Ponymagine 3.1

VersionsGamma - Aki ver Enabled support for Pony LoRAsImproved limb qualityDecreased background quality due to additional merging of PonyDiffusion CLIP/TE**(Possibly) better eroticism- Some Pony LoRAs might not work fully, but at least they won't produce flaky noise anymore at least based on my testing. Any additional TE merging to improve Pony LoRA will lead to loss of Character/Artist knowledge of the Animagine model**- I recommend using IPadapters or the Worldly XL LoRA for better backgrounds https://civitai.com/models/376395?modelVersionId=420317Beta - Salome verInitial published versionLeans more towards AnimagineDoes not support all Pony LoRA. Most will result in Noise!PurposeThe initial purpose of the model was to infuse character/artist knowledge from Animagine into PonyDiffusion without using weird tokens like 'aua' and to be able to use LoRAs trained on either of the models.The end result, while it retained the knowledge, it washed up some of the character/styles however it gets 'close enough' with more tagging. It supports masterpiece tagging structures but adding some 'score' tags doesn't hurt.Merge DetailsGamma is a merge between PonyDiffusion v6 and Animagine 3.1 . After the merge, a Lycoris extracted between the difference of AutismMix Confetti and PonyDiffusion v6 was merged at 0.8 ratioBeta is a merge between a PonyDiffusion v6 mix (AutismMix Confetti) and Animagine 3.1 made possible by using DARE merging.This is a BETA version. It is not meant to be perfect and it is not perfect. The main point of it is to get the word out that Merges between Pony-styled models and other SDXL models is possible without artifacts.DisclaimerAs somebody already mentioned, not all Pony LoRAs work correctly on the Beta model ! This was achieved during its alpha merge stage but was scrapped in favor of DARE merging. I have an idea on how Pony LoRA support can be enabled again but it will require more testing on my side. Added support in GammaThis is an image with the current noise when using Pony LoRA on BetaThis is an image with a Pony LoRA on an older Alpha versionSamplesAs I am hailing from /vtai/ the samples will focus on vtubers. This doesn't mean the model is not able to do anything else. Post your own creations and I might feature them instead.The Intention of the posted Grid images is not to cherry pick and display the best images created with this model but rather to show the losses in character/style that occurred as a result of the merge. I am sure others can make better merges using this method.The Generation data shows 'Aniamgine-xl-3.1' as the model because I had 3 models loaded at the time of the model merge. Animagine, AutismMix and the resulting merge. The samples have been extracted at the same time as I was testing the merge. ComfyUI is weird in that sense.LicenseAs the license of PonyDiffusion is more strict, the model will follow the same modified version, quoting the following from the PonyDiffusion model:The following modifications have been added to Fair AI Public License:You are not permitted to run inference of this model on websites or applications allowing any form of monetization (paid inference, faster tiers, etc.). This applies to any derivative models or model merges.Merge RecipeTo find out/reuse the merge recipe you require ComfyUI and the DARE extension.After both are installed, you can just drag the model into ComfyUI to load the same workflow I had when the model was merged. (WARNING: very messy!)JSON sample also added to Model's Training FilesAcknowledgementI didn't expect so many people to download and like a model that was meant for testing purposes. I have a new build in testing that does not generate noise when using Pony LoRAs but currently their effect is dampened or not working at all. I am working on improving it to allow Pony LoRAs to work as well. The sad part is backgrounds and aesthetics seem to be affected the more LoRA compatibility there is and concept LoRA don't seem to work very well as of this moment.