If you like my work, drop a 5 review and hit the heart icon. it's free and keeps me motivated
All my models are officially hosted and maintained by me on Tensor.art . use my Exclusive and public model for free on tensor.art
06/06/25 : v2.0p released
Main focus of 2.0b is cleaning the style influence of v1.5
19/02/2025 : On-site Generation added
Private commission on fiverr
Get early access to my upcoming NSFW Lora on my Patreon.
early lora share on discord for patreon.
I was hoping you could support my work by joining any one of them and get early access to all my upcoming loras and other perks, such as fan requests and a Discord role.
Join my Discord Server
check the images for prompts
use lora at 0.7-1
Adetailer for faces
Img2img upscale
4x-ultra sharp
comment you idea or request
Suggested settings:
I had CLIP skip 1 on every image
Samplers : Eular A, DPM++ 2M
CFG : 5-6
I had ENSD: 31337 all of them
All of them had highres.fix or img2img at higher resolution.
I don't use restore faces
I use Adetailer for face details
4x-Ultrasharp upscaler
positive: masterpiece,best quality,amazing quality,absurdres,
negative: bad quality,worst quality,worst detail,sketch,censored,watermark, signature, artist name
Join my Discord Server
Description
ilxl 1.0
noob added too
testing if can continue this base coming forward
FAQ
Comments (68)
for v3, the nooob version is eps or vpred?
Esp.
If it was vpred I would have named it vpred.
I found vpred to be not very much Lora friendly. Atleast on my end. I might be doing something wrong or vpred is just that much rng.
Goofy_Ai thanks¡, yeah vpred is not that easy to work apparently
Goofy_Ai The training parameters for vpred are slightly different.
In my own testing, LoRa performance is similar to EPS. LoRa tested with VPRED works well with EPS, but LoRa tested with EPS performs very poorly with the VPRED model.
NTR_BLACK yes I tried different parameters and talked with different person about this. If you have different settings that I can try then you can dm me
Goofy_Ai
it'my setting,For reference only
Train batch size: 3
(img 4_ 8_ 16_... I have adjusted the image repetition ratio for different 3D MODEL、clothing and faces)
Max train steps: 600-1000
prodigyplus.ProdigyPlusScheduleFree
weight_decay=0.02 d_coef=0.9 use_bias_correction=True
Min SNR gamma:2 (or 1.5)
Scale v prediction loss:on
IP noise gamma:0.05
dont use any Noise offset or Multires Noise
--zero_terminal_snr --v_parameterization
i have share some v-pred lora
Another release, and V1 is still the gold standard. If you like v1, but want much more anime style in your gens, you might enjoy V3. For me, V1 still reigns supreme and it ain't even close.
edit: To offer a little more charitability, I think my dissatisfaction with every version since V1 is more a testament to how incredible V1 is more than anything. At this point, I don't think there will be anything that competes with it until the next 'big thing' comes around.
I just want to ask and request... please add some Indian places, aesthetics, clothes and environment as well. Maybe I'm in the minority, but if you can, I would be happy.
I would love to ask how version 1.0 was made, what mix was it? It's simply fantastic.
great
Will you do an anime finetune of Qwen Image?
how many buzz to bid for generation?
as many to be in the top 200
still can't afford,see u 3 days later
V1 is the best.
All later versions are horrendous at a 2.5d art style.
Even with "Realistic" prompts, the images are flat as fuck.
Pretty sure OP is making non-updates to farm buzz with these new versions....
I don't see the need of being petty here tbh, leave that attitude to the digital artists, lol.
Goofy has a lot of successful loras and models, I don't think so, I think he's just not focusing on the same exact style. You can always use v1 if you want to
i feel bad for you that newer versions are not of your liking, but my goal for checkpoints were always anime style first and to have a unique style different from other checkpoints even between each versions. and as other people said you can always use the older versions. and about the buzz farming is funny as you dont need to donate anything and can just wait and give me 0 buzz and i wont even bother. people who donate just donating for the time i spent nothing else. there should be other models that focuses more on 2.5D
Any recommended highres denoising strength?
0.3 when doing highres fix
Goofy_Ai I appreciate it 🙏
Sir, is there a way to check how many new characters were added in the new version?
I like the new version, but I miss the style of version 2.0. Will we be able to have a Lora that mimics that style?
good
Holy crap 😎🤏😳🕶🤏, amazing improvement, LET REMIX!
Hello!
Does this model use the latest danbooru dataset, or an older one?
It's merged with NoobAI so it's knowledge cutoff is October 2024
love this lora, thank you
This model is amazing!! I've tried lots. Congrats!!
Making male characters with this model is quite frustrating, it seems heavily swayed towards female characters. Even after quite severe prompting, I'm still getting male characters with breasts.
Try using images to generate images, I have already created three great male characters
Wow, this is so clean!! Love it
looks good
Out of all the other models I've tried, somehow I keep coming back to this one. I don't know what you did but keep cooking.
nice
Looks so great!
This model is great!Thank u so much!
An incredible model
The addition of Illustrious XL 1.0 and NoobAI really helped with LoRA compatibility and model flexibility in term of various topics. Great work! Looking forward for Illustrious XL 1.1 and NoobAI EPS 1.1 upgrade 🤗
Not Bad model
Pretty good
Pretty useful
NIce Checkpoint
!
nice
Cool
nice
Very nice
Really happy with how perfectly this model recreates RWBY characters. Haven't tested it extensively but it gets outfits and motifs for the main girls precisely even at lower CFGs.
Most beautiful breast generation in my opinion.
beautiful what?
nice
Please, because version 2.0 doesn't work...😭
es una pena que lo hayan bloqueado, ojala lo arreglen pronto, en lo personal, creo que esta versión es la mejor
@IndependentTart595498 Confirmo, prefiero la Versión 2.0 a la 3.0 y por mucho...
Ye i need v2 back…
very good
does this model use booru tags?
Loved V2, why disable it :(
bring back V2
bring back ver 2
OMG!!!
I love it!!!
This needs v2.0p back, badly
Gorgeous work, i like it!
NEED v2.0p back
The differences in quality of photos from v2.0p to v3 is HUGE, and not in a good way.
Photos from v2.0p are much more consistent in style when comparing to v3, and is WAY better at handling lighting in multiple different situations.
From what i can see v2.0p is better in every way.
v3 feels more like a cheap alternative to v2.0p however it isn't even cheaper OR faster to generate with.
very good model