Good video for beginner. Here some tips: Hand detailing can be done without PS. Just after first inpaint result that you like switch to "inpaint only masked" with the same area selected to process that with higher resolution. Denoising can be lowered to 0.5. That way fingers and nails will be drawn almost perfectly. Also instead of control images editing, you can tweak pose bones with openpose editor extension right in A1111 window.
DWPose is a blessing indeed. I'm not a fan of manual painting stuff so in ComfyUI I use now two samplers linked via a OpenPose C-Net with DWPose most of the time as my starting stage. This doesn't only fix hands but also extra limbs. Not always of course and sometimes it makes it only worse but overall I think this approach works better than any LoRa and you don't have to mess up your prompt.
Appreciate the video - never had a good handle on this. Did want to mention you say to set the "CFG" level to .9 twice (ie: 01:48) - I'm sure it's obvious by the onscreen and basic knowledge, but you mean to say set Denoising strength. Cheers.
I was cringing so hard when you brought openpose into photoshop instead of using the built in open pose editor inside A111. Don't do tutorials if you don't know know what you're doing.
this is great, i just have this problem where when i follow your instructions and others... I get a blurry shape as my result. The hand will look like a ball of flesh or some type of shape.
same method that i did, but there is one more method as well that I did sometimes, with Lora hands + Neg hands embedding for inpainting. I use this when i really want good hands. Mu fast method was, Turn on the Adetailer (in this step from model 1-5) BG (using full body masking then invert it), Full body, Wardrobes (Using deepFashion2 model), and then Face. After this just upscale it if it's good enough. or just did the inpainting for hands for example, using the Lora + Neg Hands Embedding. For basic generation i rarely use the hands embedding and the hands lora.
I'm trying to understand your fast method. You use ADetailer for background, full body, wardrobe, and face? Like for BG, you use person_yolov8n-seg and select 'Merge and Invert' under Mark Preprocessing in ADetailer?
@@Darkwing8707 yes, set to higher resolution. and when it done it will be lowres in mycase 512x768 but with more clarity and details, and from there just do the upscale, instead of relying on hiresfix.
Followed this to the letter and now instead of weird hands I get smears in the masked area or pieces of my character's body missing where the mask is. I suspected this would be bad but it's worse than useless.
Good tutorials! Do you know of any AI that can edit the voice in a long video, running on my PC? Could you make a video on how to train artificial intelligence locally, providing information that we want? Thank you very much and keep it up!"
Great tutorial BUT it went a little fast and because the interface text was faint it was hard to keep track, not knowing what tab he was navigating on too, also an important step was missing and that is to install: control_v11p_sd15_inpaint
@@manhvu3673 They are two of many other than understanding how negative prompting works which helps without them. Also models like Flux do 99% better right out of the box.
📢 Last chance to get 40% OFF my AI Digital Model for Beginners COURSE: aiconomist.gumroad.com/l/ai-model-course
Good video for beginner. Here some tips:
Hand detailing can be done without PS. Just after first inpaint result that you like switch to "inpaint only masked" with the same area selected to process that with higher resolution. Denoising can be lowered to 0.5. That way fingers and nails will be drawn almost perfectly. Also instead of control images editing, you can tweak pose bones with openpose editor extension right in A1111 window.
nice tips thanks
DWPose is a blessing indeed. I'm not a fan of manual painting stuff so in ComfyUI I use now two samplers linked via a OpenPose C-Net with DWPose most of the time as my starting stage. This doesn't only fix hands but also extra limbs. Not always of course and sometimes it makes it only worse but overall I think this approach works better than any LoRa and you don't have to mess up your prompt.
im not not a fan, but i cannot draw!
I've tried this approach and used adetailer, and the results turned out pretty good. No need for additional editing in Photoshop. Thanks!
This was actually really good! For Comfy UI this concept works really well
Now we just need to fix the feet next
and teeth 😂
Appreciate the video - never had a good handle on this. Did want to mention you say to set the "CFG" level to .9 twice (ie: 01:48) - I'm sure it's obvious by the onscreen and basic knowledge, but you mean to say set Denoising strength. Cheers.
4:16 after generated a preview. you can click "edit" button in the "preprocessor preview" box to easily drag and move each position
"easy method" and "photoshop" in the same context :D
1:51 I did everything according to your video tutorial, but when I set this parameter to 0.9 like you did, I end up with porridge instead of hands!
