/trash/ MEGA-Megacollection (WIP)
Yiffymix (yiffy e18 and Zeipher F111)
Yiffymix recommended vae: https://civitai.com/api/download/models/4053?type=VAE&format=Other
Yiffymix 2 (based on fluffyrock-576-704-832-lion-low-lr-e16-offset-noise-e1) (Use Clip Skip 1)
Yiffymix 2: https://civitai.com/api/download/models/40968?type=Model&format=SafeTensor&size=pruned&fp=fp32
YiffyMix2 Species/Artist Grid List [FluffyRock tags]: https://mega.nz/folder/UBxDgIyL#K9NJtrWTcvEQtoTl508KiA/folder/YNhymCLY
Merge of Yiffy and Anything, posts from the archives indicate that the hash from the one below vary from the usual hash? Either way, you can either download the one below or merge it yourself: Select Yiffy in box A, AnythingV3 in B, the leaked novelai anime model in C, multiplier is 1, interpolation is add difference
Explanation as stated by anon:
>Reports stated that the hash is different from what is expected
automatic1111 changed the hashes a couple weeks ago, they used to be 8 characters long ever since they were introduced, they're 10 characters long now, which are just the first 10 characters of the sha256 hash
every old link you'll find mentioning a hash will show a different hash in the webui now
7th_furry tests (seem to be merges of the 7th layer models)
7th furry testA: https://huggingface.co/syaimu/7th_furry/resolve/main/7th_furry_testA.ckpt
NovelAI Leak + VAE
Config (yaml) file, put it in your model folder next to the above: https://civitai.com/api/download/models/5927?type=Config&format=Other
Lawlas's Yiffymix 1 and 2
2 has been merged with AOM3 and other anime models, hence the described need for high weighting of furry. I personally prefer 1, but try both and see what you like more.
The mentioned embeddings are on huggingface. Easy_negatives is on CivitAI, but shouldn't need an account.
Description: Hello and welcome. This is my custom furry model mix based on yiffy-e18.
It's able to produce sfw/nsfw furry anthro artworks of different styles with consistant quality, while maintaining details on stuff like clothes, background, etc. with simpler prompts.
I personally use it with these settings:
Size: 512 x 704 or 578 x 768 (then upscale it. Some of the example images here are upscaled.)
Sampler: DPM++ 2M Karras or Euler a
(Clip skip 1 is recommended. 2 also works but makes the style different)
For better results, it's recommended to use it with [anything v4 vae file](https://huggingface.co/andite/anything-v4.0/resolve/main/anything-v4.0.vae.pt) or you can use versions with baked-in VAE.
Starting the positive prompt with "uploaded on e621" and negative prompt with"(worst quality, low quality:1.4)" help too.
PS: Consider using bad-artist embedding or boring_e621 as well. it's optional to use bad-artist embedding and you can try other textual inversions as well. If you see this, don't be an idiot like me. the way to use bad artist embedding is by adding bad-artist instead of bad_artist and you'll see it doesn't nessesarily improve the quality of the result. Boring_e621 is overall more recommended.
fp 16 baked-in vae (no vae needed, if you want to use your own do not download this one): https://civitai.com/api/download/models/15584?type=Model&format=SafeTensor&size=full&fp=fp16
fp16 no vae pruned: https://civitai.com/api/download/models/5370?type=Model&format=SafeTensor&size=full&fp=fp16
Description: Hello and welcome. This model is an upgrade from my previous model Lawlasmix. I used AOM3 and miscellaneous models during the making of the model, so don't forget to check that out.
It's capable of generating sfw/nsfw furry artworks in general with consistent quality, while having decent details in hands, clothing, background, etc. Compared with the old Lawlasmix, it's overall more anime style oriented, and doesn't need artist tags all the time (It can still change the styles if you use them anyways). It can be tricky to write prompts for the model if you're new to it, so make sure you read the tips carefully!
