Great comparisons, thanks! I see you are upscaling a 512x512 image, while SDXL works best for 1024x1024 images, is that important? Also, for the A1111, did you use the same model to upscale (Juggernaut in this case)? That's important because AI upscaling is regenerating the image and the model has a lot to say about it. If you didn't use the same model, an interesting test would be to use one model in both programs so that the model doesn't make the difference, only the program. Anyway, thanks for the comparison, I'm enjoying your videos! P.D. I don't have A1111, so if you test the suggestion of same model, you don't need to make a new video but I would like to read your conclusions, thanks!
To be honest...I had just done an A1111 upscale video using that image, and then decided to do an upscale comparison video for Fooocus. So I decided to reuse the previous A1111 result, that used a different SD 1.5 model, so I could show the comparison at the end. In fact, if I had used an SDXL model in A1111, the A1111 upscaled result would have been even better. I was happy to do this, as previous tests using same SDXL models and 1024x1024 image sizes in both Fooocus and A1111, pretty much gave the same comparable results, between Fooocus Fast x2, Fooocus Standard x2 and A1111. You can actually see the generally lower upscale results in the video intro. These were all 1024x1024 images generated in Fooocus then upscaled with same model using Fooocus Standard x2. Even on screen you can see the less than perfect results. Good thoughtful questions though, and glad you liked the video!
@@FoxtonAI Thank you. Well, I said that not only because of the difference SD1 and SDXL could have, but also because models are trained differently, if you use one with lots of good images of animals it will be better generating (and upscaling) fur for animals, for instance. But yes, the point of Fooocus is to be easy to use, you can generate great images without complicated setups. For finetuning and greater control, programs like Automatic1111 or ComfyUI are king.
This is a LIvePortrait question. You should have CUDA installed already on your system. Launch the command prompt, and type in "nvidia-smi", and in the top right, you see your CUDA version. If you need to install CUDA, then just do a Google search for "nvidia cuda", and select the NVIDIA CUDA download link (currently v12.6), and then download and install from there. Then it's a straight install for LivePortrait.
Great comparisons, thanks!
I see you are upscaling a 512x512 image, while SDXL works best for 1024x1024 images, is that important?
Also, for the A1111, did you use the same model to upscale (Juggernaut in this case)? That's important because AI upscaling is regenerating the image and the model has a lot to say about it.
If you didn't use the same model, an interesting test would be to use one model in both programs so that the model doesn't make the difference, only the program.
Anyway, thanks for the comparison, I'm enjoying your videos!
P.D. I don't have A1111, so if you test the suggestion of same model, you don't need to make a new video but I would like to read your conclusions, thanks!
To be honest...I had just done an A1111 upscale video using that image, and then decided to do an upscale comparison video for Fooocus. So I decided to reuse the previous A1111 result, that used a different SD 1.5 model, so I could show the comparison at the end. In fact, if I had used an SDXL model in A1111, the A1111 upscaled result would have been even better.
I was happy to do this, as previous tests using same SDXL models and 1024x1024 image sizes in both Fooocus and A1111, pretty much gave the same comparable results, between Fooocus Fast x2, Fooocus Standard x2 and A1111.
You can actually see the generally lower upscale results in the video intro. These were all 1024x1024 images generated in Fooocus then upscaled with same model using Fooocus Standard x2. Even on screen you can see the less than perfect results.
Good thoughtful questions though, and glad you liked the video!
@@FoxtonAI Thank you. Well, I said that not only because of the difference SD1 and SDXL could have, but also because models are trained differently, if you use one with lots of good images of animals it will be better generating (and upscaling) fur for animals, for instance.
But yes, the point of Fooocus is to be easy to use, you can generate great images without complicated setups. For finetuning and greater control, programs like Automatic1111 or ComfyUI are king.
Make another video of liveportrait, from scratch, I had a lot of problems installing and configuring cuda, so the tutorial didn't work for me
This is a LIvePortrait question. You should have CUDA installed already on your system. Launch the command prompt, and type in "nvidia-smi", and in the top right, you see your CUDA version. If you need to install CUDA, then just do a Google search for "nvidia cuda", and select the NVIDIA CUDA download link (currently v12.6), and then download and install from there. Then it's a straight install for LivePortrait.