1. We've had very few donations over the year. I'm going to be short soon as some personal things are keeping me from putting up the money. If you have something small to contribute it's greatly appreciated. Please put your screen name as well so that I can give you credit. Click here: Donations
    Dismiss Notice

Video Upscaling software. (Before & After pics)

Discussion in 'Tilted Gear' started by MeltedMetalGlob, May 4, 2025.

  1. MeltedMetalGlob

    MeltedMetalGlob Resident Loser Donor

    Location:
    Who cares, really?
    I didn't know where to put this, and I also didn't want it to seem like I'm spamming, but recently I've been experimenting with upscaling software to update my older projects- it's a mixed bag and the tech is getting better, but not quite there yet.

    I'm using a short video I assisted on as a test. The story is about a reporter who realizes his government is starting to crack down on individualism; I play the reporter's manager, a supporting role.

    The short was shot on Hi-8 video.

    The shots:
    (The software upscales & tries to provide detail only; the color correction is my work.)
    [​IMG]

    [​IMG]

    [​IMG]

    This one has incredibly accurate detail:
    [​IMG]

    My appearance: (I was 22 when this was shot back in 1994)
    [​IMG]

    The faces have an unfortunate AI-generated quality to them. Also, my skin was NEVER that perfect. If you examine the top pic closely, you can still see my acne & pockmarks.

    I've noticed in a few shots that the program does weird things- one character does not wear glasses, yet for a few seconds he's clearly wearing spectacles. Facial features also distort slightly in wide shots where information is pretty much guessed at.

    When I'm done with this, I may upload it for viewing. It's only 15 minutes long. Acting & writing is done by teenagers, so take that into account. (I was in film school and assisting the director with this project.)
     
    • Winner Winner x 2
  2. fflowley

    fflowley Don't just do something, stand there!

    Whoa.
    Totally out of my league. It’s unbelievable what technology can do today.
     
    • Agree Agree x 1
  3. Chris Noyb

    Chris Noyb Get in, buckle up, hang on, & be quiet.

    Location:
    Large City, TX
    Very kewl. Please do upload the end result.

    Is it possible for us to view both versions once your done?
     
    • Like Like x 1
  4. cynthetiq

    cynthetiq Administrator Staff Member Donor

    Location:
    New York City
    [​IMG]
     
    • Agree Agree x 2
  5. Chris Noyb

    Chris Noyb Get in, buckle up, hang on, & be quiet.

    Location:
    Large City, TX
    Or at least some of the original?
     
    • Like Like x 1
  6. MeltedMetalGlob

    MeltedMetalGlob Resident Loser Donor

    Location:
    Who cares, really?
    I don't see why not. Don't watch it expecting Oscar-winning performances, though. :p

    On a related note, I'm also experimenting with AI generating software and it's fairly disastrous- I can't get it to do anything worthwhile, but on the other hand I can make it do completely crazy nonsense, and I may just combine the two to make a laugh riot out of this project. (This project featured a brief fight scene, and I tried to lengthen it slightly with AI, but all it did was make the two combatants engage in a dance-off, and no, I'm not kidding about that.)
     
    • Like Like x 2
    • Informative Informative x 1
  7. Chris Noyb

    Chris Noyb Get in, buckle up, hang on, & be quiet.

    Location:
    Large City, TX
    I've read other comments about AI creating content that wasn't what the artist/creator wanted.

    Once you give AI an idea, do you no longer have control over what AI creates? Are you able to edit? redirect? during the process? These redundant questions sound good in my head, not so good in writing.
     
  8. MeltedMetalGlob

    MeltedMetalGlob Resident Loser Donor

    Location:
    Who cares, really?
    I honestly don't know how other people manage it; I'm playing around with a program called RunwayML, and any time I attempt to create something, I get random garbage or the image in question doesn't do anything other than "shift" slightly. (If I take a photo of someone and tell the program to make the person wave, the person will just move slightly and their face will change... but nothing else.

    Recently I went back after a few months after not using it and it's improved slightly, but I still can't get it to do exactly what I want. I don't know if my prompts are not verbose enough, or too much verbosity.

    This is their website (This is not the video software I'm using, that's an entirely different animal.)
    Runway | Tools for human imagination.
     
    • Informative Informative x 1
  9. martian

    martian Server Monkey Staff Member

    Location:
    Mars
    Yes. But also no.

    Prompts are self-contained because the AI is a prompt engine. It takes a prompt, transforms it into an output, process complete. But you can refine outputs, because the teams behind these things have put a lot of effort into making their use intuitive, and that includes what's called a "context window" where the AI will process previous prompts and outputs alongside the current one, meaning you can describe what you want in a conversation. The model will never intentionally subvert what you're asking or because the model doesn't have intent to begin with, but it's prone to interpolative errors, which can range from anywhere to imperceptible to egregious. These are basically inherent to the model and will probably never be fully solved, though they can be minimized with careful training.

    Images are harder than text because they take a lot longer to generate. So you get an output and then a correction requires another iteration, but you can't really break it down; it's going to generate the entire output every time. So that's just a longer cycle and more tedious.

    It's like trying to make a picture of your cat by describing it to someone who's never seen your cat, or maybe never seen a cat at all, and having them draw it for you. If your description is very detailed and accurate you can potentially get something very close to what you want. But there are always going to be some points of mismatch because the artist might not interpret your instructions in the way you intend. And also the artist is a baby. And maybe slightly insane?

    Anyway this stuff is still evolving.
     
    • Informative Informative x 1