--- PONY v2 ---
Illustration re-dialing.
Better Pony prompt adherence
General Refinements
Rebalancing of art sources.
--- PONY v1 ---
I finally finished translating this model to PonyXL. It took a while, sorry for the extended update time - it'll be updated more frequently moving forward. The entire data-source had to be reimagined and recreated in high resolution. This model is trained on the base Pony 6 model, along with 25 years worth of my own design and artwork. It's not perfect and it really likes to generate people. It is capable of everything Pony is typically capable of, and does a pretty good job at it. I'm very happy with the final aesthetic here, and while it differs from my personal art it has created something new that I've become a big fan of. As always, please let me know if you run into anything I should know about - or if you have any thoughts or ideas on how to improve it. Mostly, have fun with it and share your artwork so I can admire your creativity!!!
--QP.
--- SD1.5 v1 ---
This is a model that a friend convinced me to upload here. It's trained on some of my personal illustration work (around 150 pieces) - some of that was corporate work, but a lot of it was panel art or print art. You may notice that I'm using a sampler that you may not know about, "Euler_Max". You can install it as an extension for Automatic1111 if you do the "install from URL" and paste this in:
https://github.com/licyk/advanced_euler_sampler_extension
Anyway, I've been using it a lot lately behind the scenes, and I just updated it. It's pretty flexible and can move from more detailed illustration work to comic section-style art fairly easily with a few prompts. Negative prompts have a big impact on this model, so I would recommend not using an embedding. Adding "photo" into the negatives can help if it insists on being a little too interpretative.
I'll be updating this periodically, but not on a regular schedule. I use it to help me sketch out ideas and tend to update it as I complete enough new art.
Let me know what you think, this model won't be perfect since it's based on my Arrogant Bastard - Pulp Fantasy model along with training on my own personal artwork (thus the Calamity AND the Junction).
Description
FAQ
Comments (5)
This continues to be one of my favorite models for character generation. Any plans for an Illustrious or Noob VPred version?
I've been experimenting with Illustrious a little bit but haven't gotten to anything that I feel really benefits from it. I haven't spent any real amount of time with v-prediction.
(for other readers)
v-prediction is some math behind the scenes.
SDXL doesn't really support it, as you pointed out NoobAI does. Flux is using something called flow-matching. re:NoobAI is primarily a comfy thing and I try to make sure my models will run on A1111/Forge so the most people can benefit.
I have been doing a lot of work on SD3.5 though, and while I'm not ready with any of that stuff yet... it's very promising. I'm still waiting on either A1111 or Forge support before I release those. I think I'm skipping Flux, just due to licensing between Flux-D and Flux-S, and the difficulty of really fine-tuning Flux-S which has better licensing.
Maybe that's not the best news to hear. I am working on some faster SDXL models though with lower step requirements (I'm just not happy with output yet).
@QuadPipe All great news. I'm not capable of running FLUX at a decent enough speed to make it worth while. And I was only asking about illustrious out of morbid curiosity. This model, very particularly, is my absolute favorite one and I don't need anything more than this one right here <3
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.