Personally I use it with these settings:
Batch count: 6
Clip skip: 1 or 2 (for different styles)
Size: 512 x 704 or 512 x 780 .etc (then use hiresfix, or SD upscale in img2img )
Sampler: DPM++ 2M Karras or Euler a
The model can make amazing results with very simple prompts so long as you use it right. It uses both danbooru and e621 tags, like 1boy, 1girl, solo_focus. It's recommended to start the positive prompt with "(furry art, uploaded on e621:1.4)" and negative prompt with "(worst quality, low quality:1.4)". Since the merge has significant amont of anime models in it, you need to give more weight to furry related tags. Use tags like (anthro furry:1.6) in the positive prompts and (human:1.6) in order to get it to make furry art. The model performs rather well when you specify the color of the fur. For instance, (detailed red scales:1.5), (detailed blue fluffy fur:1.5). Feel free to check out the example images for their prompts to help you write your own.
If you don't use any vae with the original version, the outputs of the model may suffer from loss of colors. As a result, consider using anythingv3 vae or orangemix vae. I provide the download of orangemix vae here. However, if you don't want to use any VAEs or can't get them to load, you can now choose the versions with a VAE baked in the model.
Feel free to use other textual inversions, but I strongly recommend boring_e621. It's trained specifically on furry artwork and works great in this case. Try EasyNegative as well
Since the has AOM3 mixed in it, it's very common for it to generate results in which characters have "M" shaped hair. Consider using tags like (unique hairstyle, unique fringe, unique bangs:1.6) or using img2img to generate alternative images to help you get to the desired results.
The model has reported to be temperamental so to speak. I suggest setting batch count to more than 2 so you have more results to choose from.
Here are models used as far as I can remember:
My previous model
I apologize for not keeping a record of the models I used. Without their amazing work, this model wouldn't have even existed. Kudos to every creator on this site!
baked-in vae: https://civitai.com/api/download/models/15460?type=Model&format=SafeTensor&size=full&fp=fp16
no vae: https://civitai.com/api/download/models/15288?type=Model&format=SafeTensor&size=full&fp=fp16
AbyssOrangeMix2 (for those without a Huggingface account)
AOM VAE (rename it the same as the AOM model you use)
Frankenmodels (Yttreia's Merges)
The filename is the recipe. Minus symbols are averaged, plus symbols are added.
Tried avoiding any models that need special VAEs.
Uh, no real comments otherwise, my Twitter is https://twitter.com/Yttreia
Crosskemono (CivitAI, links last updated: 03/28/2023)
furry_kemono.pt (Hypernetwork, should be the same one as the one from the NAI leak): https://civitai.com/api/download/models/17114?type=Model&format=PickleTensor&size=full&fp=fp16
Crosskemono 2 (with added E621 Tag support)
Same VAE as V1, see above. TEST features e621 tags, TEST_2 Booru tags, Full both types.
2.5 has added Noise Offset for brighter brights and darker darks - keep this in mind if you want to use it for any merges.
Crosskemono 2.0TEST: https://civitai.com/api/download/models/27823?type=Model&format=SafeTensor&size=full&fp=fp16
Crosskemono 2.0TEST_2: https://civitai.com/api/download/models/28447?type=Model&format=SafeTensor&size=full&fp=fp16
Crosskemono 2.0 Full: https://civitai.com/api/download/models/32830?type=Model&format=SafeTensor&size=full&fp=fp16
Crosskemono 2.5: https://civitai.com/api/download/models/47368?type=Model&format=SafeTensor&size=full&fp=fp16
If you make some cool gens with these, feel free to post them over on the Crosskemono CivitAI page and leave a rating - the author machine-translated his way onto /trash/ to ask for feedback and examples, and is bummed out he barely gets any feedback about what people think about the model over on CivitAI.
PC 98 Model
Direct DL link from CivitAI for r1 e20: https://civitai.com/api/download/models/80182?type=Model&format=SafeTensor&size=pruned&fp=fp16
Seperate Fluffusion rentry maintained by the model author (?): https://rentry.org/fluffusion
Below are links to the Prototype r10 e7 model; you probably won't need it, but it's here for posterity.
