Much appreciated! While I'm working in SwarmUI after a brief stint in A1111, Comfy and Inference, this has helped me to understand the IP-Adapter mechanism. The face detection mechanism in Swarm is basically just a frontend for Comfys implementation. One thing I do note is how this function tends to struggle with people who aren't in the checkpoint vs celebrities who are. While the face swapping of someone like Brad Pitt works well, this is because Monsieur Pitt is already in most SD checkpoints. Without a properly trained LoRA, I find that SD struggles to replicate reliably the appearance of most unknown people
Hey i was wondering what might have gone wrong, because i did everything you did aswell in the img2img thing but the face was just covered in black? any ways to fix this?
Thanks for your video. I've experimented with various settings, including those you recommended, for the Brad Pitt face swap. Unfortunately, regardless of the input photos used, the generated results consistently output generic checkpoint model photos. Edit. It appears there might be a bug. I had to alternate between using single and multiple inputs to make it work. The behavior is inconsistent-sometimes it works, and other times it doesn't.
Watching and installed Automatic11 in April 2024, my local install of Automatic1111 does not show Sampling Method DPM++ 2M SDE Karras as it does in the video, can someone tell me why that is?
Hi man. I’m sorry for asking. I currently using A1111 v1.8 and I have issue with ERSGEN when I tried upscale with 4x-UltraSharpe or any upscale models it show error message. Do you know anyway to fix that?
@@BitesizedGenius It said: PytorchStreamReader failed reading zip archive: failed finding central directory. This message only shown up when I used Upscaler Models in ERSGAN folder.
Let me point out a typical problem of AI images, that I’m surprised that until today nobody fixed. The frowning, or the wrinkles between the brows. This is what create inconsistency when you run a batch sequence. 18 months and still nobody fixed it…..why? Because AI is made by programmers and not by CGI artists….that’s why the transition to usenAI professionally is looming in darkness, at least for the time being…..I gave up explaining this to the makers. They just don’t get it…..
Much appreciated! While I'm working in SwarmUI after a brief stint in A1111, Comfy and Inference, this has helped me to understand the IP-Adapter mechanism. The face detection mechanism in Swarm is basically just a frontend for Comfys implementation.
One thing I do note is how this function tends to struggle with people who aren't in the checkpoint vs celebrities who are. While the face swapping of someone like Brad Pitt works well, this is because Monsieur Pitt is already in most SD checkpoints. Without a properly trained LoRA, I find that SD struggles to replicate reliably the appearance of most unknown people
Does this work with SD-Forge all I can find in the controlnet pre-pocessor dropdown is Insightface+Clip-H (Ipadaptor) is in brackets
its named differently, the one youve named is the correct one.
In a1111 for sdxl, can you face id 2 faces in 1 generation?
Hey i was wondering what might have gone wrong, because i did everything you did aswell in the img2img thing but the face was just covered in black? any ways to fix this?
Thanks for your video. I've experimented with various settings, including those you recommended, for the Brad Pitt face swap. Unfortunately, regardless of the input photos used, the generated results consistently output generic checkpoint model photos. Edit. It appears there might be a bug. I had to alternate between using single and multiple inputs to make it work. The behavior is inconsistent-sometimes it works, and other times it doesn't.
Watching and installed Automatic11 in April 2024, my local install of Automatic1111 does not show Sampling Method DPM++ 2M SDE Karras as it does in the video, can someone tell me why that is?
mine also not showing.
mine too
how do you use stable diffusion ? im a total noob and cant seem to get it
ua-cam.com/play/PL27Xzrmm9vtLEmTFcy0KqwRjSvQ_nnYrI.html These might help, i've also got an installation and prompting technique video on my channel
Hi man. I’m sorry for asking. I currently using A1111 v1.8 and I have issue with ERSGEN when I tried upscale with 4x-UltraSharpe or any upscale models it show error message. Do you know anyway to fix that?
Depends, whats the error message?
@@BitesizedGenius It said: PytorchStreamReader failed reading zip archive: failed finding central directory. This message only shown up when I used Upscaler Models in ERSGAN folder.
@@GriezoPheaHmm not sure, probably best to paste that into google and see if its on reddit
@@BitesizedGenius nevermind, thanks man. I already research but still can't fix. Maybe because of V1.8.
It happens when the file you are referencing is corrupted. Download it again if you can.
Bro, 00:57 you didnt provide the link for that
nor do you say which of those files
Let me point out a typical problem of AI images, that I’m surprised that until today nobody fixed. The frowning, or the wrinkles between the brows. This is what create inconsistency when you run a batch sequence. 18 months and still nobody fixed it…..why? Because AI is made by programmers and not by CGI artists….that’s why the transition to usenAI professionally is looming in darkness, at least for the time being…..I gave up explaining this to the makers. They just don’t get it…..
With time they will fix everything, AI is expanding very fast in everything.