LOFI-EVERYTHING (LFE) is my DynaVision model with some creative LORA merging to create a cool LoFi style. It works for both TXT2IMG and IMG2IMG, and can create some pretty eye catching illustrative lofi style scenes. LFE is very coherent and is easy to prompt, and is pretty good with hands. It does it's best work at 768 or higher resolution, and I've included several native 1024 output (no HRF) images so you can see how well it works. There is no keyword for LFE to do it's thing, though it doesn't hurt to add LOFI to your prompt to help guide the output towards the established style.
Description
This model includes a baked vae (mse 840k) - Included both pruned and unpruned, both are FP16.
FAQ
Comments (4)
This seems like it should be a lora, so the weight can be adjusted to change the effect
Sure, though fair warning I'm going to scrape it off of my own DynaVision model that I created it from, so the results may be a bit different when used on other models. I'll update this listing to include the LORA here in a bit.
@socalguitarist Yeah, I get you... I mean, I have previously tarined Dreambooth Vs. LORA and still, even you do the cleanest and focus train possible, you will never get the same aesthetics vibes -and maybe variations on isolated prompts- as in Dreambooth, it's impossible. Same for extracting from it. LORA's are limited but for size, use, etc. are perfect. I will wait for the LORA too, wil be fun to test it.
@LDWorksDavid I'm actually having issues pulling it out into a LORA, running into some errors that I can't seem to get around. I haven't given up yet, but need to try to find another way to extract the LORA, as it's just not wanting to play nice with my attempts to scrape DynaVision out of it.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.