New uses and methods with Self Trained AI

[KageRyu]

I had thought there was already a thread discussing AI theory and use, but I can not seem to find it now. I found this following video very interesting in it's theories and applications - and the discussion of the experimentation and work needed to not only custom train the AI to get the results they wanted, but post work to create a unified look. Of course, since it involves using AI to make animation there are lots of "reaction" videos calling this offensive, laxy, worthless, etc... but to me it is clear that the makers of many of those videos did not watch this video or did not understand it well. I have yet to watch the animated clip they made and discuss here, but this makes me wish I had access to the diffusion tools (none of my machines are able to use them). To convert my videos to toon style I still need to do it with batch scripts in a graphic processor (and I lost the scripts I had been making in that HD corruption last year).

 

[Electro-harpist]

Thanks KageRyu, that's certainly an interesting workflow they have there. The ability do 'stable' eyes and their expressiveness, from frame to frame, would be the important part. For making comics rather than animation, the 'believeability' and 'human-ness' of expressions would also be key.

Hopefully we'll soon have a Stable Diffusion plugin connector for PostFX in Poser. Send PostFX a quick basic render made with Poser's real-time Comic-book mode (in Poser 11 through 13), and get back a professional "slick anime" AI makeover of the image. Thus giving character-consistent AI makeovers for Poser renders. How that would run for animation, rather than comics is unknown. I mean in terms of temporal stability (i.e. flicker and wobble between frames).

In other posts you say you have an old workstation and you seem to use Poser. If your machine is a HP Z600 with its original drivers, the ideal for 'old Poser', there should be no problem about putting an RTX 3060 12Gb card in there, and thus getting Stable Diffusion up and running with something like the InvokeAI package. The only other thing you'd need would be a '10cm PCI Express PCIe 6 Pin to 8 Pin Graphics Card Power Adapter Cable' to connect the PSU to the card.

 

[KageRyu]

@Electro-harpist it is also a question of OS and finances. Currently I cannot make any upgrades.

 

[KageRyu]

@Electro-harpist By chance, if you have used a HPz600, is there a specific graphics card in that series you would recommend. I was looking a a few , but I am not as up to date on newer tech and have not done much actual work on rebulding computers since 2012, except my own machines which most are old single and dual cores from pre 2010. Except my HPz600. That one I have never really opened up or dug into, and I can not seem to find reliable specs online (lots of conflicting information online).

 

[Electro-harpist]

NVIDIA RTX 3060 is the sweet spot between power and price. You're in luck... "Putting a 3060 card in a Z600" Putting a 3060 card in a Z600 – MyClone Poser and Daz Studio blog and "Installing and troubleshooting a RTX 3060 graphics card" on a Z600 Installing and troubleshooting a RTX 3060 graphics card – MyClone Poser and Daz Studio blog

And if you have Win 7 on it, as you should for the dedicated drivers, and want to do AI image generation, you need this... Invoke AI 3 For Windows 7 SP 1 : Free Download, Borrow, and Streaming : Internet Archive

The problem with a HP Z600 these days, for more than AI images, is the Xeon CPU's lack of AVX2 processor support. This means you can't run local AI LLMs (large language models, aka AI chatbots), and likely other cool AI stuff that is coming in the near future. The best options there seems to be a new PC for dedicated AI work with either the Winux 7 OS (Linux Mint made to be as much like Windows 7 as possible, and with a built in Win 10 VM), or Windows 11 and strip out all the privacy invasion crud with Powershell scripts - and then also wait until there's a Win 7 makeover mod for it (coming in the summer).