do you know any good loras for this?

tried to play around with this a bit, didn't have amazing results

>>3185 Sadly no. As far as I know, no one has made one. Which is partially why I made the thread; my thinking was that I can't be the only one who has tried to do this stuff. I would try to make a lora, but I'm not that knowledgeable in how to make good ones.

>>3197 as far as I know, the main issue is going to be teaching a model the body shape? Which is probably going to be hard without a lora or a model or something similar. If I knew how to make good loras or the like, I'd probably try it myself, but I've never had good results.

There's an artist called HGSwells over on DA that does really good berry content. He has some tutorials but a lot of it boils down to "just do it in photoshop." https://www.deviantart.com/hgswells9000/gallery

>>5800 That's a fairly good shape, tbh. I'm curious how that was generated, and whether it required lora training or the like.

>>5810 Yeah, some of my pictures in my op were made with similar methods to his. I think his process of doing it is still the best, though he has a greater focus on 'realistic' stuff.

reupping these two that i thought fit

>>6666 I agree! I really think that some of us really need to find a way to train like, a round body LORA or the like. That would certainly be helpful I think, to generating different body sizes. I've been trying to train one, but I've not had any luck getting good results yet.

getting better results with a few tweaks

>>6753 I agree with you this pic's so much better you might be onto something Here

>>6704 Fats can already get pretty round. It might be hard to remove the limbs to turn them into berries.

>>6754 Thanks, made another blueberry girl recently. pretty happy with the results

>>6888 I'm very interested what your settings are, and what your prompt is. Because you're right, these results ARE pretty good. I'm still trying to train a blueberry lora, but the big problem is A. different shapes and B. it needs a lot of manual tagging, which takes a long time.

>>7042 Thanks! Glad you're enjoying my generations, I post most of them over here: https://www.deviantart.com/stuffeddeluxe/gallery I've been using images from artists and then running them through a Lora. The first one was based off the johnny swell blueberrgirl comic

We could use a Blueberry character.aI here…

>>7072 out of curiosity, what LORA? one you trained yourself or a specific other one? or do you mean taking inflation/blueberry art and then running it through img2img using a lora? Because that I've had less luck with simply with proportion recognition.

>>7768 It was LORA I made myself based on some images I ran through a LORA Generator. I use this one for reference: https://colab.research.google.com/github/hollowstrawberry/kohya-colab/blob/main/Dataset_Maker.ipynb

i'll give it a try

>>7773 I'd be curious to see results. Let me know how it goes!

👍

>>7900 alright ! Blueberry lora, made from 1.6k images from r34 & e621 Works at strength 1, the important tags are "blueberry inflation, spherical inflation, full body inflation, immobile, blue skin, transformation, lactation, juice, bodily fluids, fat, obese". Full tag list in the MEGA Use 'duo' instead of '2girls' and for negatives, you will want your normal negatives, plus "(multiple views:1.2), speech bubble, (tail, snout, furry, animal ears:1.3), colored_pencil_\(medium\), crayon, sketch, marker_\(medium\), flat color, sweatpants" The furry negatives are especially important, it has a furry bias because of the ~1k images pulled from e621. Consider leaving out "tail, animal ears" if you want to prompt animal girls Proportion control is very, very difficult, and it's honestly RNG as to whether or not you'll get the body you want, regardless of what your positives and negatives are. you can thank r34 & e621's HORRENDOUS tagging for that one. For v2 I'll review the tags and do some pruning Does better on bbw models than on normal models. Does not play well with artist loras in the slightest. Works better with non-latent upscalers on a denoise of .4-.5 https://mega.nz/folder/KdkgQIYA#ffmZL4HZROotnYIT4ZFBWA

>>7900 Who’s the first image of?

>>7900 >>7908 Holy shit, these turned out superb! Bumping this thread so more people can see the results and lora, and hopefully more art from this in the future.

Trying the lora to revise some of my old crusty NovelAI gens.

>>7908 I enjoy the huge blueberries this LoRA made at low LoRA strength, lol

>>7908 wow good work! granted, the issues that I've found are the ones you articulated: that it's far too preferential to furry stuff, and sizes are all over the place. Though I think a big issue is that it doesn't really understand positioning, such as front or above. I browsed r34 just to see what kind of images were there and... yeesh. It's not just the tags that are bad. Virtually all the stuff we would consider 'good' art are absent from there. I was already trying to construct a dataset that was closer to around 160-200 images, and basically none of them are in the dataset you used by my estimation. The real problem is that tagging things manually takes an age and a half, and there's no good way to easily tag things manually other than typing it all out one at a time by my estimation. Still, the fact that you've gotten half decent results on such a bad dataset means it can be done, and I'll look forward to seeing the v2 version. Also, with the 4060tis rolling out in a few weeks, hopefully I can actually try training something as well. Or at least, try to figure out some kind of tagging method that works.

