Tantabus
Filters (Default) SettingsRegisterLogin
Rules Update - For a short summary on changes, check Here
Welcome! Please check this thread

Model and Lora (and related) Discussion

Creative Corner » Model and Lora (and related) Discussion Search Posts
Started by Lord Waite
Posted
20 replies
Subscribe
Lord Waite

Given this site is specifically for AI art, I feel like we should have a few threads for discussing things like prompts, and the models we use and such, and I felt models and loras deserved one (and embeddings, but it doesn’t seem like those get used as much any more).
So, what models and loras do you find yourself using often?
The obvious, of course, is Pony Diffusion v6 XL, and while there are a million spinoffs, I’m still highly likely to use the original.
https://civitai.com/models/257749/pony-diffusion-v6-xl
I’ve also been finding myself liking BlueFoxMix lately, and the style it gives pictures.
https://civitai.com/models/724196
Lora-wise, I like the Pony Quality lora for making the score tags a little less necessary.
https://civitai.com/models/333358/ponydiffusion-quality-slider
I also do rather like the Wholesome MLP Mix lora for style.
https://civitai.com/models/509312/wholesome-mlp-mix-pony-v6-xl
Of course, there are other, non-pony models I use. They may not be trained on ponies, but Flux and SD 3.5 Large can do text well. Flux is pretty good at actually following your prompt, and I think 3.5L edges it out style-size.
SeaArt Furry XL is a good general furry model, though it’s definitely one of these ones that relies on artist tags too much.
https://civitai.com/models/391781
Illustrious XL has promise and has already become a new base, though not great for ponies.
https://civitai.com/models/795765
NoobAI-XL is also looking good. I do need to explore these more.
https://civitai.com/models/833294/noobai-xl-nai-xl
But what loras and models do you like, and what things do you see that you like and dislike about them? (And, honestly, feel free to go into more detail on some of the ones I mentioned. I just wanted to throw out a few of the better ones to start things off…)
Posted Report
Zerowinger

3-3/4" Army Man Fan
My go to has been Google Imagen 3.0
https://deepmind.google/technologies/imagen-3/
It’s a text to image model similar to Bing and DALL-E. Rather than using Gemini, I’ve been making images via the NightCafe creator studio. This checkpoint is very good at making ponies and anthros, and has a pretty solid idea of what the majority of characters are supposed to look like.
There are some issues with secondary characters, however, but there is an odd quirk about it: if for, example, you ask for an image of Big Macintosh, it will give you a pony that looks nothing like Mac. However, if you ask for an image of Big Macintosh, and then describe what he looks like, it will give you an image that’s pretty on point.
The checkpoint is also kind of weak when it comes to non-pony characters or creatures, such as Diamond Dogs, Buffalo, Griffons, and Yaks, and basically requires a bit more prompt tweaking to get them to come out close to the MLP style. However, it does have a good handle on Zebras and Dragons, and is able to fairly properly output images of those in the MLP style.
One other issue it has with creating anthros and some ponies is that sometimes simply describing their race (ie unicorn, pegasus, etc) isn’t enough, and will have a tendency to produce unicorns without horns or pegasi without wings, so you sometimes have to add in an additional prompt with things such as “glowing horn” or “spread wings” or some other descriptor of unicorn or pegasi elements to get it to stick properly.
Posted Report
Edited
mp40

I heard that Illustrious XL is what ponyv7 is going to be on is that true?
Also shoutou5 to autismmix/confetti and pinkiepiemix
Posted Report
Lord Waite

I’ve got a tendency not to use AI art generators that I can’t run locally, but I do remember Bing being good when I tried it previously.
As far as Pony v7 goes, that’s being trained on a different model, AuraFlow.
https://civitai.com/articles/6309/towards-pony-diffusion-v7-going-with-the-flow
There’s an interview from a few weeks ago here:
https://www.youtube.com/watch?v=8pw1LwRUGY4
Illustrious XL is more of a new XL based anime base model competing with pony. If you are generating anime/furry pictures, it might be interesting, and it seems to be big, but from what I remember, it didn’t do that well with ponies when I tried it, though I should give it another go. (SeaArt Furry XL can do ponies, IIRC.)
Posted Report
Lord Waite