Great. It really works. Thks.
Thank you.
Great vid! Thanks for a showcase.
Glad you liked it!
Is this DWPose extension doing anything? Even without the extention, controlnet has "dw_openpose_full" as selection. (Controlnet v1.1.427)
Excellent tutorial. This alone deserves a sub and did just that!!
Looking forward for more contents.
boy after 24h I finally got some results. Thank you so much^^
Well done!
I was cringing so hard when you brought openpose into photoshop instead of using the built in open pose editor inside A111. Don't do tutorials if you don't know know what you're doing.
probably he doesnt know its existence yet, he still went the right path though
I enjoyed the tutorial, at least it was short
This video is gold. thank you!
this is great, i just have this problem where when i follow your instructions and others... I get a blurry shape as my result. The hand will look like a ball of flesh or some type of shape.
same method that i did, but there is one more method as well that I did sometimes, with Lora hands + Neg hands embedding for inpainting. I use this when i really want good hands.
Mu fast method was, Turn on the Adetailer (in this step from model 1-5) BG (using full body masking then invert it), Full body, Wardrobes (Using deepFashion2 model), and then Face. After this just upscale it if it's good enough. or just did the inpainting for hands for example, using the Lora + Neg Hands Embedding. For basic generation i rarely use the hands embedding and the hands lora.
I'm trying to understand your fast method. You use ADetailer for background, full body, wardrobe, and face? Like for BG, you use person_yolov8n-seg and select 'Merge and Invert' under Mark Preprocessing in ADetailer?
@@Darkwing8707 yes, set to higher resolution. and when it done it will be lowres in mycase 512x768 but with more clarity and details, and from there just do the upscale, instead of relying on hiresfix.
Followed this to the letter and now instead of weird hands I get smears in the masked area or pieces of my character's body missing where the mask is. I suspected this would be bad but it's worse than useless.
Is this for 1.5 or also SDXL? I don't have much luck with open pose and SDXL... got any tips?
for pony 6xl ?
"Error RuntimeError: mat1 and mat2 shapes cannot be multiplied"
What do I do wrong?
How do I install DW Pose and control net in Github and huggingface?? Cause i am so confused
why is it not recognizing any models for me.. I was using controlnet before but now its none. where do I have to place the models for dwpose?
which checkpoint model you use in second image
An excellent technique, but for a video of more than 900 frames it might be a bit too tedious
at 1:50 you said to set CFG to 0.9, but them moved the denoising strenght slider instead
a high cfg setting or denoising strength?
Good tutorials! Do you know of any AI that can edit the voice in a long video, running on my PC? Could you make a video on how to train artificial intelligence locally, providing information that we want? Thank you very much and keep it up!"
There's no preview button?
sir do I have to install this to my pc?
bsolutely, stability is key, whether you're running Duffusion on your computer or in the cloud!
4:00 She got those Stanford Pines hands lmao (Gravity Falls for those of you who live under a rock lol)
it wont let me install dw openpose?
but it looks still a bit awkward... :/
Which AI voice used here?
Great tutorial BUT it went a little fast and because the interface text was faint it was hard to keep track, not knowing what tab he was navigating on too, also an important step was missing and that is to install: control_v11p_sd15_inpaint
Do you know how to install that?
Awesome! ❤ but the hand bones still doesn’t look quite natural. A little more fine tuning is required….
it may took hours of tries and probably 100 photos to get good result
Or the other option is too crop out the hands ;)
Could've sworn I heard MKBHD
it probably is, just AI voice
@@piracymoney Haha sick, do you think ElevenLabs can get this accurate with cloning?
Does it work with SDXL?
SD1.5 or SDXL works the same, but need download separated OpenPose model version (ends with _xl_)
aint no body got time for ps
ai voice with mkbhd voice loooool
no need for this anymore with all the hand loras around and better models.
which hand lora is good?
@@TheKillingPerfection Go with band hands v4 and v5 are good, also loras like unrealisticdream have hands included.
@@TPCDAZ only for SD1.5 right?
@@TPCDAZwhat you mean bad hands v4 v5? are they Lora model files or just text prompt?
@@manhvu3673 They are two of many other than understanding how negative prompting works which helps without them. Also models like Flux do 99% better right out of the box.