>>7908 Thanks for this! Here's my fav so far.

>>8183 The only solution for tagging in large numbers is to train your own image classifiers. But that's probably not worth the time for datasets < 1k images.

>>7770 >>7900 >>7908 Going to bump this thread. Don't want the compilation work this person put in to go to waste on the backpages.

>>9321 These people* I'm high

>>7908 Neat

>>9325 Holy shit that's good Great job! Any more?

>>9375 No one tell Bao about these

Bumping

My contribution to keep this thread rolling

>>10221 So kind! Thank you haha

Keepin things going around here. It's been quiet.

>>10810 To be fair, it's pretty niche even within the community. And also apparently rreeaallllyy difficult to get gens of fuller stages. But new gens are always welcome! Fantastic as always

>>7958 Does anyone have the pics from the previous thread in the inflation board that had a similar style to these?

>>12267 You might be looking for my revised novel ai gens. They're mostly not blueberry though. Here's the whole lot if you want em. https://files.catbox.moe/fievgz.zip

Fully round blueberries are hard to do. Though I can can get a pretty good inflated bodysize without to much work. So here's another!

>>12373 It's a nice coincidence this thread got bumped recently, I've just been working on getting some nice round blueberries. These ladies are starting to look pretty ripe and juicy.

>>12273 Thanks for sharing! The pics are like the blueberry ones in the files you sent, I think one of the pics made it look like the girl was biting her lip? Does anyone know how to find a thread that isn't in the archive?

>>12376 Oh boy these show some promise.

So I've been playing around with the blueberry LORA on civitai, and I discovered two things about it. First, it's very temperamental when it comes to what model you use as the base. There are some that work great and some that work not at all. The second thing is that, for whatever reason, the LORA practically seems to REQUIRE that you generate huge images as the base; these have been upscaled, but the base images I was generating were as large as 800x800 or larger for getting decent results. I don't know why this is. I should also say that while it does very decent partial blueberries, getting it to do more spherical things is something I've not really managed yet. But getting consistent results for partial blueberries like this IS possible, albeit with some inpainting.

>>12890 Author of that LoRA here. It was trained at 768x768, so I'm not necessarily surprised that it works better on higher base image sizes. As for being temperamental on certain models? Probably a side-effect of using NAI as a base model for training. The dataset it was trained on was fairly limited. I would like to refine it further but it competes with the general BBW LoRA for my time. I'm pleased you were able to get good results sometimes, at least. I often combine it with my architecture LoRA (SyMix), where that LoRA is at lower strength, in order to refine the output.

>>12376 Those are damn good. What's the prompt?

>>12890 Have you got any reccomendations for models? I've been getting terrible results

Made a Heather Mason sequence

>>16644 These look fantastic! Well done! I know berries are a bit niche, and they are hard to generate, but man this board as a serious lack of them. Makes my dick happy every time I see this thread updated.

>>16644 Mooooaaaar

>>16672 >>16793 Sure thing. I'll try posting whenever I've made something worthwhile, but 99% of character LoRAs have been fighting tooth and nail against anything spherical. Here's some Annie for the time being

>>16991 And some Ashley

>>16991 Out of curiosity, are you getting these from a LORA you trained yourself or one that already exists, because I've not been able to get similar shapes. I'd actually be happy even if they weren't entirely spherical, because I consider the gaol for now to be able to get the AI to understand how proportions for the output work.

>>16993 The blueberry LoRA I'm using is the one provided by >>7908 Arguably more important is the actual model itself. I'm using the same biggerGirlsv2/abyssOrange checkpoint merge as is recommended in the pinned guide thread. Generates way more diverse proportions, and is more sensitive to expansion-related tags. You're more likely to get the shape you want without needing to brute force it with a billion tags, which leads to a less overcooked final image. I also find that the weight slider LoRA gives you a decent amount of control over the average size of your generations https://civitai.com/models/112552/weight-slider-lora

Some Chel

>>17011 >>17013 Awesome

Junker queen

Some blonde broad

Some of the better stuff I've generated

>>17178 These are class! What's your settings for this if you don't mind posting it