No problem on the clarification.
I’ll have to try out ZoinksNoob. Of course, I also have to remember to do stuff with autismmix. I download a million models then forget to do anything with them…
I do remember EasyFluff. Haven’t used it in a long time.
Posted Report
tyto4tme4l

Something of an artist
@Lord Waite
Please keep in mind that for ZoinksNoob and other Noob-related models the prompting is a bit different than in Pony-based models. I use (masterpiece, best quality:1.2), absurdres, highres in the positive prompt and (normal quality, worst quality:1.2) in the negative with CFG 1.5-2.0 and Euler A as a sampler.
Posted Report
Siber

@tyto4tme4l
Technically, there’s a new version available, but it’s based on v-pred NoobAI, which is still in training. If I understood correctly, the author plans to remake it once NoobAI finishes training.
I only did some cursory testing on it and, ngl, it felt undercooked, whereas Zoinks just clicked with me instantly.
Posted Report
Mystic_Taurus13

@mp40
I heard that Illustrious XL is what ponyv7 is going to be on is that true?
Also shoutou5 to autismmix/confetti and pinkiepiemix
Pony 7 is in training right now using Aura Flow.
Posted Report
Lord Waite

@tyto4tme4l
Yeah, I’d read how to prompt Illustrious, and if it’s based on that, I’d expect it to use more danbooru tag style prompting. As awful as that picture I posted was, the prompt was:
perfect quality, best quality, absolutely eye-catching, ambient occlusion, raytracing, masterpiece, fluttershy, pegasus pony, my little pony, sitting in a meadow, masterpiece, safe, furry, yellow fur, fur tufts, god rays, dynamic pose, detailed background, full body, 
Negative: lowres, worst quality, bad quality, bad anatomy, sketch, jpeg artifacts, signature, watermark, artist name, old, oldest
Mostly me throwing in things from this article, admittedly:
https://civitai.com/articles/8380/tips-for-illustrious-xl-prompting-updates
In fact, I just ran the same prompt through ZoinksNoob, and it generated some fairly nice looking Fluttershy pictures.
Posted Report
AIPonyAnon

@tyto4tme4l
SD 1.5 based EasyFluff is probably still the best model for very painterly styles, probably because it has a very good understanding of many traditional artists. However, being 1.5 based it’s a lot worse with anatomy and coherence in general, though significantly better than most 1.5 based models.
Posted Report
AIPonyAnon

I tried out NoobAI-XL v-pred 1.0 yesterday, https://huggingface.co/Laxhar/noobai-XL-Vpred-1.0:

Wildcarded through a fair few combinations of artist styles when generating this image. Judging from that, it seems to have a better understanding of style and texture than the eps-pred version. It also seems to be slightly worse at fine detail, but that may just be a product of its better understanding of style. It needs further testing with artist tags that do highly detailed works to see if that is a limitation of the model.
Posted Report
Edited
truekry

Not an Artist
I’m actually in a hate/love relationship with Pony (SD XL) in general. In my honest opinion, SD 1.5 was so much better and generating detailed backgrounds. That’s why I, for quite a while actually, used both (like here). Back then I used Aniverse/Anitoon a lot in various versions and just used Pony to do the Characters.
But since it was a huge hassle and my pc crashed 50% of the time switching models one the go, I turned to PonyV6 full time. I used that an Autism for a while, got stuck on T-Ponynai3. It has this light “dirtyness” of left in sketch lines. And it can produce so much more detailed background than PonyV6 in my opinion. Here is an example. The light roughness makes it look a little more hand drawn in my opinion. Pics not done, so that is that.
I only use Loras for human chars. Since most people generated are either muscled dudes or near anorexic big boobed girls, I need them to stray away from that. For ponies I use non at all.
Edit Preview
Syntax quick reference: **bold** *italic* ||hide text|| `code` __underline__ ~~strike~~ ^sup^ ~sub~

Detailed syntax guide