py", line 118, in colormatch

out = torch.stack(out, dim=0).to(torch.float32)

^^^^^^^^^^^^^^^^^^^^^^^

Nice workflow and thank you :) I do have one recommendation though, too many required settings are hidden away. With the Florence one the WD14 tagger and then the Reactor node both gave me errors. I was able to tack them down and fix both, but with the WD14 node behind one of the comparer nodes and both minimised it took some work to identify them.

Same for the SUPIR nodes, anything that has drop downs for models or samplers should be visible. I get this was probably optimised for your system and it's all neat and tidy, but when released for general public use there are going to be a lot of different configs where it won't work without some tweaking.

Hope this doesn't come across as too harsh a criticism as it is a great workflow and I do like it, just trying to be constructive.

I am facing a CUDA related error and a PATH but that PATH doesn't exist on my Windows PC. How do I change that path? Here is the error:

=====

Error occurred when executing WD14Tagger|pysssss: D:\a\_work\1\s\onnxruntime\python\onnxruntime_pybind_state.cc:857 onnxruntime::python::CreateExecutionProviderInstance CUDA_PATH is set but CUDA wasnt able to be loaded. Please install the correct version of CUDA andcuDNN as mentioned in the GPU requirements page (https://onnxruntime.ai/docs/execution-providers/CUDA-ExecutionProvider.html#requirements), make sure they're in the PATH, and that your GPU is supported.

=====

I'm assuming this is using Custom Nodes? Where could I snap them?

Any Mac users in the house having issues running this? Help

Somehow I can't properly install the ReActer Node for ComfyUI, I always get "Import failed" :/

Hello! Thank you for your wonderful workflow. I've been having a great time with it this week. Could you please help me set up the best settings for restoring old pictures using img2img with the SUPIR technique, but without using Detailer?

Error occurred when executing VAELoader: 'NoneType' object has no attribute 'lower' File "/opt/rd/apps/ComfyUI/execution.py", line 152, in recursive_execute output_data, output_ui = get_output_data(obj, input_data_all) File "/opt/rd/apps/ComfyUI/execution.py", line 82, in get_output_data return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True) File "/opt/rd/apps/ComfyUI/execution.py", line 75, in map_node_over_list results.append(getattr(obj, func)(**slice_dict(input_data_all, i))) File "/opt/rd/apps/ComfyUI/nodes.py", line 704, in load_vae sd = comfy.utils.load_torch_file(vae_path) File "/opt/rd/apps/ComfyUI/comfy/utils.py", line 33, in load_torch_file if ckpt.lower().endswith(".safetensors") or ckpt.lower().endswith(".sft"):

(Rundiffusion.com)

If you get a WD14 tagger error. Try the fix found here: https://github.com/pythongosssss/ComfyUI-WD14-Tagger/issues/60#issuecomment-2267108746. It worked for me.

I really nice the workflow. Very clean and easy to follow. The only problem I'm seeing is that I'm getting 1.27s/it the first time I use the workflow and if I run it again it jumps up to 7s/it. I've tried enabling the VRAM and cache purge as well and it doesn't help. If I kill it and start again it will be fast again until I try to run it a second time. This happens both with img2img enabled and disabled. I'm sure it isn't a workflow issue, I just wanted to share my experience. I will need to keep looking for something that works for me with my 16GB card,

Very good workflow! It takes a while to figure out all the file paths that need to be adjusted to get it to work, but when it runs it manages to achieve amazing results with little additional effort.

There are only two problems I have not yet found a way to solve with this workflow. (Edit: found a solution to one of them, link below)

Problem 1: after the first successful run it gets extremely slow on the next runs - at least with my 16GB VRAM limit. It is actually faster to just restart Comfy and go for a next run from scratch. I've tried duplicating the "Purge VRAM" node, adding another one to the very end of the workflow, however it apparently does nothing at all. I still need to restart.

Problem 2: There is a loader for the NF4 version of FLUX, however it does not appear in the custom nodes as something that needs to be installed. (Edit: I managed to get it to work, here is the link with instructions https://www.reddit.com/r/comfyui/comments/1eq8puq/error_when_trying_to_use_flux1devbnbnf4safetensors/)

Despite those small inconveniences, this is a workflow clearly made by a professional and I intend to study it. There is much to learn from this workflow. Thank you!

