With this explosion of AI, I thought I would start a thread with ideas as to how flame could better embrace AI. Almost an AI feature request thread. We should then look to add a whole lot of the popular ones as feature requests. My first request would be to have a decent sized AI R&D team within the Autodesk Flame R&D team. I have no idea of the makeup of the team or how many are across AI but whatever it is, I’d love to see it ramp up.
Feature Request 1 - AI text to particles
As mentioned endlessly in Logik, the particle system is a dinosaur compared to other systems. How cool would it be to have a text to AI particle generator. So you could type snow, rain, small flame, bonfire, etc;
The node could potentially be linked to a gmask or axis so AI knows where it needs to be generated from. Not only that, how cool would it be for a text to control generator. For instance, you may decide you want speed, number/density, size, direction. Generator size, colour. You could type those in so the AI tool will generate a slider or value system so you can adjust the results. Then you could potentially have child AI particle generators, to add smoke to a fire, or to ask for rain to bounce/splash, create puddles. Then have controls for that.
It would also be cool if there was a way to feed the AI particle tool some reference imagery to help it’s starting point, whether that be a still or moving video.
I would have thought the above would have been unachievable only a couple of years ago. I kind of feel it is now not unrealistic and how cool would that be?
Feature Request 2 - AI excel or csv to end credit roller/cards
I’d love to be able to import an excel or csv file into flame then let AI do it’s stuff to break that out into an editable text setup (or sequence of text setups). Where you can input somewhere that you now want Comic Sans as the font and it adjusts everything for you.
This is just a start. I have plenty of other ideas but I thought I’d open it to everyone else on Logik.
We are still fighting to have a Paint module that doesn’t corrupt itself, or Motion Vectors that don’t forget themselves, or a Full Range SDI signal. We are fighting with basic software functionality, and Nuke is realizing the future.
Conform. Here’s the xml, here’s the reference video and this is the source. Please conform it in a 10bit HD timeline with BFX setups and with sources in full res. Also create social versions…
I completely appreciate where you are coming from Alan. The playback issue for longform on Linux is also concerning. One of the reasons we moved to finishing in Resolve is because of this. Plus lack of Dolby Atmos support. Plus inability to create IMF masters or supplementals directly from Flame without paying for an add-on that has very limited functionality. Plus the Text tool having limited functionality. Then there is the wide array of add on tools that simply aren’t available for Flame. On the flip side though, there are other tools have a multitude of their own issues that really aren’t great (Nuke Studio is a dog’s breakfast and can have its own playback issues, Fusion setups have a habit of getting corrupted and the cache tool in Resolve is untrustworthy). I think we are at a point of time where there is a real opportunity for Flame to become a leader again but if there isn’t a bit of a refresh on its approach it will slowly dwindle away when it reaches a point where there aren’t enough users to justify the cost of developing it. I am not suggesting that will be any time in the next 5 years or anything. It did not fill me with hope to hear of Grant Kay leaving either as Flame is lacking pathways or easily accessible learning versions or marketing to develop the next generation of Flame artists.
Just imagine if Flame even have a simple interface where Flame could connect with something like runway.ml and treat it as an OFX plugin. I’d prefer something native of course, but I’d settle with something that resides within the ecosystem. That would be something worth marketing.
A node within Flame that is an interface/gateway to runway.ml or other similar online AI (or other) tools.
I kind of think that having a system that has all features natively developed for it is not going to be financially viable moving forward. Having a really solid way of interacting with tools outside of the flame ecosystem would be amazing, especially with cloud based offerings such as runway.ml
In a slightly different way to how things like Omniverse and USD are enabling this more in a CG environment.
Yeah this is my feeling. The development team could end chasing their tail up trying to mimic every AI development being created now. I don’t know exactly how it would work, but some way to harness these developments and better integrate them with Flame would be amazing.
Loading PyTorch models like Nuke is a good start, along with a widget creation setup that artists would also have access to in PyBox and Glsl as well.
That would open flame up to a bevy of community driven developments while shielding autodesk from the liability of where the datasets and resulting models come from…
Matte painting AI … much like the many AI image generating tools, this one would generate matte paintings based on descriptive input.
Snow covered mountains? Forrest? City scape? Suburban street on Christmas night? Airport terminal view? Etc.
After using certain AI Tool in Nuke for FG extraction from BG in non blue/green screen plate I would say adding a similar feature to GMask Tracer just makes sense.
