What is the correct answer to the question, "I don't have a powerful GPU, how do I generate AI art?"
InkBunny's rules, as they're stated, require tools to be "open-source" (in broad sense). This makes answering the question about what tool to use rather hard. 😆
What's currently known:
1. 👍 (DO USE; RECOMMENDED) Graviti Diffus A1111 WebUI — publishes all code under AGPL, is a fork of A1111 SD, embeds full metadata in files. One of the best tools in existence, but it has recently banned free NSFW generations and even generating SFW art requires providing a credit card, which some people aren't okay with, even if they aren't paying anything.
2. 👍 (DO USE) StableHorde Image Alchemy — open-source crowd-sourced distributed cluster running SD and LLMs. Your prompts are executed on machines of other users in the horde. The choice of models is limited to whatever other users have and the waiting times can be very long, but it's completely free and has few restrictions on content. All code is licensed under AGPL, so it's undoubtedly compliant with InkBunny rules.
3. ❓ (AVOID USING) CivitAI Generate — unknown closed-source implementation, very limited functionality. Metadata is mostly fine: even though it doesn't include model names, all referenced models are essentially direct download links. The main problems are unknown implementation (and thus syntax) and max resolution of 512x768 (at the time of writing, upsclaing hasn't been implemented yet). The main advantage is that it's absolutely free and has very few restrictions on content.
5. 👎 (DO NOT USE) SeaArt — heavily customized closed-source image generator, doesn't provide metadata in a meaningful way. Was fine before it decided to ban NSFW. Extrapolating from known examples, administration won't find that website acceptable.
6. 👎 (DO NOT USE) PixAI website & app — closed-source implementation running only models located on the website itself (which includes both private models and models copied from other sources). On top of these attempts at isolating content, metadata isn't embedded into the files and restrictions on content are very strict (even suggestive poses are policed and checked). The website allows generating about 25 images of decent quality per day for free users.
8. 👎 (DO NOT USE) NovelAI — explicitly listed in the rules as an example of a prohibited website. It uses closed-source implementation of SD using private models and provides access only for paying users. At its peak, it was considered top tier furry porn generator, but currently it's largely obsolete. However, if you don't plan to publish on InkBunny, their current text model is reportedly competitive.
So, uh, what's left? How am I supposed to answer the question about generating AI art? "Get your credit card ready"? 😐
If you're aware of any decent online tools which fully comply with InkBunny ACP, please list them in the comments. 😁
What about civitai? As far as I know, their surface or basic structure is exactly the same as automatic1111. You can use it completely free of charge, unless you want to use special functions, which is more about training AI. Perhaps you can tell us more?
What about civitai? As far as I know, their surface or basic structure is exactly the same as automa
Oh, right. Haven't tried CivitAI's generator yet. All I remember is people complaining that its queue takes forever. 😁 Its contents restrictions are cubs and scat, I believe?
Hm. The horde is a thing too...
Oh, right. Haven't tried CivitAI's generator yet. All I remember is people complaining that its queu
If you use it privately, you can actually do anything. You just can't publish it. As far as I know... I could be wrong. In any case, Cub is not forbidden, because you can even use it in the generator. In fact, almost all known models.
Just be careful not to make the mistake of making it public. Fortunately, this is strictly separate.
If you use it privately, you can actually do anything. You just can't publish it. As far as I know..
Tried CivitAI's image generator. It's, uh... complete and utter trash? 😆 The maximum resolution is 768x512, no upscaling, no variants, full syntax is undocumented, embedded metadata doesn't include model names, the list goes on and on. No wonder it's free. 😆
Sadly, I can't recommend this tool. Not only it's bad, but it also doesn't satisfy any of ACP requirements.
The only real bummer to me was finding out how much electricity it uses to gen your own images. I was doing a lot of prompt and lora testing, but had no real idea the cost till my month's power bill went up 70 bucks! I've had to tone it back and instead try to plan images using sketches with controlnet and be frugal, no more wildcard randomness just to see what it makes. And some might not be aware, but SDXL uses up to 4x more electricity per image instead of normal 1.5 SD. So it's just something to keep in mind, even home gens are not really "free"
The only real bummer to me was finding out how much electricity it uses to gen your own images. I wa
Huh, how much and with what GPU did you work? o.o Let's assume 333W of power draw which is really high. That gets you 3h per Kwh which is normally between 15 and 45ct depending on your country.
