Xordel
patreon
Renders + IA?
🕑 Added 2023-08-09 13:05:56 +0000 UTCHi guys! I have been testing the AI, I have been trying to improve some very old renders with AI stabble diffusion and controlnet, and although I still have a lot to learn I have share with you some interesting results
I will try to don't do an extremely long article xD so I will try to resume, I have discovered that like any tool it has pros and cons.
Pros:
- By using an image as reference pose. is possible create visually spectacular renders.
- Being able to reimagine old renders and turn them into incredible posters.
- Using low strength values you can keep the essence of the original render.
Cons:
- If high values are used the result can be radically different from the original render.
- Tons of inconsistencies in the images and many errors in the hands.
- A lot of editing time is required to create a piece free of visual errors and inconsistencies
- The debated topic that being models trained with images from the internet, the result of the content is not "original", however, as you will see in the examples I did, the renders are still very similar to the originals.
Example 1: with a really old render, in this I only try to improve only the face and genitals
Original:
Original + IA:
------------------------------------------------
Example 2: In this one, I try to do some similar with color adjustment to try the tool
Original:
Original + IA:
Notes: The nipple details are much better, open the image larger so you can see the details
----------------------------------------------
Example 3: In this one, although the face is not identical, I think it is much more in line with what I expected to get at the beginning of the render, it's almost like the "expectation vs reality" meme π€£
Original:
Original + IA:
Notes: You can see how much the whole image improved, textures, details, face, lighting, shading, basically everything is better
------------------------------------------------
Example 4:
Original:
Original + IA:
In this one, open the image in larger, the datails are much better
------------------------------------------------
But not everything is to make the renders always exactly the same, also is possible to give little more freedom to the IA and you will get different results, for example the imput for the next example was it:
And the results was the following:
01:
NOTE: As you can see, the base essence is still in the render but it has a lot of changes, it's also clearly visible how many inconsistencies they have in the image, hands, face, foot, etc.
02:
NOTE: Inconsistencies continue in hands, face, foot, etc. but to generate this image they have new elements like rain and wet floor.
03
NOTE: the same place and girl but with diferent pose in a rainy night
04:
NOTE: completely new place this time is a beach, but keep the base composition
In this last examples It doesn't keep the dva style, but if I wanted to, I could, just this example is to show that you can do any style, even manga or anime.
So to finish, you can see that from a base image, you can have infinite images and styles that you can work on to bring them to do a final polish.
If the faces scare you, it's because I don't work on them, these are sketches to continue working on once you've selected the image that you like, in other words, these images only serve as a basis to continue working on them
The faces are very easy to repair and place the character you want, you just need a lot of patience and time, but very good results can be achieved, perhaps a more annoying and more complicated problem to solve is the issue of the hands but there are ways to do it anyway.
So, wha do you think? would you like to see renders finished in high quality with the help of AI? Or do you prefer to continue enjoying the "original" renders? Let me know in the comments, since for my part I would like to start using the tool to create much better content, π
Comments
Jim
Go with it (AI) !!
MpMan
with some mastering of ia generated image, you could stop entirely to do render but if you go full ia (or 90%) , video with ia is still not good
