Downloads · 30 days
12
3% of all-time downloads
AlekseyCalvin/HSTcolorFlux_LushApparitionLoRAII
HSTcolorFlux_LushApparitionLoRAII is a text-to-image model from AlekseyCalvin. Use it when you need an image from a text prompt. It is set up for diffusers. The card lists the license as apache-2.0.
Downloads · 30 days
12
3% of all-time downloads
All-time downloads
367
Public
Repo size
120 MB
Likes
1
Public
Click a slice to open those files.
.safetensors118 MB · 98%
From the Hugging Face model README
Our two HSToric Color Lush Apparition HSToric Color low rank adapters series are belatedly shared byproducts of our early Flux training experiments and projects Flux circa summer of 2024. Basically, we wanted to reposition the definition of "quality" when it comes to photorealistic inference using the open/distilled Flux variants, particularly Schnell. <br>
To achieve this, we turned to the very beginnings of historical photography in color: the archival legacy (from circa 1900s-1910s ) of the photographers Sergey Prokudin-Gorsky, who traveled and photographed widely in those years whilst perfecting implementations of a pioneering 3-color-composite photography technique.<br>
This LoRA represents one of our many attempts/approaches to charging the representational powers of Flux with the uncanny character and finesse of Prokudin's visionary captures. <br>
This is because whilst producing these two LoRAS we were experimenting with targeted fine-tuning solely over specified layers/components of the base model. <br>
Trying out all sorts of layer combos, we strayed far from the commonplace quick-train surface targeting character LoRA templates. <br>
Naturally, these trainings were also high rank. See the configs for more details. <br>
As an unforeseen side-effect of the above training regimes, Lush Apparition Version 2 (this one) approaches text prompts as seldom more than vague suggestions; mere non-obligatory nods towards this or that direction or theme. <br> But in this LoRA's special case, that demonstrably isn't always or neccessarily such a bad thing, as the imagery it produces tends to be aesthetically compelling and even nominally photorealistic, with a clear (if unintended by us) bias towards a sort of photography often spotted at various MOMAs or/and fine art photo galleries. <br> And sometimes this adapter invents truly phantasmagoric creatures and scenes, drawn from even the most vanilla oat milk-&-toast prompts. <br>
On the other hand, this LoRA's slightly older sibling, Lush Apparition Version 1 takes prompts fairly well, if sometimes injecting its extrapolations thereof with surprising spins. <br> Finally, you might also consider trying out some of the LoRAs we personally consider the most successful T2I productions of our HSToric Color series. Among these are: <br> HSToric Color Dedistilled <br> HSToric Color Schnell <br> HSToric FOCUS <br> HSToric Color SD3.5L V.1 <br>
Also check out our new Text-to-Video additions to the HSToric Color series: <br> HSToric Color Film WAN14B v1 (Rank 256) <br> HSToric Color Film WAN1.3B v2 (Rank 128) <br> HSToric Color Film WAN1.3B v1 (Rank 32) <br> And more on the way! <br>
You should use HST style autochrome photograph to open a vivid window into a chromatically bewitching ever-futurous past.<br>
from diffusers import AutoPipelineForText2Image
import torch
pipeline = AutoPipelineForText2Image.from_pretrained('black-forest-labs/FLUX.1-dev', torch_dtype=torch.float16).to('cuda')
pipeline.load_lora_weights('AlekseyCalvin/HSTcolorFlux_LushApparitionLoRAII')
image = pipeline('your prompt').images[0]