Prototype r10 e7 640x
e621 tags with post counts used in Fluffusion (could use it for looking up stuff, making wildcards or for use in Tag Autocomplete, put it in \extensions\a1111-sd-webui-tagcomplete\tags for that):
0.7(revAnimated_v11) + 0.3(fluff_proto_r10_e7_640x): https://easyupload.io/0eqrwu
Revfluff : https://pixeldrain.com/u/KJ1TKS26
0.8 (fluff_proto_r10_e7_640x) + 0.2( revAnimated_v11)
0.75 (fluff_proto_r10_e7_640x) + 0.25( kotosmix_v10)
Tism Prism (Sonic characters)
Repository (GO HERE FOR DOWNLOADS):
Tag Autocomplete CSV:
Crookedtrees (Full Model)
Use crookedtrees in your prompt
0.3(acidfur_v10) + 0.7(0.5(fluffyrock-576-704-832-960-1088-lion-low-lr-e22-offset-noise-e7) + 0.5(fluffusion_r1_e20_640x_50))
Use mmind and argon_vile in your prompt, on_back, high-angle_view, etc. Ghost_hands and disembodied_hand are hit or miss.
BB95 Furry Mix
LORAs from /hdg/
Mokochi/Neferpitou (among others) LORAs apparently taken from /h/
Ber00, Elora, Fredina, Quotefox, Ralsei Meme, Rikose, Syuro, Hariet (Mario Odyssey), Double Deck, Volcarona, Claire (The Summoning) in the style of Anonymouse
Pokemon, Kikurage, Sindoll, Tsampikos and Hizake among many others (Professor Harkness)
Midna LORA from /h/
Amad, Kishibe, Pochincoff
Binah, Felicia, Meicrackmon, Taokaka
Moths, Pokemon (Style), Mai Shiranui (DubAnon) (New Link)
Another Fredina LORA
Matuura's Fox Girl (removed from CivitAI)
Triggerword is matuurafox
Splatoon Characters, Various Pokemon (Froslass, Jirachi, Toxicroak...), Youjomodoki (style) (AmmoniaM)
Miraheze (Ankha, Lopunny, Argonians)
Pale Kemono (Amari Akitsune)
Shortstacks, Coco Bandicoot, Elora, Midna, Mirko, Priscilla, Ranni, Gwynever Dress, Goblins, Mothgirls, Yordles, PlantPenetrator, Pochincoff, tangobat, wamudraws
Blowup Doll Concept
Human art by Shinodage and Pestil (Keyword is "pestino")
Assorted HDG characters and LORAs
Guangjian Huang (@hgjart)
### About keywords: No particulars. I removed the artist name to make it implicit. It works very well with "armor" and "weapon" since most pictures include at least one of those tags. I think "greek" will also set off a couple of images.
Epoch 6: https://files.catbox.moe/asbbkh.safetensors
Epoch 12: https://files.catbox.moe/fpdliw.safetensors
Cervina 7 (readme+Epochs 1-15)
Dr. Comet (Epochs 1-5)
Udders (from /d/)
Keyword: Bubsy the Bobcat
kame_3t (Artist), Nanachi (Made In Abyss) + Various Furry Artists (among others)
Trish the Opossum
Activated with "Trish the opossum"
155 images of varying aspect with bucketing for 256 to 768. Tagging was rather haphazard and manual, not following any particular convention except to give a general idea of the content and associate it with "ocarinaoftime", "n64", "3d", etc.
>screenshot of nintendo 64 game released 1997, legend of zelda: ocarina of time, n64style, adult link, blue eyes, pointy ears, hat, blonde hair, 3d, long nose, green tunic, solo, ocarinaoftime, n64, lowpoly
Characters included: Link, Impa, Malon, Nabooru, Ruto, Zelda. Ruto may have overfitted results for "monster girl", but lowering the strength or using a character LORA fixes it. Training uses 0.1 offset noise. Currently trying to determine best checkpoint out of 250+ but it seems epoch 64 was a good one.
