[Megacloud] Daily Upload Post September 15th & Access Member Previews (Patreon)

The Megacloud Link is here......unless it stops working again xD: (Accessing this site as an Access Member will be denied due to Sign In with Patreon lock for only cloud and up members)

https://rule34diffusion.com/rule34-diffusion/

===== [Anime Day 2 Content] =====

A note on AOT Characters*(Read at the bottom)

Mikasa Ackerman (Fixed) Sexy Poses [Gen 2] (I redid her again just in case, I had a hell of a time with the AOT girls, read below to see why)

Mikasa Ackerman (Fixed) Full Body [Gen 2]

Mikasa Ackerman (Fixed) Bleach Poses [Gen 2]

Ymir Sexy Poses [Gen 2]

Ymir Full Variety [Gen 2]

Ymir Beach Poses [Gen 2]

Esdeth Sexy Poses [Gen 2]

Esdeath Full Variety [Gen 2]

Esdeth Beach Poses [Gen 2]

Jolyne Kujo Sexy Poses [Gen 2]

Jolyne Kujo Full Variety [Gen 2]

Jolyne Kujo Beach Poses [Gen 2]

==== [Character Request] ====

[Cloud Plus Tier Request]

(I thought the first masturbation option was supposed to be standing...so I have to fix it)

[Character Request] Ultimate Viessa [Gen 2]

AI Problems xD

So it seems like some characters still had a problem, Ymir still didn't work, but worked when using the built-in prompt that came with Pony Diffusion, she was the same case as Zero Two, where no matter how much I trained they didn't show up without the built-in prompt.

The issue with the built-in version of the characters is their eyes are terrible(Zero Two's eyes were horrendous), and Ymir looks off overall on the face in this case(I had to weaken it to mitigate the issue, without it, she doesn't look right).

Seems to me that Attack on Titan was much trickier than I had thought, Annie also had this same issue as well and I was forced to use the built-in version, however, she wasn't too bad as the other two luckily, I really wanted to make her right.

So yes some characters are somehow impossible without using the built-in versions already trained on the base model. Some characters look great, and some are just not so great.

Another interesting thing was Attack on Titan style was built into the pony model as well, however using it too strong gave them a very strange look, this was noticeable when I first attempted to make the girls last week.

and well there's more but too much to explain, it's honestly wild how mixed up things can get when you run into many problems and try to figure out which one is which xD

I am looking forward to the new Pony Model as the base one I use was trained on 2.6 Million images, this new one has hit 30.5 Million..........so far......xD But has been trained with better quality dataset and better organization of everything.

I do believe this model can be about 15GB in size if it will be this large, luckily the UI I use is Forge and it's amazing at handling models well, my Old GPU the 8GB Vram 3070 was able to run a 12GB Model (it was fp32 so the size was not because it was much bigger in scale but just better accuracy, fp16 is about half the size with less accuracy) and not loose much performance, note AI models get loaded onto the GPU's Vram, and Forge has some amazing optimizations that allow lower Vram cards to run the bigger and better models.

I do plan to use the new Pony Diffusion once the creator has finished training and beta testing, and It will be the backbone on my Gen 2.5 works, unless it's a huge improvement I might call my works Gen 3.

If you read this far thank you for listening me ramble on about AI xD also, Sexy Poses will finally get Dynamic locations and I plan to implement a way for people to choose them for Custom Sexy Poses.......and I might allow ANOTHER Custom Poses set for people who already requested the new poses & outfits on the same character again. I think it's only fair for people to finally get to choose locations.

I plan to add more as I do more research on what to add before I release it as an option for custom poses first though, so stay tuned!