CC License
This model may have not been released for quite some time if it wasn't for the recent developments going on with the HLL finetune models which have recently continued as "lyco/lycoris" models, the main development with HLL6 specifically was one model which was made to accompany the furry finetune model "Easyfluff" which is not only impressive when combo'd with HLL6 but is one of the notable V-Prediction models, however I had tons of trouble making a proper "V-Prediction" Based model so that will be avoided until there's a NAI/Anime finetuned V-Prediction SD 1.X model to make the merging process a lot less headache inducing (I was trying to come out with something decent for days but I just gave up and made a normal NAI based model) the main thing about this model was using a merging script for Lycoris/Loha models and merging HLL6 (the NAI) version into the models I was using for the mix to ensure that the final product would still have the capabilities of the lycoris by itself, what are those capabilities? Well, Hll6 both the Fluff and NAI version has a bunch of artist/vtuber knowledge built in, so like in the example pictures the main purpose of this model is to have access to a lot of artist and vtuber knowledge so you can just use tags and generate certain styles you want right away while still having the great compatibility that previous Based models had with LORAs. Once again thank you to HLL's creator for giving me the foundation and inspiration for creating the Based mixes, the next installment with major changes may be able to finally implement V-Prediction however if that's not the case the next Based model might be released by the time HLL7 is released.
IMPORTANT
HLL6 NAI's ability to prompt CERTAIN artists correctly is a bit iffy compared to HLL6 fluff with easyfluff, however most artists can be prompted with no adjustment to the strength of the tag (some artist may need a strength of 1.2+ to look more precise)
To use this models' built in knowledge of artists specifically there exists two files, one is for the wildcard extension or you can simply look inside the .txt file to get a list of artists, but this link also includes the .csv tag file based off all the tagging data within the HLL6 lycoris, be sure to add this to the "tags" folder for the autotag complete extension
https://pixeldrain.com/l/M81RuJK6
Future Notes
I'll be keeping a close eye on comments regarding this mix from here and outside of here to keep notes on what I can improve on, it might lead to future proofing for Based69 or just another version of Based68. Please be sure to share your feedback and test as much as you can!
Support
Some time has passed since I've been active but I have gained back some of my enthusiasm for AI image generation, a lot of my public platforms are still dormant however I will try my best to be active again, Ko-Fi is available for any donations, my Linktree has all my other social media platforms if you wish to visit them, otherwise even if my public platforms are still dormant I'm still open for personal requests to make LORAs for a certain price (I would like to use Fiverr again however the 2 wait week to cash out is not to my taste) so if you can just send me a message on
Discord: anonymousm_ai
with your request and I'll get back to you to see what we can work with from there.
Description
FAQ
Comments (12)
I think I understand the steps?
1. DL the extensions (wildcard, autotag) 2. DL pixeldrain files and put the "wildcard tag" in wildcards and the other two in "tags" in the autotag complete folder. 3. DL Hll6.3-fluff-a5a-nai? 4. use the model (No not use the lora with the model)
It must be working it is changing with different artists.
Yeah but basically all you want are the tag/wildcard files inside the pixeldrain link, you don't need the individual lyco for HLL6 NAI, unless you want to try out the more precise Easyfluff/Hll6-Fluff combination however the tag.csv file for autotag complete and the wildcard extension will work with it as well.
based
So what I'm getting from the description is that this model is a v-tuber version of easyfluff which is on furries
Almost correct, to be precise it's
"Based64 merged with other models with the NAI version of HLL6 which works better on NAI 1.X Anime based models compared to HLL6 Fluff which only looks good on Easyfluff"
You can prompt this the same way if youre doing hll+easyfluff right?
From what I recall the only thing that makes this different from the hll6+EF combo is that there's no Furry knowledge in this model so there's no need to do the "human" tag at all.
I love your checkpoints but as much as I'm excited for every new release I still find Based66 the best one as it is insanely loyal to the Loras I train and test, or just use.
Not sure if I'm doing something wrong with this one (based68) but models look quite way off original compared with same prompts and seeds in 66.
Any thoughts?
Most likely the burned in merges I did with HLL6 NAI being a factor that influences the model to not output the LORA faithfully like B66, happened with B65 if anything I can looking into version 2 fixing that issue however if it ends up looking way too different from the current model I'll just do Based69
B66 is also my favorite, but I actually like B68 being so different: same prompts give very different composition styles and is quite refreshing (I use the same list of seeds+prompt to test multiple models). Tried looking into vprediction models myself but couldnt figure out the magic behind it (I dont have much technical knowledge of it all) and merging them seemed to mess everything up.
V2 is definite improvement upon V1, but still not really there as B66 is.
I don't know the details how you did it but somehow you got a golden hit with B66 to be lora loyal to the extreme.
B.t.w why did you deactivate comments on B66, always wondered about the one?
This is the best model which I have test for drawing group-sex pictures. ^y^
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.