>>17432 For the realistic ones: young BBW with beautiful slim face, ((detailed surprised face)), light hair, curvy, standing up, (((completely nude))), (bare breasts), <lora:blueberry_inflation_v1-2-000060:0.4>,(hyper_blueberry_inflation), ((blue skin)), juice puddle, ((nipples and pussy leaking blue juice)), <lora:hyperfusion_550k_128dim-LoCon_extracted-v7:0.5>, ass visible from front, colossal breasts, topheavy, bimbo, hyperbreasts, colossal ass, colossal belly, massive thighs, hyper thighs, breasts on belly, fat belly, breasts apart Negative prompt: (Worst Quality, Low Quality:1.4), (multiple_girls:1.2),border, grayscale, loli, paintings, sketches, (worst quality:2), (low quality:2), (normal quality:2), lowres, normal quality, ((monochrome)), ((grayscale)), bad anatomy, ((abs)), ((muscles)), ((multiple navels)), (speech bubbles), facing away, looking away,tilted head,lowres,bad anatomy,bad hands, text, error, missing fingers,extra digit, fewer digits, cropped, worstquality, low quality, jpegartifacts,signature, watermark, username,blurry,bad feet,cropped,poorly drawn hands,poorly drawn face,mutation,deformed,worst quality,low quality,normal quality,jpeg artifacts,signature,watermark,extra fingers,fewer digits,extra limbs,extra arms,extra legs,malformed limbs,fused fingers,too many fingers,long neck,cross-eyed,mutated hands,polar lowres,text,error,missing fingers,missing arms,missing legs,extra digits (extra fingers, deformed hands, polydactyl:1.5) Steps: 35, Sampler: DPM++ 2M Karras, CFG scale: 9, Seed: 2492493431, Size: 512x768, Model hash: 93e6545565, Model: cyberrealistic_classicV20, Denoising strength: 0.51, Hires upscale: 2, Hires steps: 10, Hires upscaler: R-ESRGAN 4x+, Lora hashes: "blueberry_inflation_v1-2-000060: c593ce8fd0c5, hyperfusion_550k_128dim-LoCon_extracted-v7: 66d481cf230a", Version: v1.6.0-2-g4afaaf8a For the anime looking ones: pale chubby young woman with short blonde bob haircut, pageboy hairstyle, ((detailed face)) ((((shocked scared face)))), ((((one hand grabbing breast, breast grab)))), light hair, curvy, standing up, (((completely nude))), (bare breasts), waddling, legs spread apart <lora:blueberry_inflation_v1-2-000060:0.5>,(hyper_blueberry_inflation), ((blue skin)), juice puddle, ((nipples leaking blue juice)), wet pussy <lora:hyperfusion_550k_128dim-LoCon_extracted-v7:0.6>, ((ass visible from front)), ((colossal breasts)), topheavy, bimbo, (hyperbreasts), colossal ass, colossal belly, massive thighs, hyper thighs, breasts on belly, fat belly, breasts apart, (background is inside of a dark industrial factory with pipes and hoses), (dark blue dim lighting) Negative prompt: (Worst Quality, Low Quality:1.4), (multiple_girls:1.2),border, grayscale, loli, paintings, sketches, (worst quality:2), (low quality:2), (normal quality:2), lowres, normal quality, ((monochrome)), ((grayscale)), bad anatomy, ((abs)), ((muscles)), ((multiple navels)), (speech bubbles), facing away, looking away,tilted head,lowres,bad anatomy,bad hands, text, error, missing fingers,extra digit, fewer digits, cropped, worstquality, low quality, jpegartifacts,signature, watermark, username,blurry,bad feet,cropped,poorly drawn hands,poorly drawn face,mutation,deformed,worst quality,low quality,normal quality,jpeg artifacts,signature,watermark,extra fingers,fewer digits,extra limbs,extra arms,extra legs,malformed limbs,fused fingers,too many fingers,long neck,cross-eyed,mutated hands,polar lowres,text,error,missing fingers,missing arms,missing legs,extra digits (extra fingers, deformed hands, polydactyl:1.5) Steps: 33, Sampler: DPM++ 2M Karras, CFG scale: 9, Seed: 289088916, Size: 512x512, Model hash: ee70fda123, Model: bigbellybabes_v10, Denoising strength: 0.44, Hires upscale: 3, Hires steps: 20, Hires upscaler: SwinIR_4x, Lora hashes: "blueberry_inflation_v1-2-000060: c593ce8fd0c5, hyperfusion_550k_128dim-LoCon_extracted-v7: 66d481cf230a", Version: v1.6.0-2-g4afaaf8a

