Post n° 19
I'm still looking for a completely free and open-source program (installable locally) that can help me process old static images. There are currently many programs online that do this, but they all come with significant usage restrictions. Besides blackmailing you into creating accounts controlled by limited credits, you have no idea where (or who) the images edited online end up...
In recent months, I've had the opportunity to try Invoke AI (locally), but it turned out to be a disaster on all fronts. Invoke has incomprehensible features and functions, named with acronyms and abbreviations that are impossible to understand. Invoke requires you to download tons of packages and libraries with obscure names that make no sense, are meaningless, and lack any intuitive meaning. It's a program that can take a whopping 40 minutes to do a simple static 2x enlargement, and once it's finished, the result is always a disaster: fake eye pupils, shaped eyes instead of round ones, a lack of graphic detail and definition, crooked fingernails, unfinished, poorly rendered fingers, crooked feet, poorly rendered toes, etc. ETC. You also need decidedly expensive video cards to achieve decent resolution times. I spent €200 on a video card (with 6 GB of VRAM) to still have something that takes almost hours to render... This is absurd! Do I have to spend at least €800 on a video card? You're crazy!
I need a ready-to-use program, a program that restores images logically: I insert the photo and then, in a single text prompt, I write what I want to do with the image. This all needs to be done quickly (less than two minutes) and in a single operation.
Example: I upload a small (345 x 986 pixel) photo from the 1990s and start typing the entire text prompt: Resize this image to 2500 x 2500 pixels by changing the background behind it and replacing it with a sunset at the sea. Improve the quality of the foreground subject by adding detail and good definition to the entire body. Change the subject's clothing. Complete the subject's head (in many cases, the final part of the head is incomplete or cropped...). Create a smiling face... ETC. ETC.
This should be the procedure for a program like Invoke as well. Currently, Invoke does none of this. Furthermore, the libraries to download should be significantly fewer and significantly smaller in size!! There's no point in downloading dozens and dozens of GB of libraries (and various Lora packages...) to still get decidedly mediocre results. The program needs to be much more automatic, it needs to give you the highest quality in every case, not that I have to make hundreds of attempts (varying all those incomprehensible parameters) until I get the results I want, because doing so takes me days and days to get the image I want.
Currently, Invoke (like Amuse) is an "assisted abortion"!
The functions I need in photo editing software:
- Magic Eraser (to remove parts of drawings or superimposed text)
- Background Change/Removal, to replace or remove: backgrounds, objects, clothes, details, particulars.
- Enhancement: Enhancement includes improving the details and image quality.
- Image Enlargement: Enlargement must be proportionate to quality: the more I enlarge the image, the more detail quality there must be within the image itself.
- Generative Expansion: The program must add parts of drawings to achieve the image format I want.
All of this should be written in a single text prompt, a prompt that I write right at the beginning without having to go through a lot of mechanical steps (as unfortunately happens now in Invoke...).
It's obvious that all of this should be done through vector graphics, as I've already explained in my text... HERE. Until the communication bridge between vector graphics and raster graphics is created, there will always be many problems to solve...
Nessun commento:
Posta un commento