Huh, how much and with what GPU did you work? o.o Let's assume 333W of power draw which is really hi
I have a evga geforce 970, when I use SD it's set to the lowvram and xformers on, it takes quite a long time to make a picture usually between 30 seconds and 1:45 minutes depending on the size I chose and if I decided to use any other small features with it such as controlnet. Let's just say that when I use the program I have a little lamp on my desk, it actually causes it to flicker a very tiny but noticable amount as the GPU thinks.
And the really crazy part, is that my house has solar panels and batteries, it actually ate up all the stored solar credit first from over the summer and then the 70 was additional. So I have no real idea overall how much it was using, just that it was quite a lot.
I have a evga geforce 970, when I use SD it's set to the lowvram and xformers on, it takes quite a l
Hmm, that should be below 300 watt. Maybe you should get an "electric meter" to check the power draw! Do you also have air conditioning in your house/room? Since that might have needed to cool down the 300W of warmth spilled into the room.
Hmm, that should be below 300 watt. Maybe you should get an "electric meter" to check the power draw
Do it more in wintertime if the power bill is a concern. Most of the power used by a computer turns into waste heat, so you're basically using a space heater to make what you want and using that power twice, and you can turn the thermostat down a bit to compensate. XD
Do it more in wintertime if the power bill is a concern. Most of the power used by a computer turns
Well, I've just learned to be a bit more careful about what I make, and try to make very good images. Planned out, with controlnet and so on. Only a few tests here and there, I used to test Loras and prompts, and the new controlnet features on end for quite a lot of time. I just try to make others aware that if you do generate a lot, for many hours, overnight, very high res, etc. that it can be quite significant. And that SDXL, does actually use around four times more power per image.
When careful, everything is fine and likely not much different then just playing a game for a while. But if you go really overboard, training or creating for hours a day it does show. Not hating on it in any way, it's more just an awareness thing that it's not entirely free at home on your own pc, that's all ^^ All the personal home benefits, likely do fully justify the cost too.
Well, I've just learned to be a bit more careful about what I make, and try to make very good images
I think people got SD running even on ancient GPUs with 4GB VRAM. It may be rather slow and painful, but definitely not impossible. 😆 And if you have recent hardware capable of running modern games, running SD should be no issue at all.
I think people got SD running even on ancient GPUs with 4GB VRAM. It may be rather slow and painful,
16Gb and it's more than enough. My bottleneck is vram. Very limited picture size, long wait time, control net adds like 15min so I don't use it at all, and of course no fancy sdxl models either.
16Gb and it's more than enough. My bottleneck is vram. Very limited picture size, long wait time, co
The problem is that "just running SD on PC" isn't all that accessible due to many reasons ranging from an old and weak GPU to not having a PC in the first place. So we do need online services to let people "try before buying". 😁
The problem is that "just running SD on PC" isn't all that accessible due to many reasons ranging fr
It's most flexible to generate using your own hardware.
Even something as old as a GTX 1080 will work although slowly. If buying, I'd go for something newer than that, and a 12 GB VRAM model if possible for flexibility.
You can also use GPU instances. You must be very careful with those, as if you forget them they can cost you a lot of money.
It's most flexible to generate using your own hardware. Even something as old as a GTX 1080 will wo
Flexibility depends. If one is stuck with 8GB RAM and GTX 1060, then using all of the newer tech like SD XL and ControlNet becomes a royal pain in ass, even if technically possible. A lot of people don't even own a PC nowadays. Overall, we do need some alternatives to local installations. 😁
Flexibility depends. If one is stuck with 8GB RAM and GTX 1060, then using all of the newer tech lik
I found https://frosting.ai/. It is free and dose not allow underage material. However there is a work around. It is not the greatest and will definitely not comply with InkBunny's rules. You can sign up, but not much in doing so. The only up side is you can delete all of what you generated instead of deleting everything one at a time with out signing up. And there is a cap for how many words you can put in. There is unlimited prompts but you got to pay. I haven't paid yet for all the stuff to get better pics generated, I like it for my personal pics that I will never post. Got some good ones and a ton of bad ones. You can get some R34 characters generated like Renamon, Muk, Eevee, Duchess from Aristacats, Torial, and more. Getting Ranbow Dash to propery generate is a frigen nightmare,
I found https://frosting.ai/. It is free and dose not allow underage material. However there is a wo
I forgot to mention. You can also add another picture for a starting point. And add various artists too. What ever you do not add in the prompt: teen, underage, child, school, young or anything like that. And in negative prompts do not add: Amateur or adult either.
