Man you are the BEST!!!!!! I have looking for simple solution to just upscale image without promt (because i have bunch of them and i dont remeber what promt have i used) and your first method worked like a charm!!!! Thank you so much!!
Thanks for this video, mate. I am trying to create a "celebrity LoRA" which needs high-res photos. The ones I found, a lot of them were pretty ordinary. I tried your setup and setting but whilst the photo did get larger, the quality didn't improve, noticably, at least. For my purposes, this seemed to work: * RealisticStockPhoto_v1.safetensors * sd_xl_vae (as a separate vae) * RealESRGAN_x4.pth (Upscale model This gave me less detail in the skin, bust as I'm after face shape primarily, those details weren't all to significant. The Add_DetailXL LoRA didn't add anything. So, I get tack-sharp hair, eyes and edges and soft-focus skin. Happy with that.
Also a json Flow for a batch SVD would be handy, imagine having 5 images, I need a sequence SVD from 1 to 2, 2 to 3, 3 to 4 and 4 to 5, would this be possible to make a setup in a flow?
It would be helpful is you explained the purpose of adding the KSampler node and why you chose the 0.15 values instead of 0. So what is the purpose of the ksampler node and why did you choose 0.15?
There is something called Pixel Space and Latent Space. K-Samplers work in latent space. When doing image to image its pixel space going directly to pixel space for upscaling. Text to Image upscaling requires denoising which is done in latent space. Since the image has to be generated first by the K-Sampler from the text prompt. The K-Samplers (laten space) will Decode and generate the image (pixel space) where Ultimate Upscale SD can be used to just upscale or add details while upscaling. Denoise strength of 0.15 means whatever is the image output by the text prompt, will not change much. The Ultimate SD upscale will have only 15% of pixels to add details to. So this will just refine some details. Upto 0.3 value you will just notice refinements. The image will start changing at higher values. A Value of 1 would means a completely different image. This value applies only when the mode type is linear or chess and not "none". None makes it ignore the denoise value and just upscales it.
Well this is an old tutorial and looks at techniques at that period, although still valid we now have multi options. First let me know if you are upscaling image to image or text to image generations. The final way to approach both of them would be different without adding much details.
@@controlaltai well both, I started with a 1000 frames Animdiff, lowres, mp4, when upscaling I got memory issues, tried a lot but no solution. The way that it works now, I think, is creating a batch folder with my 1000 frames in it, upscaling frame by frame and saving new batch of frames.
Re confirm by going in comfy manager on the to left from all select installed, see if it shows up. If it does and the node is not available, uninstall and re install, restart, open comfy manager again, select update comfy, check if any update is there or not. Update, restart. Open comfy manager again, click on update all, then go in custom nodes, select update from top left and update custom nodes.
Are you asking that after upscaling, you are getting this effect? If you check the denoise level for upscaling, if in linear. If you only want upscale then in ultimate upscale turn mode type to "none".
Hi, I found something regarding batch upscaling, however I am in the middle of a video for this week. I have to try and test and come up with a workflow. I will make a video about it, will let you know once ready.
@@controlaltai ohh great dude, when stay sure please share me the json file, thank u very much, i would like use it for upscaly many chapters of a mange, it have a recursive structure of folder
Man you are the BEST!!!!!! I have looking for simple solution to just upscale image without promt (because i have bunch of them and i dont remeber what promt have i used) and your first method worked like a charm!!!! Thank you so much!!
I watched many videos, but your video was the easiest and best.
Thank you!!
Thanks for this video, mate.
I am trying to create a "celebrity LoRA" which needs high-res photos. The ones I found, a lot of them were pretty ordinary. I tried your setup and setting but whilst the photo did get larger, the quality didn't improve, noticably, at least. For my purposes, this seemed to work:
* RealisticStockPhoto_v1.safetensors
* sd_xl_vae (as a separate vae)
* RealESRGAN_x4.pth (Upscale model
This gave me less detail in the skin, bust as I'm after face shape primarily, those details weren't all to significant. The Add_DetailXL LoRA didn't add anything. So, I get tack-sharp hair, eyes and edges and soft-focus skin. Happy with that.
