If you find the video useful and would like to tip you can buy me a cup of electricity for all these image generations. It is greatly appreciated! ko-fi.com/kleebztech
Negative Prompt : (Hands:1.5), (hands on hips:1.4), hand on hip, (holding:1.4), hands on chest, arms out, holding hair, hand on hip, hands on waist, thumb, (wrist:1.5), fingers
hand negatives were pretty good advice. I put stuff like grabbing, holding, grasping in it at 1.4 weight and they came out much better over all. I was already messing around with negative "hands" with different weights to try to find a sweet spot and it did improve it more or less, but this so far by a few testing looks very promising. maybe i try to do both. Thanks!
I've been going into photoshop and doing some painting of the hands to get the structure the way I want and then back in to use the detailer, your method looks a lot less intensive and I'll be giving it a try :)
I almost included that into the video as an option but was trying to keep it mostly using inpainting and Fooocus. Sometimes that method works well. But if its pretty easy to just generate a bunch then it really doesn't usually take much.
You can do a tutorial to generate images with custom sizes and not just those that appear on the list. By being able to generate larger images, for example 1216x1216, it is possible for the hands to look better structured as well as the details of the faces.
Improve details works better with using "woman hand" or "man hand" - it does make sense. Plus it is easier to sketch basic hand in photoshop/gimp and "improve details" on it.
Did you watch my video on the basics of that? I may do one that goes a little further into it now that I know how to get my computer to upscale and record at the same time. ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-9FZe7n9VO-Q.html
@@KLEEBZTECH I’ve noticed that the face changes completely when I try the upscale options (low, strong). I would just like to make the image sharper and the skin to match the faceswap when I upscale
You can use the FaceSwap when doing upscale and variation. Just need to go into debug options and check of the option to use FaceSwap and Upscale variation together. That should help.
This is the worst part for me. Not even so having to go through the process, but I'm on a low-end GPU with only 4G of RAM. I have to run multiple instances to fix things and it takes several minutes each. I'll spend 30 minutes on one image just trying to fix the hands. Doesn't leave much time to actually create.
Another option is to use other tools that have Openpose support like A1111 and such but those tools are much more complex. They can be a little better with getting the hands correct but you spend more time on that part instead of generating as much.
Could you maybe mix the inpainting with a photo of a real hand in ControlNet (PyraCanny)? Then the point is having a database of real hand photos in different poses...
Yes you could do something like that. I do have some videos that cover creating text and poses that use some inpainting tricks that could be used for that as well.
If I want to improve details of background but when I upscale it, it becomes more likely unrealistic or is it work for improving bg quality like if I select the some part of the bg and then put improve details??
Hard for me to answer that on here since I would need to understand more what you are getting. I am a regular visitor of the Fooocus Subreddit and answer on there often. Easier to help with that stuff if I can see what you are currently getting. Or if you want to email me I do have my contact info on the channel page.
I cannot see what you are doing through most of this video. You are moving stuff and switching views too fast and often. I had no idea what settings you were using or inpaint/outpainting or improving detail or what.
Your neg prompts about hands are useless and hiding hands isn't the answer because we don't want to generate people with hands in the pockets. First, use a better model for hands (Proteus 0.2 by example, most of the time you've got five fingers, even with anthropomorphic animals). But If you are stuck with Fooocus, the only way to do better hands is to use another software (Comfy or WebUI) with controlnet. Using Fooocus for this case is a waste of time.
I wouldn't say useless. And not everyone has the ability to use all the different tools. But if that works for you then I would suggest going that route.
It's quite easy to use ControlNet pose control (available with the Fooocus CN branch, so there's no steep learning curve). Alternatively, give Forge WebUI a try (same developer as Fooocus): hands can be corrected with almost a single click (AfterDetailer, hands). It's not very difficult to use if you already know Fooocus. If you're teaching people how to take good photos, don't let them go in the wrong direction and waste time. Stick to what we can truly achieve with this tool and don't push the limits too far. :) If Illyasviel invented Controlnet, it wasn't to waste hours on composition issues. However, he also doesn't want to overload the engine with bells and whistles for beginners, keeping it accessible to all GPU computers. Fooocus is a really good tool, but it has some limitations. Maybe Illyasviel would add other CN controls than canny, depth, and IPadapter. But not now, except in Fenneishi's branch (only a general pose controller: face, body, hands). At the very least, try other checkpoints with better hand control if you don't want to learn new tools. @@KLEEBZTECH
I use many of these tools and originally used ComfyUI for a while as my primary. I do the videos targeted towards people who do not. As for "better" checkpoints for hand control... Well I tried the one you mentioned but did not notice any difference.
