>>1040809 Nice.

>>1040471 >>1039405 >>1039792 >>1039393 Different poses, but very embarrassed. 1st attempt I think the hair wasn't quite right, 2nd attempt was too ambitious with multiple prompts so it took ages to fix things. 1940s Illustration mixed with Alkemanubis for the style,

I feel bad asking this anywhere, but I want to get started learning this stuff, and I have no idea where to start. Everywhere I turn, people seem to already know basics that I do not comprehend. What websites do I go to? What programs do I use? What do various terms mean? What are the free options and how badly do they suck compared to paid options? What are techniques that you guys consider important? All I know is how to search "free AI image generation," click on websites until I find one that is actually free, then type in words and be disappointed with results. >tl;dr: spoonfeed me newfag info. I've clicked on the resources pastebin and it's not spoonfeedy enough. A complete retard's guide to this would likely be something good to put in the OP. You might get more activity that way. I'd join in, at least.

>>1041006 The good online ones are usually dead or a pain in the ass to use, I still weep for nemusona

>>1041006 >programs At this point you want PDXL ("Pony"), and/or ILXL ("Illustrious") for most purposes, with Flux being a more GPU intense option that has some advantages (acceptable text handling, ability to pull off some interesting things with natural commands) at cost of extreme disadvantages (awful license, much more resource expensive to run, poor with non-realistic styles without being taught to copy a specific style). SD1.5 is lighterweight and has a more open license as well as a lot of plugins made for it, but it has some very frustrating limitations and is outdated for most things. If you want the absolute basics of "What is all this?" level, Civitai's intro video is decent but a bit outdated in software/model recommendations (despite not even being a year old. This field moves fast) https://youtu.be/IIy3YwsXtTE?list=PLlVSBOQWxoKntlHmA3eLvODWb7uCttbfK Civitai also has a free generator with customizable settings (for web) and relatively loose restrictions (essentially no real people, no loli/shota, no bestiality, no guro). The main catch is the "buzz" use, which is really just site credit. You get ~220 a day(rollover at midnight GMT) from doing basic site interactions (takes just a couple minutes) and need 3 to generate most images (just set the images returned to 1 and tip to 0). You can get more by browsing without blocking ads, having users interact with stuff you've posted on site (as an established user, I get maybe 150 a day from this on a good day) or (outside the scope of free) buying it at a rate of 10 USD for 10,000 buzz.

>>1041145 I prefer running local to avoid censor limitations & pay-pigging, but there are shortcomings. Firstly, it's harder to make the installation of everything required idiot-proof. Even supposedly simple one-click installers can wind up with an incompatible version of Torch or something that you'd have to remove & swap manually. I've had plenty of issues where something that should work doesn't & I have to either reinstall everything or search everywhere for a solution that requires more effort & most often doesn't solve the issue. This leads to paranoia over updates so I'd get left behind by staying on an old install that I had stable & working. There's also people who simply can't learn or require explanation on a level that is like trying to navigate a boomer through tech troubleshooting over the phone. Nightmarish. One guy early on became a meme for not comprehending things that were spelled out in many ways, to the point where a video was made to spoonfeed such people. Outdated now as most things, but the process is still similar & at least demonstrates a "git clone". I'd recommend Forge over Automatic1111 at this point as Forge is much faster. Forge can't use Regional Prompter though, it has to use Forge Couple (if you want to do multiple prompts for multiple characters, otherwise this doesn't matter). Even if you're not being nickel & dimed by some online AI site so they can comfortably over-fund their upkeep, there's still a cost with local AI since you're best off with an expensive high-end GPU. Sure you can use mid-lvl hardware, but you're gonna be waiting a while or else making smaller images to compensate. Regardless, a high-end GPU can do far more than AI & you actually OWN the hardware, for many people owning the hardware & not being subject to any censorship is non-negotiable. I'm likely stubborn enough that if my GPU failed I would not be using the online stuff & instead save that $$ for another GPU.

in the mood for some citrus!

>>1041231 >SD1.4 >2022 dates in explorer That's even more outdated.

>>1041271 Post this shit on Facebook claiming to be a young African artist and you'd get 50000 comments all saying "amen!! amen 🙏"

>>1041231 >Sure you can use mid-lvl hardware, but you're gonna be waiting a while or else making smaller images to compensate. By "smaller images," do you mean in terms of resolution? Because that might not matter for what I have in mind. Basically what I want to learn to do is do basic sketches (though I'm not good, I could at least do layouts or whatever), then have AI improve it so it looks like a reasonable drawing, then I can manually touch up the AI result, even tracing it fully if needed. I even figure if doing multiple characters at once is hard, I can do them one at a time, then edit them together in GIMP or whatever. I can use AI to make sketches better, right? I'm more interested in that than I am in text to image generation. But when I look up tools, they all want money, or at least accounts, and I really don't know where to turn. I don't want to shell out money or even make an account when I'm just blind guessing at what is actually going to be a good tool to use.