Also a json Flow for a batch SVD would be handy, imagine having 5 images, I need a sequence SVD from 1 to 2, 2 to 3, 3 to 4 and 4 to 5, would this be possible to make a setup in a flow?
Ermm I am confused. Your comment is referencing SVD but is posted in an upscale video. Please clarify.
@@controlaltai sorry, offtopic
It would be helpful is you explained the purpose of adding the KSampler node and why you chose the 0.15 values instead of 0. So what is the purpose of the ksampler node and why did you choose 0.15?
There is something called Pixel Space and Latent Space. K-Samplers work in latent space. When doing image to image its pixel space going directly to pixel space for upscaling. Text to Image upscaling requires denoising which is done in latent space. Since the image has to be generated first by the K-Sampler from the text prompt. The K-Samplers (laten space) will Decode and generate the image (pixel space) where Ultimate Upscale SD can be used to just upscale or add details while upscaling.
Denoise strength of 0.15 means whatever is the image output by the text prompt, will not change much. The Ultimate SD upscale will have only 15% of pixels to add details to. So this will just refine some details. Upto 0.3 value you will just notice refinements. The image will start changing at higher values. A Value of 1 would means a completely different image.
This value applies only when the mode type is linear or chess and not "none". None makes it ignore the denoise value and just upscales it.
Whats the difference between using other upscale models? I need to upscale 4 times but without new details but need a sharp new upscaled image.....
Well this is an old tutorial and looks at techniques at that period, although still valid we now have multi options. First let me know if you are upscaling image to image or text to image generations. The final way to approach both of them would be different without adding much details.
@@controlaltai well both, I started with a 1000 frames Animdiff, lowres, mp4, when upscaling I got memory issues, tried a lot but no solution. The way that it works now, I think, is creating a batch folder with my 1000 frames in it, upscaling frame by frame and saving new batch of frames.
@@SjonSjine 1000 frames, you will run out of vram, Split it in batches in seperate workflows, Do like 100-200 at a time.
Hello, sorry to bother you. For some reason when I open ComfyUI I don't get the manager button.
Hi, You need to manually install that. Please watch from 1min 30 seconds: ua-cam.com/video/hTSt0yCY-oE/v-deo.htmlsi=6PkShHBV7Tcsfhu0
😂😂😂😂😂
i installed ultimatesd upscaling but i cant find the nodes isnt working why?
Hi, After installing anything on Comfy, You need to close browser and the command prompt, then relaunch. Please verify if you have done that.
i have i see the file but i dont have the nodes@@controlaltai
Did you install it from the Comfy Manager? Node should be available in right click images upscale or double left click search enter upscale
yes and i dont fint@@controlaltai
Re confirm by going in comfy manager on the to left from all select installed, see if it shows up. If it does and the node is not available, uninstall and re install, restart, open comfy manager again, select update comfy, check if any update is there or not. Update, restart. Open comfy manager again, click on update all, then go in custom nodes, select update from top left and update custom nodes.
Why does my images look like its either have a green overlay to it or look like a bad painting? edit I'm stupid its because I use the turbo one
Are you asking that after upscaling, you are getting this effect? If you check the denoise level for upscaling, if in linear. If you only want upscale then in ultimate upscale turn mode type to "none".
Thanks bro!!! Do you have some workflow that can share for simple batch upscale for upscaly folder with subfolder (recursive)? I need some like this,.
Welcome. Let me check and see if I can find anything for batch upscale.
Hi, I found something regarding batch upscaling, however I am in the middle of a video for this week. I have to try and test and come up with a workflow. I will make a video about it, will let you know once ready.
@@controlaltai ohh great dude, when stay sure please share me the json file, thank u very much, i would like use it for upscaly many chapters of a mange, it have a recursive structure of folder
@@controlaltai would love to see batch upscale!
Sorry about it, I was working on it. I remember. I will continue working on it. Stay tuned please. Thank you
I upscaled a photo of a shoe but the result got hallucinated. ❌
Did you use none for mode instead of linear in the ultimate SD node? It will hallucinate only if details are added.
Should 16gb vram be enough for this ?
Yeh