Hi
this is clean and sharp mix of base Stable Diffusion 1.5 model which can do 3d/illustrations/photo etc and there are no analogues, made to be close to original model so no surprises here
Styles comparison
vector illustration
3d render
photo
STABLE v2.2 - fix for v2 for illustration bias shift in some cases, also overall quality improvements, check comparison grids
UPD: there are still some small misbehaviours occurring occasionally due to influence of loras, should be fixed in next iterations
____________________________________________________
previous checkpoint changes full size1(ezra) full size2(santa)
STABLE v2 - starting new version of checkpoint based on test mix STABLE_src with step-by-step tuning. now it's tuned with samdoesarts_v2 lora which gave this mix nice illustrations, but the bias shifted to art where it's not needed and will be fixed with next release soon
STABLE v1.5.1 - tuned with samdoesarts lora v1, improved quality of illustrations
STABLE v1.5 - tuned with girls showing tongue lora v2, better overall quality
STABLE v1.4.2 - tuned with walking robots lora, slightly better overall quality
STABLE v1.4.1 - slightly better overall quality
STABLE v1.4 - more realism, more art, more overall quality improvements
STABLE v1.3 - one of the test mixes gave overall improvements, so I made another release
STABLE v1.2 - made another merge with sd 1.5 (due to tongue v1.1 lora merge, all girls looked more like bimbo then natural beauty)
STABLE v1.1 - now it's more sharp
STABLE v1 - starting new mix with fine tuning by own loras
Description
FAQ
Comments (7)
The name of this model trips me up every time because it looks like one of those pointless spam uploads.
you got the vibes
Gah, you got me again! Incidentally which version is the finished version, 1.0?
Edit: I think it's JJ2 because it was uploaded soonest.
I agree totally and not even sure why I gave it a chance, but I do not regret it. Can't wait to see what the update does!
@tombot jj2 latest one for the moment, makes slightly better compositions and adds more details in comparison with jjc2.
@MysteryWrecked new version produces slightly clear and more detailed pictures. In rare cases some pictures could be broken due increased amount of details, so it needs reconfigure the token wights or steps
to be honest I didn't expect the update, but jj2 is definitely works better than jjc2, so I released it
Хорошая модель! Генерирует более-менее правильные тела, нормальные руки. Цветность и детализация выше чем у других подобных моделей.
спасибо
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.