@fredwarren@Slabrie & @LouisM … I just used the new Morph tool on a job for the first time and OH SWEET BABY JEEBUSS!!! You and the team really knocked it out of the park on this one. Thank you, and please thank all involved!
I have to say - I’m so thrilled with the quality of the native MLTW and now the ML Morph tool. It’s fantastic to have ML tools that reliably and predictably work. So many ML/AI tools out there, especially for things like roto, fall short of the level of quality that is needed for high end VFX. No criticism…it’s just the nature of the beast/state of the art for AI/ML. Bravo for focusing on tools that work so well!!
If you have content that could be shared so we could see the issue that would be awesome. Of course, we would not train our model with your content but re-create asset to handle the content you have seen issues with. As we stated, these ML tools can only get better when trained with the right content.
But there is the crux of the issues with AI/ML. The legal implications of us sharing any data with you is problematic. It would involve productions paying lawyers to go through legalese to work out how and how not any footage could be used, accessed, stored, etc; With a strong anti-generative rhetoric most studios are more wary than embracing.
What I would suggest the Autodesk M&E product team do would be to have some kind of accessible document that clearly outlines how footage would be used to train ML, outlaying everything. I would also suggest that you could store LLMs based on what rights each bit of media is given to. Then, could Autodesk potentially have a LLM per studio based on rights and access?
All that sounds pretty complex right? Unfortunately, I think that is going to be a legal reality where Disney, Sony, Warners, Netflix, Amazon Studios, etc; will have their own LLMs that you will be limited on using depending on which studio you are working for. As for footage from/for TVCs, wow, that’s even more complicated.
The content request is not for training our model but to see what we need to create as far as content to handle you use case. We would never request content for training since we are very tight on what is used for training. We do not borrow or license content.
We have pretty tight GDPR rules at Autodesk so you can trust us. But we also understand that it might be impossible to share assets event for analyzing so no worry.
