Created this LoRA as a first test of training with FLUX.1-dev - thank you to fal.ai for providing the compute!
Trained with activation token "ASCII art" but I recommend using this entire prompt schema for best effect: ASCII art on a white background, made of letters, numbers, and other symbols, dithering effect,
To test this LoRA, I used this workflow (not using the Upscale pass). Flux Guidance of 3.5, euler 40 steps, using fp8 unet and clip - Tenofas v1.0 - FLUX with LoRA's and Ultimate SD Upscaler | ComfyUI Workflow (openart.ai)
Obviously, it's not perfect ASCII. I'm still impressed with the performance VS previous test models trained on SDXL and SD1.5. Hit rate can be prompt dependent.
Can also be downloaded from huggingface if you prefer.
v1 - trained with fal.ai's flux dev lora trainer. Used 12 images, 4e-4 lr, 2500 steps. Settings were picked somewhat at random, as this is my first flux lora, so don't try to read too much into this.
1. 재게시된 모델의 권리는 원 제작자에게 있습니다.
2. 모델 원작자가 모델을 인증받으려면 공식 채널을 통해 SeaArt.AI 직원에게 문의하세요. 저희는 모든 창작자의 권리를 보호하기 위해 노력합니다. 인증하러 이동