Captain Amelia (Treasure Planet)
No-Swift, Miles-DF, Ruaidri
I uploaded my Ruaidri and Miles-DF LoRAs alongside my No-Swift LoRA. All are trained on top of Gay621 and untested elsewhere. No-Swift and Ruaidri are activated with "by no-swift" and "by ruaidri" respectively (in line with Gay621's conventions).
Miles-DF was the first one I made and I didn't know what I was doing, so there's no activation keyword -- it still kind of works, but it's somewhat shit. I don't actually like his art style so I'm unlikely to ever remake it either.
Blaze the Cat
Stylized 3D Anime LoRA
Narse, Skygracer, Jay Nailor, Raiji, Zourik (Artists)
Use keyword "<artist>mix"
Fluff Trained: pixeldrain.com/u/XC6FgTVT
NAI Trained: mediafire.com/file/ikbsqolr8mucmot nai.zip
NAI was uploaded to mediafire as pixeldrain was failing to finish the upload for some reason
Mei_Panda (Turning Red)
Feral Horses (from /vt/ of all places?)
Responds to typical roll db tags + "rollchan".
C'mon with another gift, I threw a bunch of feral pics into the lora colab and out popped this. Feralbois v1. Trained on NAI with E621 tags. The dataset covers quite a bit though you may get mixed results on the anthro on feral positions. Its also a bit biased to felines...I was a lion king kid.
Dataset: pixeldrain com u Jc3svbHe
Fluff Trained: pixeldrain com u V6cosxNn
NAI Trained: pixeldrain com u kS25QCQY
Based off of yiffye, but seems to work with other checkpoints. Trigger word is "redxiii"
Tsukino (Monster Hunter Stories)
Waai Fu (Arknights)
My 4th try (holy fuck) at it and i wont probably try this anymore, this was done on Hollowstrawberry's lora trainer: https://colab.research.google.com/github/hollowstrawberry/kohya-colab/blob/main/Lora_Trainer.ipynb
Trained with 61 images at 768 resolution for 5 repeats and 10 epochs, used flip_aug and min_snr_ganma, 2e-4 unet/1e-4 text econder lr, batch size 2.
Getting waai fu right is a nightmare and i think part of the problem is the models not being good at it, i only tested with 7thanimev3c: https://huggingface.co/syaimu/7th_Layer/blob/main/7th_anime_v3/7th_anime_v3_C.safetensors but if someone has a merge with a furry model and it looks better on those please do tell me.
More ammount of time went into pruning this time around, i decided to swap 1girl for furry female since leaving it before was detrimental, other than that the minimum tags to prompt her are "waai fu, furry female, tiger ears, body fur", if you want her glasses add "orange-tinted eyewear, round eyewear" and for her default outfir use "default china dress, capelet, pelvic curtain, elbow gloves, orange belts", add or remove "animal nose" to positive/negative prompts depending on what you want. Additionally you can use tiger girl, tiger tail to help prompt the tail but expect random tails around and tail limbs, i honestly dunno how to fix those.
Princess Luna + Twilight Sparkle
Steve Martin Style
Probably so-so, didn't test it well enough
Digby (Animal Crossing)
Able Sisters (Animal Crossing)
Trained on SD 1.5: https://files.catbox.moe/1zno17.safetensors
Trained on NAI: https://files.catbox.moe/79u1qa.safetensors
> I suppose you use the e621 tags for each sister?
Yeah, sable able, mabel able, and labelle. I combine booru tags and autotags in my loras, it's extra pruning work but it seems to do the job.
3D Midna (trained on Fluffyrock lion low-lr e22 offset-noise e7)
Brandy Harrington, Rebecca Cunningham, Slappy Squirrel, Miranda (Wakfu), Xerneas, Berri (Conker's BFD), Lammy and Rammy and many others
Use "by nomax".
