tl;dr : Lugianon Mix is a FURRY-ORIENTED merge based on multiple NoobAI models, the main quality of this model is to ally pristine artist style faithfulness and excellent anatomy and image composition.
V2 VERSIONS ARE OUT! Try V2_Ouro for better cartoonish/lineart styles, and V2_Boros for better realistic styles.
If you're just looking for the version that matches the V1 style the closest, it's Boros.
Summary (for V1)
This very simple merge is an attempt at fixing the main weaknesses of two furry models that I use regularly : 3wolfmondai and manticore. NB : The V2 merges use other models.
3wolfmondai has trouble with artist style faithfulness, but has way better general composition and anatomy than manticore
Manticore struggles with some concepts and composition in general, but has excellent artist style faithfulness
After testing, I found a ratio which keeps the best of both worlds while reducing the weaknesses of both original models. The result is a merge that handles realistic and traditional artstyles very nicely.
Also the merge is named Lugianon because that's what people have been calling me on /sdg/ since I mostly post Lugia gens.
Use this as you would any other VPred NoobAI-related model, just wanted to have this out there and I'm very happy with the results from this merge.
Important Information
These are my personal settings, but i use as a base
Positive prompt : masterpiece, best quality, furry,
Negative prompt : lowres, {bad}, error, fewer, extra, missing, worst quality, jpeg artifacts, bad quality, (signature, watermark, patreon logo, patreon username, artist name, artist logo:1.2), extra digits, artistic error, username, scan, bad anatomy, bad hands, worst quality, low quality, normal quality, mutation, mutated, extra limb, poorly drawn hands, missing limb, floating limbs, disconnected limbs, malformed hands, mosaic, bad faces, bad face, {{{bad eyes}}}, bad feet, extra toes, earrings, {{{@_@}}}
50 steps, cfg 4.0, rescale cfg 0.6, euler_ancestral, sgm_uniform
Description
FAQ
Comments (10)
What a wild model! Great variety with what I prompt for. Gives me results that I'm genuinely not used to seeing at all from any other model. A huge first! When I ask for a specific pose, angle, look, style, feeling - it's followed as commanded.
I'll give a more earnest thought as I continue playing with it. But as for now? This is a righteous model that allows me to work with things that I feel not even the models you used to merge together could do efficiently. I'll be posting some of the things I genned with this model eventually.
Whoa, What an amazing merge! It really like Manticore but way better.
My recommendations
1) (vpred = on, zero SNR = on)
2) Make your prompt like this:
(masterpiece:1.3), (best quality:1.2),
then
sensual /sensitive / explicit / general / safe / uncensored
then
Your prompt combined of plain text and -booru tags
3) Don't use new lines. Put all positive prompt into one line - it's very efficient. You can use dots or semicolumns for logical separation but it will worsten consisstency.
4) Don't use negative prompt at all, empty nega makes consistency on the next level.
5) Sampler - Euler Negative Dy (built in reForge)
6) Scheduler - Normal (all others - separate characters, make blank faces and scrambling bodyparts)
7) Steps >95 (yes yes, I know, but you will be amazed how Euler Dy - which means 'dynamic' will dynamicaly adopt the picture closer to the end).
8) CFG =3 (lower values makes picture inconsistent, bigger values emphasize tags with bigger weights to much)
9) DPO lora worsten quality, cfg config lora - messing with style, cfg rescalers kill palette, but characters lora works ok.
10) Resolution - 1024x1024 only, any other will mess with character positioning.
11) Instead 1girl, 2boys you can try duo, trio - it helps picture composition even if focus on / closeup on are used
The prompt comprehension on this is very good compared to the other models. I hope thats a feature which continues to improve and I hope this model becomes a mainstay on the site.
bro your model is not available for generation on civitai (only download), please return, your model is AWESOME
i paid 38000 buzz to have it featured for a week, i don't have anymore buzz to invest so it's not up to me. If you have a few, consider placing a BID to help the model stay featured and thus enabling onsite generation.
@lugianon_sdg heres hoping 3k is enough to get it goin. big fan of this one, its been getting what i want w/o needing like 5 different loras slapped on
@lugianon_sdg Unfortunately, I have very little bazz, sorry
@driftingphoenix thanks, i've reinvested the buzz from onsite gens into bids as well, we should get it back in about 3 days
@lugianon_sdg I hope at some point this model really catches on because it's fantastic at prompt adherence.
yes
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.