AI tools that you use (or want to use) for editing

[INIGHTMARES]

After taking a quick glance at the Ebsynth website, which I was unaware of before, the way it turns live action footage into moving water-color paintings is so cute

That's not what it does technically. With EbSynth, you can draw over a frame in any style and the TLDR is that when you run that frame through EbSynth all the frames in the shot will look like what you drew. It doesn't just make everything you throw at it look like a water-color painting.

 

[MightyAttackTribble]

After taking a quick glance at the Ebsynth website, which I was unaware of before, the way it turns live action footage into moving water-color paintings is so cute. I could see a 'What Dreams May Come' edit utilizing this to great effect.

EbSynth is a very useful tool, but you do have to put a lot of work into it to get the best results – it's saved me from having to re-colour about 60% of the frames thus far in my Transformers edit.

It's also worth noting that it's not an AI-based app and does not claim to be, as it functions closer to an automated batch script in Photoshop than a true generative AI.

 

[INIGHTMARES]

It's also worth noting that it's not an AI-based app and does not claim to be, as it functions closer to an automated batch script in Photoshop than a true generative AI.

Ya I sort of mention this in my first message. But you're right.

The term AI is pretty general if you think about it but I am wondering about modern AI tools that are cropping up

 

[Zarius]

Am amateur writer, had a novel almost published a while back but the agent dropped me without much warning. I do use AI generated imagery for some accompanying art that goes with my work, but for the most part it's never 'generate pic -> slap title -> publish', I always tweak it in some form through Photoshop, be it correcting image mistakes, but more commonly just generating individual elements and compositing them into something I deliberately made.

I publish a Halloween webnovel, my first 100 pages are out, and I used AI generation to make things like the cover (heavily edited) and the scene break images. But this is something I justify by telling myself it's a free project that anyone can access. If and when I get published, I'd be paying for covers by smaller artists.

At least you have the skills and resources to make that happen. I'm not that gifted, but, as you say, if and when the time comes to publish an original work of mine, I know who to contact for art.

Let me know how you get on with publishing sometime.

 

[tremault]

EbSynth is a very useful tool, but you do have to put a lot of work into it to get the best results – it's saved me from having to re-colour about 60% of the frames thus far in my Transformers edit.

It's also worth noting that it's not an AI-based app and does not claim to be, as it functions closer to an automated batch script in Photoshop than a true generative AI.

That's really interesting, looking at it now, it seems very much related to that data moshing. since video encoders already work by taking keyframes and applying motion vectors to the pixels, it's not so unusual that you could take that motion data and apply it to a different image. That seems like an incredibly useful tool and it certainly is not generative AI.

 

[Moe_Syzlak]

Gah, this forum software is definitely not intelligence of any sort.

 

[Moe_Syzlak]

I had some time to think on an eight hour drive yesterday. And, while “AI” is certainly all the rage these days, I’m curious as to what really constitutes AI. I was using Google Maps to drive back to Germany from Italy at the end of a big holiday week here in Europe. The app was constantly assessing real time traffic data, making predictions about whether the traffic would change (for example a crash that was still several hours ahead of us on the route that Google deemed would be clear by the time we reached it), and suggesting alternative routes and updating our predicted travel time. This is data in/data out. Similarly I’ve used a music program (mostly for jazz) called Band In A Box for maybe 20 years. If I want a band to comp me I can choose the instruments, the style, even a player to emulate (so say I want Art Tatum playing a Moog synthesizer I can get that sound). Again, data in/data out. But none of the so-called AI things I’m seeing seem like more than that to me. It’s still data in/data out. It’s just that tools have a much broader base of knowledge to draw from (more data in equals better data out). It seems like we crossed some major threshold with the way everyone is talking about AI, but to me it seems like we’re just on a continuum and it’s just all a relatively competent Clippy. What am I missing?

 

[tremault]

AI is a nebulous term indeed. I've heard people say that once a thing becomes reality, it's not AI any more.
It's mainly being used at the moment to describe neural nets that are trained to create a highly complex algorithms, also called learning models. It's just a whole process of take an input and showing the neural net the required output and letting it use trial and error to compare it's algorithm's results to the required result. So the input they give is the description of an existing image and the learning model tries numerous different permutations of an algorithm until that input text results in an image that matches the actual image. It's a really high tech mimic.

I think a lot of people get confused because this mimic is at such a level that it can produce expected results in such a way that it appears to be thinking, but it's actually just producing a randomised version of an expected response. They aren't able to reason so they aren't actually intelligent.

we have to consider the question though, what is intelligence?

Our human brains have a number of various processes that work in unison to produce our ability to operate. we have image recognition systems and are able to predict things to a remarkable degree of accuracy, we're able to analyse and reason. these AI programs are achieving one small subset of what we actually do and there is no real progress in producing AI that can do all the other more complex stuff we can do.
The AI is still dumb and nowhere near actual intelligence.

 

[Moe_Syzlak]

I think there are some incredible (and alarming) things being done with AI recently, but I agree with your assessment. It’s still pretty dumb. It’s not Clippy dumb, but maybe that’s why we’re impressed; low expectations.

 

[Zarius]

WOO HOO. I'M A HUMAN BEING!

 

[INIGHTMARES]

I usually don't even read the updates to Topaz, but in one of the recent updates it added the ability to add LUTs to the render. I'm gunna have to try it since I can't do that in my current NLE.