Rauru (Zelta TotK)
Trained on Fluffyrock
I tried including all possible urethral scenarios in the dataset (tentacles, rods, penis in penis, tongue, arm) but im afraid im gonna have to make one for each. The results are a lot more coherent then base fluffyrock, but still fucky.
So here it the urethral penetraion lora, use with "urethpen" tag among others, for example "penetration,urethral, urethral penetration, urethral bulge, urethralpen" will get the job done. Trained on fluffyrock lr-e27 with 350+ images (did some cleaning up), over 11k+ steps.
trigger word is pup_mask with the underscore
Beverage in Sheath
LORAs from the Discord
Various Characters (FinalEclipse's Trash Pile)
• Dawn Bellwether (Zootopia)
• Esix (e621 mascot)
• Fidget (Elysian Tail)
• Freya Crescent (Final Fantasy)
• Gadget Hackwrench (Rescue Rangers)
• Gazelle (Zootopia)
• Human Male x Female Anthro
• Jenna (Balto)
• Juno (Beastars)
• Katia Managan (Prequel webcomic)
• Krystal (Star Fox)
• Kurama - Female (Naruto)
• Kurama - Male (Naruto)
• Lamb (Cult of the Lamb)
• Maid Marian (Robin Hood)
• Master Tigress (Kung Fu Panda)
• Millie (Helluva Boss)
• Nicole Watterson (The Amazing World of Gumball)
• Porsha Crystal (Sing 2)
• Rivet (Ratchet and Clank)
• Roxanne (FNAF)
• Toriel (Undertale)
• Tristana (League of Legends)
• Vicar Amelia (Bloodborne)
Here is a LoRA, BulkedUp, that was made with Kohya's GUI. The purpose of this LoRA was to create bigger buff dudes on different Stable Diffusion models. I personally use between 0.2 to 0.5 strength, with 0.2 strength adding a bit of muscle and 0.5 going even bigger. Compared to the Hypernetworks I have worked on, I believe that LoRAs are a great alternative for training with shorter training time and better generations. However, from what I've seen on how this LoRA behaves, it seems to reflect the art style of its respective artist that it was trained on at 0.6 strength and above. Due to this discovery, I will provide the training dataset for the LoRA in the link.
Using default E621 tags with spaces, like huge muscles, works really well with the LoRA
One thing I would like to mention about this LoRA is that if it output dudes that are too huge or have bizarre anatomy, sending them to img2img or inpaint with high denoising strength (between 0.4-0.7) could really help fix them.
Here is the link to the LoRA, model formula, training dataset, and images of the examples:
NOTE: The merged model the LoRA was trained on requires the VAE, vae-ft-mse-840000-ema-pruned.ckpt, from stabilityai
protogen - Obsolete version trained on 2400 steps
protogenv2 - Newer version trained on roughly 3200 steps
protogenv2-0004 and protogen-0005 - If standard v2 feels too overfitted/overtrained to you, use these
Activation keywords are:
protogen, protogen visor, protogen face
Mr. Wolf (The Bad Guys)
Mr. Wolf from The Bad Guys, but he's a LoRA now.
Responds very well to higher weighting, like :1.3 or :1.2. The LoRA is trained at 704 resolution, so it works best at that size.
Issues: It needs testing on the paws, something seems to be up with them. Also his suit usually doesn't make sense if you look too closely
Also, yes it does nsfw.
This model is trained on Gay621
~300 images from Wizzikt.
Download link: pixeldrain.com/u/yqadCyMz
beeg wolf wife generator (Sligarthetiger)
My first attempt at a LoRA. This is LoRA Trained on 150 works by Sligarthetiger at around 4000 steps for 6 epochs. Contains two versions.
v2 is trained on Lawlas Yiffy Mix. It isn't as stable as v3, however is more accurate to the training dataset I feel.
v3 is trained on a certain anime model. More coherent and stable and personally probably better, although sometimes isn't too accurate to the data at times.
