Showcase images are in PNG and include metadata & workflow
All licensing conditions for the Flux Dev model apply. You can find the details here.
Description
weighting between 1.2 - 1.6 ( you can test outside of this interval)
Showcase Images were created with Flux1.dev fp16.
LoRA training images base were created with Pixel Alchemy.
In order to explore the extent of the possible renderings of this LoRA, it is recommended to use these tags which were used in the training. Do not of course use them all at the same time, but depending on what you are looking for : Perspective, Proportions, Construction Lines, Shading, Freehand Drawing, Rendering, Architectural Details, Schematic, Loose Lines, Abstract, Conceptual, Minimalist, Gestural, Outline, Impressionistic, Visual Brainstorming, Multi-View, Composite View.
The model seems to show skills outside of architecture and interior design, yet he was only trained on that. Don’t hesitate to test it on other themes.
FAQ
Comments (10)
#AwesomeSauce
Chapeau! 💖 It is well trained on portraits, took me some to get story drawings generated, but succeed! This is a great LoRA!
Thanks for the feedback
Very adaptable, will do a lot more than portraits, mixes brilliantly with other loras.
So much the better, yet the model was only trained on portraits.
@Adel_AI I have several loras that work in that way, Eldritch Parchment is another... the question is why.
@robadams171 I have tested other models trained only on portraits, with a base containing about ten images. As soon as I try to do something other than portraits, I get anatomical aberrations and disproportions in the sizes
@Adel_AI Perhaps the learning rate, I get anatomical problem at 8e-4 but not as much at 3e-4. More steps needed of course. Flux is hard to predict though, it will often do the unexpected.