FinalCut 4k Fantasy
Fully custom training enforcing natural language.
FLUX meets PONY
High PONY and SDXL LORA capability
Description
FAQ
Comments (9)
Very impressive model. My first impression is that finally someone has fixed flux-face. But maybe you just bypassed it instead, which is clever.
It is SDXL/PONY based with a large amount of natural language training but it does not have a LLM like FLUX or the ability to generate single words like SDXL base. There is a limit to feature extraction/cosine similarity at 77 tokens but I am not sure we have reached it with SDXL
Would it be accurate to describe this model as an attempt to find the sweet spot between XL, Pony and Flux architectures?
@akdcpk SDXL had single word recognition, which was likely just extensive text/image pair training, where as flux has LLM guidance on its side. I am hoping to get a smarter SDXL/PONY base but might be pushing the limits on CLIP
I notice the initial and upscaled images are slightly different proportions, with the upscaled image being wider than the initial image.
you can use 1 to 1 but in my testing the offset can provide a touch of extra realisim
The speed at which this model is able to generate high-quality 4k images is amazing.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.