Can use Pony in the same way as Animagine XL 3.0/3.1
anima_pencil-XL based Pony
Both Pony and Animagine XL 3.0/3.1 LoRA/QualityTags maybe work.
License: Fair AI Public License 1.0-SD
You should share the merge recipe if you release a model merged with pony_pencil-XL.
This license does not add any restrictions on your generated images.
For more details, please see the license section of ANIMAGINE XL 3.0 / Pony Diffusion V6 XL
Please see the "About this Version" section for information on each version.
HuggingFace: https://huggingface.co/bluepen5805/pony_pencil-XL
Description
# Checkpoints
# LoRAs
self-made QualitySlider (Concept Sliders)
# Recipe
anima_pencil-XL-v3.1.0 + ( Pony Diffusion V6 XL - SD-XL 1.0-base ) x 1.0, Add difference(trainDifference)
anima_pencil-XL-v3.1.0 + [1], Weight sum(cosineA), Elemental: "BASE:conditioner.embedders.0:0.35,BASE:conditioner.embedders.1:0.5"
( [2] x 0.65 ) + ( PD for Anime v2.0 x 0.35 ), Weight sum(cosineA)
( [3] x 0.9 ) + ( AiPonyAnime v2 x 0.1 ), Weight sum(cosineA)
( [4] x 0.85 ) + ( MomoiroPony v1.4 x 0.15 ), Weight sum(cosineB)
( [5] x 0.25 ) + ( anima_pencil-XL-v3.1.0 x 0.75 ), Weight sum(cosineB), Elemental: "BASE::0"
( [6] x 0.75 ) + ( [4] x 0.25 ), Weight sum(cosineB), BASE::1
( [7] x 1.0 ) + ( [2] x 0 ), Weight sum(normal), Elemental: "BASE:conditioner.embedders.0:0.45,BASE:conditioner.embedders.1:0.65"
[8] + ( anima_pencil-XL-v3.1.0 - sd_xl_base_1.0 ) x0.85, Add difference(trainDifference), Elemental: "BASE::0"
( [9] x 0.75 ) + ( anima_pencil-XL-v3.1.0 x 0.25 ), Weight sum(cosineA), BASE::0
[10] + QualitySlider1 + QualitySlider2 + Aesthetic Anime LoRA:0.3 + Artist Coloring Style Lora - SDXL Anime Style v2.0:0.2 + XL Outline / 轮廓线 v1.0:0.25
( [11] x 1.0 ) + ( ebara_pony_1 x 0 ), Weight sum(cosineA)
( [12] x 0.98 ) + ( anima_pencil-XL-v3.0.0 x 0.02 ), Weight sum(cosineA), Elemental: "BASE::0"
[13] + ANIME DETAILER XL LORA:1.0 + flat_sdxl v1.0:-1.2 + lora eyes xl v1.0:0.6 + Better Blur Control / Less Depth of Field XL LoRA v1.0:0.2
( [14] x 1.0 ) + ( [10] x 0 ), Weight sum(cosineB)
[15] + QualitySlider1 + Artist Coloring Style Lora - SDXL Anime Style v2.0:0.6
FAQ
Comments (14)
Finally!! you did it!! I was wondering why it took you so long XD I'm going to test it and I'll make a review on my youtube channel! Thanks!!
I was very excited to try this model but I cannot get it to draw men at all. even with a LORA, even with male focus, 1boy, man positive tags and 1girl etc negative tags - nothing works. :(
Please wait for the next version. 🥹
@blue_pen5805 will do, thank you! I love your other models so I can't wait for this one to draw boys. <3 I appreciate your response ^_^
@blue_pen5805 Works amazing now drawing boys. Thank you so much, the model is perfect now! ^_^
Software translation:
Very good, just a simple test, unfortunately when used with my own Pony lora, the result is not enough to restore the artist's style. It's more kawaii style. Well, after all, your checkpoints shouldn't count as pony series?
Hey mate your checkpoints are good, no need to give a crying emoticon, just a little less friendly to lora. Would change lora style a little bit.
amazing! I've actually been looking forward to something exactly like this. But it unfortunately loses scope of some of the more obscure characters from Animagine, any chance you'll bring them back?
nice job
Solid model. However, I need to voice a singular issue: the style of the model is erratic.
It puts out varying styles, randomly. For me, I can try to make it more consistent with some tinkering but that has limits.
It goes for good outputs to randomly very weirdly colored gens, almost as if it were a failed run.
I think ALL pony-based models have this problem to some degree = especially if you copy booru tags with "_" symbol in them - then the model starts to "remember" which images it was trained on. So it is not a big deal - but a more universal problem of the base Pony (unfortunately)...
@OliviaRossi I did a pretty basic run. Other models would have been consistent.
@User37 maybe this model needs more in-depth training, since Pony is "Anime"-ish already, so it is hard to tell what is and what is not within a certain style
@OliviaRossi given a new version was released, I will need to test it again. For now, while I view this model negatively, I believe that it has potential.