This is just unbelievable! I took an image of Nicolas Cage where you can barely recognize him, ran it 3 times on the results, and the final output is him in super high resolution. Crazy! I love it, amazing job brother! Thanks for sharing!
Thank you! I downloaded and tried many Supir workflows, and this is the only one that works great with amazing quality. Your workflow is very easy to understand and well-organized. I used the RealVisXL V50 SDXL model on my RTX 3060 with 12GB VRAM, and it took 4 minutes. The quality is definitely worth the wait.
I've tried numerous images, I keep getting error "stack expects a non-empty TensorList". I updated all, and using the same ForRealXL ver .3 and added VAE Any ideas on this error?
Great video's @Jockerai. The 3rd picture featured in the beginning of your video is an old B&W photo that turns to color with the slider. What setting or checkpoint is that found in? Did I miss something in the 2 video's? I tried to have ChatGPT make a description for full color but that didn't work. Keep up the great work sir! Love your video's!
@@Darkprayers you're welcome bro. The upscaling was with the spuir but colorizing not. For colorizing that image I used an ai website that is especially doing that. I forgot the name of that just searched in google
this could be either because there's not enough precision to represent the picture, or because your video card does not support half type. try setting the "upcast cross attention layer to float32" option in settings > stable diffusion or using the --no-half commandline argument to fix this. use --disable-nan-check commandline argument to disable this check.
@@Jockerai The prompt is: "A high-resolution photo of a beautiful girl with soft blonde wavy hair, wearing a blue dress. The girl has striking blue eyes, full natural lips, and a calm, confident expression." and when the upscale finish the final image has faces all over the picture, I dont know why
Thank you for videos. It seems that UltimateSDUpscale isn't working with Comfy anymore. It will no longer install. Do you know of this? Perhaps of another place to download a version of it that still works?
@@josephrguezglez8670 yup, change dtype to bf16 and fp16, anywhere that is on auto in the flow, still i hit another error after this, and after ksampler done, mismatch color error stuff
this could be either because there's not enough precision to represent the picture, or because your video card does not support half type. try setting the "upcast cross attention layer to float32" option in settings > stable diffusion or using the --no-half commandline argument to fix this. use --disable-nan-check commandline argument to disable this check.
updated all, installed missing but get this errors -- Prompt outputs failed validation VAELoader: - Value not in list: vae_name: 'ae.sft' not in ['ae.safetensors', 'taesd', 'taesdxl', 'taesd3', 'taef1'] DualCLIPLoader: - Value not in list: clip_name2: 't5xxl_fp16.safetensors' not in ['clip_l.safetensors'] UNETLoader: - Value not in list: unet_name: 'flux1-dev.safetensors' not in []
click the link below to stay update with latest tutorial about Ai 👇🏻 :
www.youtube.com/@Jockerai?sub_confirmation=1
This is just unbelievable! I took an image of Nicolas Cage where you can barely recognize him, ran it 3 times on the results, and the final output is him in super high resolution. Crazy! I love it, amazing job brother! Thanks for sharing!
Nice work buddy. Happy to hear you liked this method. It is the best for now
Excellent work and many thanks for sharing your workflows!!!
Thank you! I downloaded and tried many Supir workflows, and this is the only one that works great with amazing quality. Your workflow is very easy to understand and well-organized. I used the RealVisXL V50 SDXL model on my RTX 3060 with 12GB VRAM, and it took 4 minutes. The quality is definitely worth the wait.
How many steps, you tried, with pics where there are medium-small distance faces, i note it changes the original features faces a little.
Great to hear! 🤩
I've tried numerous images, I keep getting error "stack expects a non-empty TensorList". I updated all, and using the same ForRealXL ver .3 and added VAE
Any ideas on this error?
Thanks!!!
@@MJ12GRAVITON 🌱🙏🏻
Great video's @Jockerai. The 3rd picture featured in the beginning of your video is an old B&W photo that turns to color with the slider. What setting or checkpoint is that found in? Did I miss something in the 2 video's? I tried to have ChatGPT make a description for full color but that didn't work. Keep up the great work sir! Love your video's!
@@Darkprayers you're welcome bro. The upscaling was with the spuir but colorizing not. For colorizing that image I used an ai website that is especially doing that. I forgot the name of that just searched in google
@@Jockerai TY sir! Great work!
@@Darkprayers you're welcome buddy
SUPIR_conditioner
module 'tensorflow' has no attribute 'Tensor' error, I have the 4090 and I did follow all the steps
Спасибо, все получилось отлично.
English please
Any chance of creating a video on how to get Flux installed and working in Forge UI. Thanks.
You don't need any video really, FLUX works by default with the latest version of FORGE since some time now.
Yes there is I try to do that and make a video
hi, everything is fine but in the middle of the scaling it gives me this error: mixed dtype (CPU): expect parameter to have scalar type of Float
this could be either because there's not enough precision to represent the picture, or because your video card does not support half type. try setting the "upcast cross attention layer to float32" option in settings > stable diffusion or using the --no-half commandline argument to fix this. use --disable-nan-check commandline argument to disable this check.
Hey master, I have a problem, when upscaling, the final photo is enhanced but there are faces all over the photo, why does this happen?
@@davidmanas95 make sure to use very proper prompt. What is your prompt that makes such result?
@@Jockerai The prompt is: "A high-resolution photo of a beautiful girl with soft blonde wavy hair, wearing a blue dress. The girl has striking blue eyes, full natural lips, and a calm, confident expression." and when the upscale finish the final image has faces all over the picture, I dont know why
@@davidmanas95 I have to see your whole workflow can guess what is wrong
Thank you for videos. It seems that UltimateSDUpscale isn't working with Comfy anymore. It will no longer install.
Do you know of this? Perhaps of another place to download a version of it that still works?
you're welcome bro. No it is still working and last update is for 2024-07-17
hi, i get this error in ksmapler:
RuntimeError: mixed dtype (CPU): expect parameter to have scalar type of Float
I have the same error, could you fix it?
@@josephrguezglez8670 yup, change dtype to bf16 and fp16, anywhere that is on auto in the flow, still i hit another error after this, and after ksampler done, mismatch color error stuff
this could be either because there's not enough precision to represent the picture, or because your video card does not support half type. try setting the "upcast cross attention layer to float32" option in settings > stable diffusion or using the --no-half commandline argument to fix this. use --disable-nan-check commandline argument to disable this check.
@Jockeri getting this error. Any idea. SUPIR_sample
Trying to convert Float8_e4m3fn to the MPS backend but it does not have support for that dtype.
Pls send a screenshot of your wf in telegram
❤❤❤
❤️❤️🤞🏻
👏😎
😎🤞🏻
updated all, installed missing but get this errors
--
Prompt outputs failed validation
VAELoader:
- Value not in list: vae_name: 'ae.sft' not in ['ae.safetensors', 'taesd', 'taesdxl', 'taesd3', 'taef1']
DualCLIPLoader:
- Value not in list: clip_name2: 't5xxl_fp16.safetensors' not in ['clip_l.safetensors']
UNETLoader:
- Value not in list: unet_name: 'flux1-dev.safetensors' not in []
Make sure you select every model (vae, text encoders...) one by one and have them in proper directory folders
@@Jockerai Thanks man, that was my problem, fixed now!