>>17491 Cheers v much for posting, had far better results using those settings

I'm super late to this party. I didn't even know this board existed. I've been using the blueberry inflation LoRA from Civit AI, which I just discovered like a week ago. Is the one posted in the thread >>7908 an earlier version of that? Or are those two completely separate projects? Anyway, glad I finally found where I'm supposed to be. Here are posts I made on the wrong board: >>/inf/101422 >>/inf/101423

>>17663 Dang it: >>>/inf/101422 >>>/inf/101423

Anyway, massive thanks to whoever trained that LoRA. The results have been fantastic. For those curious, the models that I have found work best with the LoRA are: > CarDos Animated 3.0 > RealCartoon3D > RealCartoon-Realistic I have had some success with the more realistic models, but some degree of 3D or drawn appearance is usually best. It's also been my experience that the LoRA is more effective when at least one side of the first generation is ~768 pixels.

Couldn't decide which face and dialog I liked better so you get both. Choose your own adventure. The left panels are RealCartoon-Realistic and the right ones (full berry) are Dreamshaper 8.

>>17765 >>18034 Cute

Merry Christmas! Sequence plus individual panels. >>18061 Thanks!

One more of the Christmas berry.

Maksim

Appalachian permaberry boyfriend

Discord groups for this?

Here is a bunch of random results from the last few days! any tips for getting that puffy look around limbs/neck? wanting to get a sort of close-up inflation with the focus being the face, swollen body filling the frame

Hey, not an art generation but I made a Blueberry chatbot as there aren't many. Enjoy! https://venus.chub.ai/characters/Chadpiss/hana-26507c66 (Warning: Lactation, Fluid Leaking)

messing around trying to get sequences and puffy faces

>>3169 So there's an update to the Blueberry Lora, though I've struggled to get it to work well. Anyone gotten good results with it?

>>19123 Having issues getting quite the same level of polish, here are two from old, and two from new, seems to really like 05' BIG shapes, but still working with it, there's also a new Fluffyrock trained blueberry one too

>>19129 *using the same prompt

Random other ones

few other club scenes with v3

>>19140 Awesome, very cool theme you've got going here. Will have to try out that v3 myself soon.

>>19171 Ah cheers v much, here's a random dump from the last few days for anyone interested

O R B S

i've not had much luck using SD to make images, been sticking with Bing's AI image creator inflation doesn't work very well with it, farthest i can get is some more realistic skin-color change, puffy cheeks, the occasional pregnant belly to simulate juice

Got lucky with Bing AI, and thought id share some

>>7958 The long haired girl in the kitchen is really good.

>>19429 these are really, really good. do you have more?

>>20463 Have a good few, but still haven't got good faces out, need to refine a bit. Honestly at a loss on what to generate, anyone with ideas/scenarios they'd like to see let me know!

>>20613 Can you share what LORA you used and what your tags are? I ask because I've had a really rough time getting the ones I know of to work well and produce good results. I'm amazed you got anything decent.

>>20619 ah no worries Baiii, All my images are just direct, so PNG info will bring back the prompts if needed, either way let me know how you get on with this one:women, chubby, fat, detailed, intricate, masterpiece, <lora:hyperfusion_550k_128dim-LoCon_extracted-v7:0.7> , detailed face, blue eyes,<lora:SnxilluminateV2:0.8> blue light, purple light,background ,<lora:blueberry_inflation_fluffyrock_v1a02:0.6>, full blueberry inflation, hyper blueberry inflation, spherical inflation,<lora:blueberry_inflation_v3a01:0.6> lactation through clothes, pussy juice, pussy juice stain, pussy juice puddle, lactation <lora:blueberry_inflation_v1-2-000060:0.6> ,puffy cheeks, puffy arms Negative prompt: (worst quality, low quality:1.4), cartoon, drawing, sketch, boring_e621_v4, By bad artist -neg, bad-hands-5 ,ng_deepnegative_v1_75t, shortstack, clear face, malformed limbs,((more than 2 legs)), cross eyed, cut off image, cropped, cropped image, lowres, bad anatomy, bad hands, text, error, missing fingers,(multiple views:1.2), speech bubble, (tail, snout, furry, animal ears:1.3), colored_pencil_\(medium\), crayon, sketch, marker_\(medium\), flat color, sweatpants,(navel),((multiple navels)), ((mirror)), ((reflection) Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 8.5, Seed: 2523549693, Size: 920x640, Model hash: 7eb8672531, Model: realcartoonRealistic_v11, Denoising strength: 0.41, Clip skip: 2, NGMS: 6.66, Hires upscale: 1.5, Hires steps: 15, Hires upscaler: R-ESRGAN 4x+

>>20621 and yes i did bring in all 3 of the current Blueberry LORAs, Throwing shit at the wall *Sometimes* makes a diamond

The background is scuffed and I wasted a lot of time trying to inpaint in extras before resorting to an image composite, but I'm pretty proud of this one

>>21011 Here's just the berry

>>21010 >>21011 These are amazing! Whatever you're doing to make these is *muah* chef's kiss. Any thoughts on different art styles, or is that not an easy task?

