Type | |
Stats | 894 534 1.7k |
Reviews | (123) |
Published | Dec 2, 2024 |
Base Model | |
Training | Steps: 2,000 Epochs: 5 |
Usage Tips | Clip Skip: 1 |
Hash | AutoV2 3BA498621D |
MacaronFLUX aims to improve seated/leaning/reclining poses and adds more of a fashion/culture magazine aesthetic. The lora improves initial generation details and decreases the amount of background and hair blurring that FLUX tends to produce. It also adds a subtle atmospheric lighting and grain effect. You do still have to prompt the poses you want.
Prompting
The example images were prompted somewhere along the lines of the following:
"A fashion magazine photoshoot. A young {race} {man|woman} with {hair-color} hair {leans against|reclines back|sits confidently} in an playful pose in a {__interior__|__location__}. The image has atmospheric lighting and a heavy grain."
Note: The dataset for this model was not very broad since I was going for a specific aesthetic and was trained on mostly images of european/african ethnicites so can sometimes struggle with asian faces 50/50.