I encountered this error, please guide me. Thank you:

Error occurred when executing WD14Tagger|pysssss: [ONNXRuntimeError] : 1 : FAIL : Load model from D:\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-WD14-Tagger\models\wd-vit-tagger-v3.onnx failed :D:\a\_work\1\s\onnxruntime\core/graph/model_load_utils.h:56 onnxruntime::model_load_utils::ValidateOpsetForDomain ONNX Runtime only guarantees support for models stamped with officially released onnx opset versions. Opset 4 is under development and support for this is limited. The operator schemas and or other functionality may change before next ONNX release and in this case ONNX Runtime will not guarantee backward compatibility. Current official support for domain ai.onnx.ml is till opset 3.

PLEASE HELP ME!!!
I have been tyring to get this to work for hours! I get this error. What should I do to fix this? Error occurred when executing SegmDetectorCombined_v2: 'NO_SEGM_DETECTOR' object has no attribute 'detect_combined' File "D:\ComfyUI\execution.py", line 313, in execute output_data, output_ui, has_subgraph = get_output_data(obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "D:\ComfyUI\execution.py", line 188, in get_output_data return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "D:\ComfyUI\execution.py", line 165, in map_node_over_list process_inputs(input_dict, i) File "D:\ComfyUI\execution.py", line 154, in process_inputs results.append(getattr(obj, func)(**inputs)) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "D:\ComfyUI\custom_nodes\ComfyUI-Impact-Pack\modules\impact\detectors.py", line 153, in doit mask = segm_detector.detect_combined(image, threshold, dilation) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

I'm missing the CheckpointLoaderNF4 node. I'm new to this genAI stuff so I don't know how to sort this error out.

I've tried downloading the missing nodes but this one doesn't show up. I can't find it when searching for nodes manually.

Anyone have any suggestions?

Is this only usable if i want to do everything and with a complete different image?
Img2Img and Inpaint is at "false"
But it takes the image and generates a new one and stops after it because of a missing mask.
XL, Detailer, USDU are at "no" but it won't let the image pass to Supir.
And even if it would reach supir, why does it use SDXL? I thought it is a Flux Supir Workflow?

Even when I try to fix all the hidden notes and relink them only with supir, it just does random weird stuff.

Love this workflow; i was able to figure out any issues I had. One question about your UI tho...what sidebar and top menu are using?

Do you think SUPIR will be updated to use Flux directly instead of having to use an SDXL model for it?

Hello and thanks, I don't know wht but the result is always cartoon, manga...

U think you could create a video of you working on a picture with this workflow? A tutorial of some sorts?

UPDATED KJNODES -problem solved.

I made a mistake by updating comfy UI while working on your workflow. stupid me now get this:

Loading aborted due to error reloading workflow data

TypeError: can't redefine non-configurable property 0lockArray@http://127.0.0.1:8188/extensions/ComfyUI-Custom-Scripts/js/locking.js:13:10 lockNode@http://127.0.0.1:8188/extensions/ComfyUI-Custom-Scripts/js/locking.js:117:14 beforeRegisterNodeDef/nodeType.prototype.onConfigure@http://127.0.0.1:8188/extensions/ComfyUI-Custom-Scripts/js/locking.js:181:13 beforeRegisterNodeDef/nodeType.prototype.onConfigure@http://127.0.0.1:8188/extensions/ComfyUI-Easy-Use/js/easy/easyWidgets.js:377:57 configure@http://127.0.0.1:8188/assets/index-CI3N807S.js:66723:14 configure@http://127.0.0.1:8188/assets/index-CI3N807S.js:66509:20 #addConfigureHandler/LGraph.prototype.configure@http://127.0.0.1:8188/assets/index-CI3N807S.js:85882:26 init/LGraph.prototype.configure@http://127.0.0.1:8188/extensions/ComfyUI-Custom-Scripts/js/snapToGrid.js:160:21 init/LGraph.prototype.configure@http://127.0.0.1:8188/extensions/ComfyUI-Custom-Scripts/js/reroutePrimitive.js:14:29 loadGraphData@http://127.0.0.1:8188/assets/index-CI3N807S.js:86334:18