>>21011 Abs unreal outputs here, nicely done Orbs are perfect

>>21013 Art styles don't have much to do with it. I've been following the advice of TheGuyWhoDidAThing from a few other threads in regards to generating realism, and his guide can be found on his DeviantArt page. The tl;dr is to generate the initial shape and proportions with a more drawn/cartoonish model, then switch to a photorealism checkpoint to iteratively upscale the image, removing superfluous tags and LoRAs as you go. The only reason I've been able to generate these shapes at all is with the new v3.0 version of the blueberry LoRA on CivitAi. Every other model generates mid-inflation shapes 99.9% of the time, but this one pumps out perfect fucking orbs as long as you're using the hyper tag. Hats off to Darkforces for training it so well. Godmode. I would also endorse Eddo's advice to incorporate several blueberry LoRAs for the first generation, but I would urge you to stop using the v1.2 of Darkforces model after that since it tends to force 2D art style and produce a ton of artifacts.

>>21024 Ayy, ty for the tip on animated/realistic models for upscaling, with give it a try later and see how we go, cheers

>>21010 >>21011 >>21012 Your results are fantastic! Amazing work. >>21024 Interesting. It must be the way I've been prompting but so far I ONLY get good results from version 1.2. The keywords I use are full and spherical blueberry inflation. It is definitely true though that the results get a lot of artifacts and a drawn look when you add any sort of background. Maybe I need to take another crack at version 3. I do like partial blueberries though and the 1.2 mixes well with the Hyperfusion LoRA. Examples attached here. The rounder berries are done with DreamShaper 8 and the more realistic images are RealCartoon-Realistic v13. >iteratively upscale the image So img2img? Anyway, thanks so much for sharing your work and your insights.

You guys were right about mixing the LoRAs. I added notes on what I did with this one, all the important parts are there. Basically I put the LoRA keywords at the beginning and include the LoRAs themselves at the end. The middle of the prompt is all specific to the particular image. To me it seems like the Hyperfusion LoRA does add a touch of realism but I don't use it above 0.2 or 0.3. I find if you use it with (round breasts) or round_breasts you can get better-shaped boobs. So thank you to >>21024 for recommending v3 and mixing LoRAs. I'd just about given up on v3 but I'd been using it by itself.

>>21142 I'm trying these out and getting nice stuff back, Cheers for the tips! will keep experimenting in inpaint

>>21192 Beautiful! So glad it was helpful. It took a long time to figure out. Something I forgot to mention was that I read somewhere the BB LoRAs were trained at 768x768 pixels, so you get much better results if you use something like that.

>>21142 I'll be honest, I've not gotten ANYTHING that looks remotely good with the spherical shapes. I've tried doing the mixing of loras, but in general if you want any kind of specific thing your tags are going to be over 150, and there be monsters once you get that that point. Forget about using real style models, I've never gotten much of anything to work with drawn ones and not for lack of trying. Perhaps this is a failing when it comes to my mentality; the way I see it, if the lora can't output what it needs on its own, it's not that great. And having talked to the guy who made it, he's not particularly interested in the concept, which is sad. Had I any skill in tagging or training, I would train it on my own system; I've got the hardware for it. Though I think that in the future it probably would be best to have a lora specifically for spherical shapes; how you could go about training that, I have no idea, since you get pretty far outside the realm of what models understand at that point. The way I see it, and this might just be me, blueberries are kind of like the last frontier on trying to really get models to generate something because it's so far outside the norm on so many levels. You have an intersecting of skin color, body shape, and clothing, and there's no real world examples to train off of. So you're dealing with a very small sample size of very small amount of artists. Which means that actually getting it to work is very hard. One last bit. I'm not going to say that TheGuyWhoDidAThing's method is wrong, but it's needlessly long and convuluted for like, 90% of generations. Rather than trying to upscale over and over, just generate at that size and inpaint, and use photoshop to correct as needed. I've generated lots of things at 'reasonable' shapes that would fit into hyper that aren't bad. if you CAN generate at say, 1k square, then just do that and inpaint at that. You don't really need to upscale so many times outside of automatic's repo. I'm also not 100% sold on his idea that prompt doesn't matter. Imo, prompt is extremely important, especially negative prompt. In fact, my methods usually revolve around generating, seeing what the model seems to struggle with, and then using the negative prompt to nudge it in the right direction. Ergo, if you can't get it to move in the right direction with 'lying on back' then using things like 'on front' in the negative solves the problem usually. But I digress. I really think we need to just find someone who can train a proper blueberry lora so we don't need to try and jerry rig one together like this.

