DGB.LOL - Upscaler models explained
HTML-код
- Опубликовано: 13 сен 2024
- A video tutorial which tries to explain to you, which upscaler model to use in which cases.
Feel free to use my completely free AI Image upscaler, and other useful image processing tools on dgb.lol
✨ A wonderful walkthrough of the tools!! Thank you, Tom!! ✨
Nice work Tom, this helps immensely, thanks so much!
Great walkthrough!! Thank you for the information.
Mate, please consider simplifying the model names - to Sharp x4, Sharp x6 and Sharp x8, for example. As your naming convention utilizes Greek letters' names, they can be a bit distracting and confusing to noob like me :) Thanks for an awesome and free service, I use it for a lot of geometric drawings. No frills, everything works. Thanks again!
the greek letters has meaning, the letter position in the alphabet is the number of trials for that model training :)
Between the DGB LOL "Lambda and LOTA" I have 2 questions:
1- Which is better?
2- What is the difference between these 2?
@@veteranocolosal8740 1) these are performing better in different image styles, for illustrations of abstract things maybe the iota, for stock images (good approval rate) the lambda, but feel free to try and compare the results. Lambda is trying to avoid the "fishnet" effects which are being genereated usually as a glitch in AI generated images, the Iota is trying to make the images more sharp and crisp, they both have additional mini-photoshop trained inside them which removes noise and obvious artefacts.
@@dgblol Thanks :)
Thank you for this useful toolset!
Could you add a tool to anonymize a photo by blurring all faces and car plates in an area (like in pictures from Google street view) ?
This may be useful to be compliant about image right management for people who don't want their face to be publicly displayed on Internet.
Hi, I am not planning it, as you should have the rights before upscaling the images. I am working now on different things, let's say more widely useful :)
Thank you for sharing so much!
I wish for an decent Upscaler that is like the others but with results that are a bit less thick, if you know what I mean.. Like the AI effects the original image's quality/texture with a bit less strength/intensity. Think like as if it were a filter with a adjustable slider. I don't mean the AI upscaler has to be adjustable, I just using adjustable filters as an example in my attempt to explain my thoughts. Some of the images I try to upscale, end up with a texture that completely ruins the image. Usually it covers the original image's fine details. It effects actual photographs of people. The skin taks on a very unnatural fake look. It almost looks like a layer of putty/plaster or clay or plastic or something. Hence my use of the word: "thick" at the beginning of this comment. Because it looks layered on too thick. So, I donno if my idea would even be possible, but I was thinking something like a lowered opacity of whatever the AI does to the image. Like a transparency filter or something. Just to reduce the "thick" plastered layer and let more of the original image's details shine through, while also upscaling the image as much as can reasonably be done. Just an idea that's been floating around in my head for a while now.. I figured I'd make a suggestion whenever I had the chance.
The upscaling is being done by a particular model, which has its own rules, there are no such things inside the models as settings for thickness. They are being simply trained for a particular setup. That's why i have so many models on dgb.lol . They are basically the settings itself :) What you can do, either before or even after upscaling is to put the image to a software like photoshop or gimp and post-process the image.
@@dgblol I don't know much about photoshop, at all. The closest thing I've done to manipulating images is messing around with brightness, contrast, or saturation in Windows default photo edit thing. That and using AI upscalers. So, I wouldn't even know how to do post-process. lol I'm not even sure I know what it is, exactly. 😅But thx for taking the time to reply. I love the site btw. 👍
@@Chadw1cKed you don't need to post process :) (edit image after upscaling) . You just simply need to find your best upscaler model, and use it, that's it :)
Thank u. This was so helpful.
Sad this AI image generator is gone from the site.
What I found most interesting. Will this come back one day?
It's back with even better results
@@dgblol Many thanks for that, Tom.
You are one of my favorites on RUclips now ☺️
I quickly became fascinated with the "engine" behind re-imagine.
It's simply incredible what I've made out of mediocre pictures.
It's like night and day with many of them - I don't understand how you manage to create something like that - and it's supposed to be even better? How 😁
I left a message on Discord a little earlier.. also thanks for the reply there..
Take care.. and thank you..
@HSP..
Thank you so much
I am using it to sell my art prints, 24x36in . Which model do you recommend?
It strongly depends what is the content and details of your images, my lifehack is, that i am using balanced+ for everything :D
I am using this for mockup picutres. For example I will be upscaling images that have clothing. Which model do you recomend for this / other mockups
I was doing the same for a while, the best for me was the BalancedPlus for that. Also it depends what is in the background, if it's a studio, or nature, and how complex it is, but i would say start with BalancedPlus, if it's not good enough, go with euther smoothlol , or lambda
Re-imagine was awesome ☹️
It's still! It's back :)
amazing
I have a question between the DGB LOL "Lambda and LOTA" I have 2 questions:
1- Which is better?
2- What is the difference between these 2?
I've answered you in that thread below :)
@@dgblol Yes I saw it hehe
Thank you very much again :)
I'm from India and I'm not able to use it bruh
what should i use for real people
?
I would recommend balanced+