This may be due to the following script:
/extensions/ComfyUI-Custom-Scripts/js/locking.js

I've gotta say this is a killer workflow. I'm getting pretty consistently awesome results right out of the box and even with my limited experience it's been easy enough to play around with.

which SUPIR model did you use (there are two models and no info on the differences)? thank you for this amazing workflow! im learning so much here

Sorry im new to this stuff, what comfyui model folders do each of the models you listed need to be placed into? I cant get this workflow to work properly. Trying to use everything in the workflow with default settings its not finding the models even though I downloaded them

I think with the latest ComfyUI update the florence is broken.

Failed to validate prompt for output 72:
* Miaoshouai_Tagger 300:

Absolutely stellar work again, best workflow out there. Interestingly I'm getting way better results with your older workflow using the SDXL refiner. I use juggernaut XL.

I've mirrored settings and not sure if it's my lora though seems unlikely. Regardless it's a fantastic workflow, will need to play around a bit to see if it is the absence of the SDXL model or not causing me the issue.

face_yolov8m-seg_60 in what folder should be place?

This version is a Masterpiece! Thank you for sharing!

Thanks for sharing this workflow. It's been giving me great results.

Theres one thing I can't wrap my head around, how is the img2img denoise and Max Shift, Base Shift interacting with each other?

Any tips regarding these settings would be much appreciated :)

Great workflow.. I didnt 2ndpass Noise injection , Lucy Sharpen and Shadow Highlight v2 in the updated workflow.
Particularly i found 2ndpass noise injection a great feature.

I like the look of this workflow, but I'm having trouble getting it to run. Even with everything turned off, it's still giving me errors about UltimateSDUpscale.

The error says "Input type (float) and bias type (struct c10::Half) should be the same"

Any idea how to fix that?

with the latest comfyUI update, the workflow broke. Something that has to do with the "SAMPLER IMAGE" node. At least for me, i don't know if anybody else has any problems. Would you mind taking a look? Or could you please upload a screenshot of your settings in that node? Because when i load it, the settings are a bit messed up after the update

Great update like always, I noticed and this may be user error that the workflow won't run unless you have an image loaded in the load image node. This is irrespective whether you have img2img set true or false. Any idea what is going on there?

I think the USDU nodes appear to be broken. (#3840 UltimateSDUpscale) is all red and (#3749 UltimateSDUpscale) is outlined red. Sorry if I'm wrong but I'm still learning and so low on energy right now I'll ask for some babysitting lol. The initial gen is awesome though, the workflow is sick. Though it's taking me 60s/it I assume because of my subpar rig(16GB RAM, 1660s(6GB VRAM)). With another SDXL workflow I tried a fully upscaled(4k) image takes me ~10 minutes lol. I've struggled with getting flux workflows to work.

I think this is broken at the moment

If you've installed SAM2 and LayerStyle, they will be in conflict with each other. I confirmed this by running the following command find ~/ComfyUI/custom_nodes -type f -iname "*sam2*" go to your custom_nodes folder and delete ComfyUI-SAM2, and make sure LayerStyle is installed. Then you should be good.

i`m having a issue with Ultimate SD Upscale, it`s installed, but when i open the workflow it`s missing a node, in the Custom Node Misssing tab show the thing, there is a button fix, but nothing happens, anyone can help to fix it?

how to turn off florence??

I'm getting:
SegmDetectorCombined_v2 'NO_SEGM_DETECTOR' object has no attribute 'detect_combined'
but am not sure where I can troubleshoot the issue. The box is outlined in purple

I've installed jiter,still shown '''No module named 'jiter.jiter' ''' .Does anyone know why?

Bro, thanks for the best workflow! Please tell me, some nodes are black, and it is not possible to configure the parameters. I would really like the network control to follow the example more closely when the photo turns black and white and the pose of the sample photo is extracted. Now the pose is not repeated, but a similar one is generated. How to do it?

How am I supposed to use it? is it just I2I? I just loaded a bunch of big black empty nodes.

after a comfyui update i cant run anymore this workflow?? someone can update??

so month ago the florence, stops working, now i none is working

Sorry but i'm confused, the download is just 4 images, 4 pngs