ComfyUI-Impact-Pack - Tutorial #2: FaceDetailer

This video demonstrates the most basic method of automating facial enhancement using FaceDetailer.

[DurzoBlunts]

Thank you for all your time/work in making this, great documentation and content covering the nodes with these videos.

[MichauxJHyatt]

There aren't many video tutorials how to set this up properly from the beginning, and the ones that do exist start off in the middle of something else. So I greatly appreciate your insight and contribution as well as being the shortest and most informational tutorial I've seen. Thank you!

[Gawdzend]

Incredibly helpful for getting things up and running from an Auto1111 mindset. Thank you!

[yossikoren]

Thank you very much ! very helpful and makes things understandable.
Where can I find the workflow json in order to have the prompts used in the video.?

[svenhinrichs]

How is ist possible to inpaint at the perfect resolution ? I d like to upscsale the bbox masked areas to the closest sdxl format and paste the scaled down inpainted to the original?

[TheQok]

Hey man, just tested the new node to convert image batch to list, works really well, keep up the great work dude!

[marcoannesebob]

great video! one question, do you think it is possible to also specify negative prompts on the face detailer? I just can't find the way

[dannywei]

Thank you for providing the usage information of FaceDetailer. I have a question, why is there no color difference at the seam after FaceDetailer is redrawn, but when using mask to redraw, the color difference at the seam is so obvious? Is it possible to solve the problem by just adjusting parameters like bbox_dilation, sam_dilation, sam_bbox_expansion, noise_mask_feather?

[АлександрКаренин-йн]

Kudos to the author! But what if I have several characters on the image (each is controlled with its own lora)?

[photo-noriter]

Thank you for your lecture. I took the course and made it with the nodes.. There's an error in the face detail part What could be the cause? For your information, comfyui has been updated to the latest version

Error occurred when executing FaceDetailerPipe:pic should be 2/3 dimensional. Got 4 dimensions.File "C:\ComfyUI_windows_portable_nvidia_cu118_or_cpu\ComfyUI_windows_portable\ComfyUI\execution.py", line 151, in recursive_executeoutput_data, output_ui = get_output_data(obj, input_data_all)File​ "C:\ComfyUI_windows_portable_nvidia_cu118_or_cpu\ComfyUI_windows_portable\ComfyUI\execution.py", line 81, in get_output_datareturn_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True)File "C:\ComfyUI_windows_portable_nvidia_cu118_or_cpu\ComfyUI_windows_portable\ComfyUI\execution.py", line 74, in map_node_over_listresults.append(getattr(obj, func)(**slice_dict(input_data_all, i)))File "C:\ComfyUI_windows_portable_nvidia_cu118_or_cpu\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-Impact-Pack\modules\impact\impact_pack.py", line 1485, in doitenhanced_img, cropped_enhanced, cropped_enhanced_alpha, mask = "C:\ComfyUI_windows_portable_nvidia_cu118_or_cpu\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-Impact-Pack\modules\impact\impact_pack.py", line 934, in enhance_facesam_mask = core.make_sam_mask(sam_model_opt, segs, image, sam_detection_hint, sam_dilation, File "C:\ComfyUI_windows_portable_nvidia_cu118_or_cpu\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-Impact-Pack\modules\impact\core.py", line 242, in make_sam_maskpredictor.set_image(image, "RGB")File "C:\ComfyUI_windows_portable_nvidia_cu118_or_cpu\ComfyUI_windows_portable\python_embeded\lib\site-packages\segment_anything\predictor.py", line 56, in set_imageinput_image = "C:\ComfyUI_windows_portable_nvidia_cu118_or_cpu\ComfyUI_windows_portable\python_embeded\lib\site-packages\segment_anything\utils\transforms.py", line 31, in apply_imagereturn np.array(resize(to_pil_image(image), target_size))File "C:\ComfyUI_windows_portable_nvidia_cu118_or_cpu\ComfyUI_windows_portable\python_embeded\lib\site-packages\torchvision\transforms\functional.py", line 278, in to_pil_imageraise ValueError(f"pic should be 2/3 dimensional. Got {pic.ndim} dimensions.")

Do you have a new FaceDetailer tutorial to match the last node ? Or do you have a Discord server ? I can made it working because problem with BBOX detector (model field is "undefined"). Thx and you do great work !!!

[BackyardTattoo]

Many times it skips the detailer and returns the same imagine, any solution, please? Thank you!

[LeKhang]

Thank you very much for sharing. In the [BASIC PIPE >> DETAILERPIPE] node what should I connect to refiner_basic_pipe please? I got stuck there. In the video that node has only 4 inputs but the lastest version has 6 inputs.

[JamesMcHie]

The face detailer added some texture, but I can't make the emotion LoRA work. Just to test it out I tried different facial expression prompts and nothing resulted in more than tiny changes from the generated image.

[kcc_netherland]

Hi, Thanks for nice video. When I load 'BasicPipetoDetailerPipe', the box appears different, 2 more bullets and 1 box to select Lora. Strangely, it show error "Required input missing - select to add Wildcard'. What it could be the cause of this problem? I stalled in here....

[AI.ImaGen]

Really good tuto, fast and clear! I was looking for that since ComfyUI install... and it's run fine with SDXL 1.0

[isomorphism__]

Thanks for the tutorial. I tried the provided workflow but it would fail with "RuntimeError: The expanded size of the tensor (63) must match the existing size (64) at non-singleton dimension 3. Target sizes: [1, 4, 79, 63]." Tensor sizes: [4, 80, 64]". I updated everything but still see the same issue. The other workflow using detailerdebug (segs) works fine for me. Have you seen this issue before?

EDIT: I tried deleting all facedetailer nodes and re-add them following your tutorial, and it works now

[billyleap]

Hey, mine load sithout the detector model in th ONNXDetectorProvider, it says undefined, i dont know what model or where to put it

[Najdmie]

Is there a way to detail the whole head including the hair? I know I can just upscale then inpaint manually, but detailer will be a time saver.

[lioncrud]

This used to work for me all the time, but recently the UltralyticsDetectorProvider has disappeared from node search. Even when I drag a noodle from the bbox_detector circle on the BasicPipe->DetailerPipe it doesn't appear. I tried uninstalling and re-installing Impact Pack both manually and then using ComfyUI Install Manager. No matter what i do, that UDP node is NOT showing up anywhere in ComfyUI