Imagine that: Stable Diffusion being lousy at generating male characters! I'm shocked. 😆

I've got my a1111 system up and running locally, and have been using it that way for many months. I've also got Comfy and Invoke and even a few other tools (Mochi Diffusion, Diffusion Bee, and a CoreML Python-only implementation), but I use a1111 for most all of my stuff. I can't feel comfortable in ComfyUI for one thing. 😜

Unfortunately, there seems to be a bug with Torch 2.0 on macOS 13.3, and it likes to throw out of memory errors when generating even a single 256x256 image. (That affects all three: a111, Comfy, and Invoke.) The only workaround I've found is to go back to using Torch 1, which is slower and doesn't have as great support for MPS (the interface to GPU on Mac, in place of Cuda). I don't really want to run on a shared system like Colab for various reasons. But it's good to know that Comfy does still work on Colab, should I need to use it.

In terms of space, I've got a 2TB SSD devoted entirely to SD, and about 4TB free if I need it on my main data hard drive, plus a couple more 4TB drives I'm not using, so space isn't really a big concern for me. (A huge change from when I first started using computers... a 60MB hard drive was huge, and 10-20MB was standard. My first PC actually didn't even have a hard drive. Just two floppies.)

@shiowoneko Hmm, yes, that's a good point. At 1.8gb, I definitely wouldn't be downloading 4000+ of them like I have for SD1.5 lol. But that happens with increased size. People think, "1024x1024 is twice the size of 512x512." But it's not. It's 4x the size. So it makes sense if the LoRAs are around 4x the size too.

It's possible I can't even do XL on my system, since I haven't even tried it yet. If so, I'll probably have to do like you and either start using RunPod... Or just not do XL.

It's crazy just how fast things like storage space evolve. In 10 years, my 12TB data HD will be puny in comparison to what will be on the market. No doubt they'll have drives in the 100–200TB range, if not 500TB. That's if they keep making HD and don't switch entirely to SSD. In terms of SSDs, probably 50TB would be pushing it. But in 15 years? Maybe they'll be breaking the PB (petabyte) range. Putting that in terms of the HDs of my early days, 1PB ≅ 10.5 million 100MB drives or 105 million 10MB drives. Boggles the mind. My first MP3 player, which I got in 1999, stored 16MB, enough for 4-6 songs. 😆 (No, I wasn't stupid enough to buy it; it was included free with a CD writer I bought.)

If we had a fast, easy, cheap way of storing things in DNA, our storage space would explode. DNA has a storage density of approximately 1TB per mg. The data on my 12TB HD would fit in 12mg of DNA. The entire Library of Congress is estimated at 40PB, so it would take 40g (1.4oz if you're American like me) of DNA to store it. But there are also newer forms of NVRAM on the horizon, like RRAM, that store data at a higher density than current SSD (NAND NVRAM) technology. So perhaps that technology will cause data storage to grow exponentially.

@shiowoneko $40,000 US. Wow. Insane. But I had no idea that 100tb was on the market already as an SSD. Or even as an HDD, actually. Guess I'm pretty short-sighted when it comes to speculating on future technology. In terms of consumer/desktop equipment, it looks like 16Tb is the largest SSD and 20Tb is the largest HDD as of June. That doesn't sound quite right, though, because they had 18Tb HDDs in 2020.

@shiowoneko No matter how much storage space you have, you can always find some way to fill it up. —Dita's Law

@shiowoneko That's how it goes, yes. I had to delete a whole lot of anime a couple years ago, and I'd rather not do that again. I've found in the last few months that I wanted to go back and watch some of it again. So now I try to buy more storage ahead of time, before i run out of what I've already got.

@shiowoneko Ah, that's a title I haven't seen in... 10 years? Maybe 12. I actually only started watching anime in 2009, at the age of 33. But I've made up for it. My MAL stats show I've seen 21,060 episodes (1110 series, 121 OVA, 93 movies, 101 specials, 340.6 days). Around 4 episodes a day on average.

@shiowoneko I don't watch nearly as much as I used to, so my list is growing a lot more slowly than it was the first 5-6 years. At that point, there was so much I hadn't seen, and I was focused primarily on older series, so I was always marathoning them. I watched all of Naruto and the first 200-some episodes of Shippuuden in just over 2 weeks. Then when I started watching seasonals, I would watch just about everything. Until a few years ago, it was pretty common for me to be watching 30-40 series per season. Now I try to keep it in the 20-25 range (ideally 21), because I mostly only watch while I'm eating. So 21 episodes in 7 days. Which is still a lot. But 20 episodes a week compared to, like, 150 is a big difference: 8 hours vs 60 hours.