Yes, the feature request is to fix this, so it’s actually usable. Forget if you love or hate it. Think if a Python shell to call an external app and pass a sequence back and forth, is the answer.
If you’re ready to build a spaceship, I won’t stop you. At least I took a step forward by filing a feature request and not just posting on the forum hoping for miracles.
you can already export files in the desired format and encapsulate that in openclip - why would anyone waste development cycles or money on resurrecting dead technology that had little or no uptake?
I rarely post on the forums asking for help. If Jon is asking for clarity on some topic, bear in mind, that without Jon there would most likely not be the role of Flame Artist. This isn’t just some random opinion. You forget and encourage others to forget… Shame…
Hey now, I think we’re all gettin a little riled up. Sounds like some miscommunications happening. You and Jan are both awesome contributors to this spot, it’s like watching my parents fight! It makes me very un-comfy
I like though how Photoshop incorporated ( as weird as the charging is ) Nano and Flux Context natively and in an other case in a paid plugin seedream +flux2 pro as well
also they went 2k with firefly .
I think Autodesk dev should get their hand on this .
Python like Steve did , API ? Not sure what to say about the whole situation .
This should be a corporate level discussion within Autodesk. As all their software packages should utilize AI models within their specific environment. It’s not just generative AI they could use but also LLMs.
I’ve been using Google Antigravity editor to prototype matchbox shaders with incredible speed. I believe the big change with AI applications is going to be AI agents that can speak to your software and work for you.
You’re generally spot on. This has been the ‘big thing’ in the last few months. Generally referred to as Agentic AI. To the point that Microsoft announced they would make Windows the first agentic OS. That didn’t go over well at all, big PR face palm.
The problem is that AI is not reliable enough, and is unlikely ever to be, as it’s not meant to be precise, repeatable, or accountable. Agentic AI is like given someone a loaded gun and hoping for the best.
If you are General Custer and nothing left other than a Hail Mary, worth a shot. But in most other scenarios best to be considered very carefully.
The have been valid exercises where folks had AI suggest the steps to be taken, then reviewed by a human and approved for execution. That’s a way of managing the risk, but you lose some of the scaling and speed advantages.
I can try to find old articles that cover this topic if interested.
A sort of plugin system that sends image information to a external app,
sounds like openFX?
if blackmagic was able to run all of resolve as ofx / why cant comfy? is there a technical limitation?
I do agree tho having a easier connector like a modern version of pybox would be nice.
But then, comp tools really need to deeply integrate these tools like comfy does. Flame would totally be a prime candidate for native AI nodes-
i imagine a new context like action but for ML, bascially “comfyui” inside of a “action type node” - the best thing is that adsk can shift all AI responsability to the user,
Yes, but OpenFX is complex C++ code. That severely limits how many of us could tinker with variations as part of the community, as compared to PyBox equivalent.
2 Likes