>>1041362 >what I want to learn to do is do basic sketches (though I'm not good, I could at least do layouts or whatever), then have AI improve it so it looks like a reasonable drawing What you want image to image or some specific stuff (this isn't an area I know well, but do know it is possible). There are local solutions, for this, but I don't know enough to guide you through it >I can use AI to make sketches better, right? Better is subjective. Be prepared to curate and correct a lot of "AIness" when doing this.

https://civitai.com/models/963390/atsuko-inspector-gadget?modelVersionId=1078606 Turns out nobody has made a LoRA for Atsuko (from Inspector Gadget) yet, so I made her. >>1039392 Atsuko holding hands with Penny.

>>1041231 >Sure you can use mid-lvl hardware, but you're gonna be waiting a while or else making smaller images to compensate >you're best off with an expensive high-end GPU. The RX 7600/3060/Arc 580 is good enough.

I'm gonna share a few BA gens that came out kind of nice.

>>>/h/5887 Fuzzy breast milk.

Can someone give me a quick rundown on how to train a LoRa on specific artists/poses/characters? I really want to make some for very niche stuff but I have no clue where to start and what are the best practices/fastest routes.

>>1041571 >3rd and 4th I wonder what the AI is trying to say here.

>>1041586 https://github.com/derrian-distro/LoRA_Easy_Training_Scripts?tab=readme-ov-file#installation I got as far as downloading this. I think anon explained how pretty well in the previous thread.

>>1041586 use the civitai trainer. you have to create an account on that site and then you may need to pay to have enough "credits" to train a lora but it's definitely the best alternative to actually buy the hardware to do it locally, also it's probably the cheapest way

>>1041586 https://civitai.com/articles/8370/ Ultimately comes down to four things you need to do >gather a training data set >use basic image manipulation techniques to "clean" training data to remove other characters/text if needed/possible >tag training data >set the settings on the trainer and hit run Poses are a bit different. You can train a pose like any other LoRA (and is likely better if you want some degree of style bleed, like for a pose from a meme), but for local gen there's also openpose, which lets you arbitrarily pose your results by moving wireframes (which can be saved and shared). Using openpose means you don't actually have to train to get a pose. >>1041639 You can easily get enough credits to train a LoRA just the daily tasks for three days. That's less than it sounds since it rolls over all at once at midnight GMT and you only need a few minutes to do the tasks (so do tasks right before changeover, wait till changeover, and do tasks again). I think you get a one time bonus for joining and that's enough to get it with just 2 days of credits. You can do it even faster if you browse with ads off.

>>1041651 but I want to train a lora for FLUX since that seems to be the best at realism and real people and FLUX training seems to cost a lot of credits, a lot more than pony and 1.5 at least.

>>1041639 Carizard.

>>1041271 >>1039392 >orange megaman I request the Yellow Devil but in a giant lemon form in that similar style.

>>1041671 >I want to train a lora for FLUX 17614f didn't specify base model >since that seems to be the best at realism and real people It's more that Flux can only really do realistic without having a style LoRA attached. In my experience it's impossible to get the thing to have any kind of consistent art style or consistently draw the same article of clothing in a particular way without a LoRA. (Honestly, I don't understand the point of making stuff with real people outside of shitpost/memes about public figures.)

>>1041680 >Stuff with real people Honestly, I can only think of porn, aside from the memes

>>1041712 Yeah, but who cares about seeing famous real people naked? They're either underage, already taken their clothes off, or are hags. Why bother when superior 2D girls exist?

>>1041717 Yeah I agree, most actresses get their nudes "leaked" by the age of 20 nowadays so there isn't really a point to actresses LoRAs. But what if the real person isn't famous, what if the LoRA 17614f wants to make is someone they know IRL?

New video game girl LoRA I made. https://civitai.com/models/960286?modelVersionId=1075145 Training data wasn't the greatest, so details are off (ears aren't quite right, facial mark isn't right at all) but overall functional.

>>1041785 >shantae based on Apostles style Very nice.

CAT - Citron Pony Styles - Realistic Anime: https://civitai.com/models/362745?modelVersionId=420176 Autumn Dark Academia: https://civitai.com/models/580598 Shadows Lighting - Concept: https://civitai.com/models/675108 I don't think these LoRA style Lain quite the same as other characters. Lain's eyes look cool regardless.