>>21531 Unrelated on the LORA point, but I've been slowly collecting screenshots from TMC videos in order to train a realistic sphere LORA, but god it is just coming out as jackson pollocks, will post results when they arent incomprehensible

>>21548 Okay so quick advice on your end. I would not try to use TMC stuff as your base for spherical. I have tried myself to train a spherical lora, and my base was using mostly things from artists like Sidkid, sickstar, ect. There are a lot of artists that work with the spherical shape, and you could in fact use the real life stuff, but I don't know that it'll give you a good base?

I recommend trying out the realistic models like Majicmix v7 and Epic Realism. Tougher to get a good result, butt the good gens are really worth it.

AniMerge 2.7 also worth trying

>>21745 I am eternally impressed on the shapes you're able to get. Spherical is definitely very hard to get where it doesn't look terrible.

>>21752 My humble thanks. Something I should mention is that I find the frame size matters a lot when trying to get spheres. 768x768 seems to make slightly tall berry, so using something like 800w by 768h tends to be the best bet for a sphere. Also, like I always say, it is a numbers game. The sphere you see above was 1 out of maybe 80. There were a few others in there that looked decent but this was the best one.

These ones are all Animerge 2.7 Been thinking about publishing techniques and even prompts (or most of prompts, the important bits like before). The only snags are: (1) it would be disappointing if, after spending time writing and publishing, nobody was willing to do all the reading and put the hours in to create images (2) I've realized that at least 50% of the work is inpainting and Photoshop. I can give guidance about inpainting but Photoshop is way beyond the scope of what I can spend time teaching. It's not something I could do for weeks but I thought I would mention it in case there actually are people who would read it.

>>22222 Checked.

going to post some random ones from the last few days

Bump

Good

>>22825 ????

>>22843 My friend are you okay, ramblings of a mad man

>>22844 Bbwchan welcomes pedophiles, rapist, killers, human trafficers. Are you ok????????

>>22844 "Dont expect professionalism here" - nerd look like mad scientist lol

Take your meds son

>>22847 Youre on bbwchan. Do yourself the favor old man.

They dont make medcine for what I got.... (Autism)

>>22849 Do autistic people have sex? How can I be autustic when I fucked more women then all of you lol

>>22850 Just my also autistic wife :) Honestly it helps, hyperfixation goes wild

>>22854 Oh that too like, whats your point big dawg?

>>22854 >>22857 Old man. Look me in the eyes I dont fear no one. Fuck u going do.

>>22858 Lol

Had some success with training a LORA (pic related, usual SD problems with hands/feet/face inpainting required). Moving to SDXL seems to have made a huge difference. If anyone here has tips for pushing training into the degenerate unknown, we should share them. Under SD1.5 it really felt like I was fighting the base models with how they viewed human anatomy. Also for those generating locally, are there any tweaks for SDXL stability? Automatic1111 is crashing on me once every 4-6 generations.

>>23738 I recommend using A1111 Forge if you aren't already. It's a branch of A1111 with a bunch of optimisations.

>>23768 Thanks for the tip, I switched over to forge and its much more stable. I really appreciate that force is smart enough to scale VRAM allocation dynamically based on image size instead of crashing on OOM errors.

Some more success I have had. Anyone experimented with regularization images for LORA training? I feel like the LORAs I have trained tend to pull in the backgrounds far more than I want.

>>24125 I'm very curious what your training settings were. You should also upload your lora to civitai so we can test it, because it seems like you're getting fairly good results. In general, the way you avoid getting backgrounds to be trained into the images is to make sure you tag for them and also make sure that you have a wide variety of them, so it understands that the backgrounds aren't what you're training.

It's been ages but here's a dump of some of the stuff I've been making lately. I've got more lying around in some folder that I still need to touch up, so I might post those later.

>>28244 These are all good, but the realistic looking ones are incredible!