I forgot to mention. You can also add another picture for a starting point. And add various artists
I used frosting.ai but still got all my posts on here banned so it may not be an allowed program. Sucks because it generated some great pic with a little work on my part by correcting what it screwed up and using the reworked image as a baseline.
I used frosting.ai but still got all my posts on here banned so it may not be an allowed program. Su
Yeah, I don't know why so many services decided to block generation of porn. It doesn't make any sense. Porn is the reason so many amateurs are interested in AI.
What people designing SFW-only services expected? That their users would be generating stock photos 24/7? 😆
Yeah, I don't know why so many services decided to block generation of porn. It doesn't make any sen
Is there any information on how exactly PerChance is running their models? To post an image on InkBunny, it's necessary for the image to be reproducible with open-source tools as a bare minimum (ideally with the tool itself being open-source).
Is there any information on how exactly PerChance is running their models? To post an image on InkB
Requests for AI art are fun and all, but I'd rather see people generating images themselves. 😁 Also few AI directors will bother with replicating exact OC characters and stuff like this, as that's a lot of work and commissions aren't allowed.
Requests for AI art are fun and all, but I'd rather see people generating images themselves. 😁 Also
Well, nobody is stopping people from tipping last I knew. Then, I'm still not sure how many people even today know good workflows for evoking OCs, and even the ones I've seen do a decent job of it have been very quiet.
It is sad that there's no straightforward mechanism to sell time and services rather than the art itself. As you said, it can be really time consuming evoking an OC, but it's worth the effort, I think.
Well, nobody is stopping people from tipping last I knew. Then, I'm still not sure how many people e
Yeah, PerChance is problematic. It may be possible to reproduce at least the easy prompts, but I don't think it's worth the trouble.
You can try running the same prompts in one of StableHorde apps. The results should be relatively similar to what you got in PerChance. Just make sure you preserve seeds and everything else. 😉
Yeah, PerChance is problematic. It may be possible to reproduce at least the easy prompts, but I don
the fact I need to hold onto so much just to share some cool stuff I genned just kinda makes me wanna just stop. I dunno just feels like theres more things were not supposed to us, or do then things we can
the fact I need to hold onto so much just to share some cool stuff I genned just kinda makes me wann
With the way InkBunny rules are written, administration is essentially heavily encouraging local installations of open-source tools (A1111 WebUI, ComfyUI, that sort of thing). Unless one is doing something weird, these local applications do exactly what the rules want, so all you need to do in simple cases is just copy-paste one block of text from PNG metadata, which takes 10 seconds.
There're several workarounds when one doesn't have a beefy GPU (Google Colab, where SD is currently disallowed, and renting GPUs in general), but among online services, A1111-based Diffus and crowd-sourced StableHorde just randomly happen to somehow fit all the requirements. 😆
Oh, it just occurred to me. There're some people in the furry community providing public A1111 WebUI instances. Advertising them in public is probably a bad idea, but feel free to message me if you're interested in trying out that route. 😁
With the way InkBunny rules are written, administration is essentially heavily encouraging local ins
I'll be honest I understood very little about some of those things like tha last bit... my big thing is I'm just someone who was makign stuff to share with a few friends who said I should share it, and apparently other people liked it but not sure if it's worth the hassle especially if in the not to distant future someone decides I'm a problem because they dislike AI art... just don't have the mental space for that.
I'll be honest I understood very little about some of those things like tha last bit... my big thing
I have my own curiosity, might be tempted to join myself since I have been exclusively using that model lately, not sure I'll share too much with them tho since they purposely filter out foalcon images training and I still defy them to create that content lmao
I have my own curiosity, might be tempted to join myself since I have been exclusively using that mo
Yeah, they went the high moral ground route, sadly. First furry models were like this too, but eventually people started dropping any sort of content filtering of e621 datasets. 😁
Yeah, they went the high moral ground route, sadly. First furry models were like this too, but event