I recommend using both and keeping v2 at a weight of around 0.10 or 0.15.
As it a style LoRA, other character LoRAs work as well.
Simply select the LoRA through whatever way you usually would through the A111 extension or its native support. No activation keywords needed, it should activate on its own.
Trained on ye-18
Various (Penis Lineup, Kass, Krystal, Loona, Protogen, Puro, Spyro, Toothless
Pic taken 2023/05/16
Looking them over, some of these are likely the same ones posted before ITT, currently filed under "Birds" up above. Gonna leave it up, for posterity's sake.
Some of these are Lycoris files; check out this extension if you encounter problems.
Anthro Griffin: https://civitai.com/api/download/models/30044?type=Model&format=SafeTensor&size=full&fp=fp16
Mae Borowski (Night in the Woods): https://civitai.com/api/download/models/30127?type=Model&format=SafeTensor&size=full&fp=fp16
Marie Itami (Brand New Animal): https://civitai.com/api/download/models/30940?type=Model&format=SafeTensor&size=full&fp=fp16
Bea Santello (Night in the Woods): https://civitai.com/api/download/models/31668?type=Model&format=SafeTensor&size=full&fp=fp16
Anthro Birds: https://civitai.com/api/download/models/32214?type=Model&format=SafeTensor&size=full&fp=fp16
Rito (Species, BotW): https://civitai.com/api/download/models/41394?type=Model&format=SafeTensor
Falco (Star Fox): https://civitai.com/api/download/models/42650?type=Model&format=SafeTensor
Coco Bandicoot: https://civitai.com/api/download/models/57895?type=Model&format=SafeTensor
Elora (Spyro): https://civitai.com/api/download/models/58081?type=Model&format=SafeTensor
Zorayas (Elden Ring): https://civitai.com/api/download/models/59321?type=Model&format=SafeTensor
Tempest Shadow (MLP): https://civitai.com/api/download/models/62278?type=Model&format=SafeTensor
Secretary Bird: https://civitai.com/api/download/models/63229?type=Model&format=SafeTensor
Anthro Corvids: https://civitai.com/api/download/models/64462?type=Model&format=SafeTensor
Brooklyn (gargoyles) USE Brooklyn (gargoyles)
Bathym USE bathym
Blaidd USE Blaidd (elden ring)
Batzz USE demon lord dragon batzz
Barrel USE barrel (live a hero)
Exveemon USE exveemon
Death USE death (puss in boots)
Dire USE Dire (fortnite)
Fox Mccloud USE Fox Mccloud
Fenrir USE fenrir (housamo)
Garmr USE garmr
Freddy USE freddy (dislyte)
Guilmon USE guilmon
Horkeu kamui USE horkeu kamui (tas)
Incineroar USE incineroar
Jon talbain USE jon talbain
Law USE law (sdorica)
Leomon USE leomon
Macan USE macan (tas)
Maliketh USE Maliketh (elden ring)
Meowscles USE meowscles
Mountain USE mountain (arknights)
Nasus USE nasus (lol)
Nimbus USE nimbus (world flipper)
Renekton USE renekton
Seth USE set (tas)
Shirou Ogami USE Shirou ogami
Simba USE simba
Skavens USE skaven
Steel USE steel (balto)
Tadatomo USE tadatomo
Volibear USE volibear
Vortex USE Vortex (helluva boss)
Wargreymon USE wargreymon
Warwick USE warwick (lol)
Weregarurumon USE weregarurumon
Wolf O'Donnell USE Wolf O'Donnell
Feral on Female
Assorted Random Stuff
X/Y Grid of Artists
(Not embedded due to filesize)
Comparison of Base SD-Artist - Furry artist combos (done on an older furry model, likely YiffAnything)
thebigslick/syuro/anchee/raiji/redrusker/burgerkiss/blushbrush Prompt Matrix comparison (done on a Fluffyrock-Crosskemono 70/30 merge)
Vixen in Swimsuit artist examples (Model: 0.3(acidfur_v10) + 0.7(0.5(fluffyrock-576-704-832-960-1088-lion-low-lr-e22-offset-noise-e7) + 0.5(fluffusion_r1_e20_640x_50)) .safetensors) (DL link can be found above)
Different LORA sliders - what do they mean?
Flick on the master switch, pick lora from the drop down, drag the slider. Flick the other switch for dual sliders. Speaking very loosely, the top slider and bottom slider affect how much the lora changes the result's shape and coloring.
Have a demonstration, picrel. Asked for a jackal in a kitchen making pizza on a mixed model (mostly anythingfurry), using a lora to try to make the jackal into a Lucario.
UNet alone needed a lot of weight, but colors the figure like a Lucario without changing the figure's body much.
TEnc alone makes the jackal gain Lucario characteristics without being Lucairo; specifically we see the hips and "shorts" turning into clothing.
Both going beyond 1 starts changing the scene, probably applying training data too strongly. Both going negative turned the jackal into a fox and made a mess of features.
What about samplers?
SDE is normally used at lower steps than other samplers.
12 steps in SDE have around the same effect as 20 steps in other samplers.
I mostly tend to use the DPM++ 2m Karras or DPM++ SDE Karras samplers, with 20 or 12 steps respectively for testing and playing around, and 35 and 20 for "serious" (lol) prompting.
Euler a at 20 steps is also pretty good at prompt testing, and DDIM, from what others in these threads say, is good for fur-looking fur.
Like any aspect when it comes to SD, there is no rught answer that always works, however.
Script for comparing models
dropping by from /g/ to drop a random technical guide for comparing the similarity of different models.
Script source is https://huggingface.co/JosephusCheung/ASimilarityCalculatior, the documentation for said script however is pretty attrocious, so I made my own.
Okay, thanks? What am I supposed to do with this?
It might be helpful for figuring out if a model is similar enough for a LoRA to still work, it could help with determining if a model is worth merging into a different model, it could help with identifying models that were used in the merging of a different model where the author refuses to share the recipe, etc.
Anyways, hope it helps someone.
Use either the Wildcard or Dynamic prompts extensions!
List of wildcards: https://rentry.org/NAIwildcards
Dynamic prompts wildcards: https://github.com/adieyal/sd-dynamic-prompts/tree/main/collections
(Work even without the dynamic prompts extension if you prefer the older one, just grab the .txt files.
For use in Blender; allows for posing for use in the ControlNet extension.
"What does ControlNet weight and guidance mean?"
E621 Tagger Model for use in WD Tagger
The WD Tagger extension as-is only generates Danbooru tags, which is great when training on NAI and other anime-based models. For models based on e621, the tags may need to be changed accordingly. For that reason, you can use the following model instead of the WD one.
Convnext V2: https://pixeldrain.com/u/iNMyyi2w
Patched WD1.4: https://cdn.discordapp.com/attachments/1065785788698218526/1067966541699743845/stable-diffusion-webui-wd14-tagger.zip
Older Deepdanbooru Model: https://pixeldrain.com/u/XTcj5GHz
Top 300 e621 artist wildcard (last updated November '22)
Upscaler Model Database
Recommendations are Lollypop and Remacri. Put in models/ESRGAN
Catbox Script for easier prompt sharing
This userscript adds functionality to upload to Catbox directly from 4chan, and to view metadata for Stable Diffusion, and now TavernAI, in JPEGs and PNGs.
Follow the instructions in the GitHub to install the userscript.
IGNORE THE BELOW; catboxanon added /trash/ support on the main repo. Leaving the fork up for anyone that already downloaded it from there, will just maintain it so it is the same as the original from now on.
Follow the instructions in the GitHub to install the userscript. Forked from the original repository to add support for /trash/.
LoCon/LoHA Training Script / DAdaptation Guide