Jordyn Jones SDXL
Trained on BigLust
Uses DMD2 LoRA (https://huggingface.co/tianweiy/DMD2/blob/main/dmd2_sdxl_4step_lora_fp16.safetensors)
Recommended:
use ADetailer for face
blur: 20-25
padding: 120-200
denoising: 0.14-0.25
LoRA strength set to 1
(optional) Hires Fix (helps for ADetailer)
upscale by: 1.5-1.75
hires steps: 0-3
denoising: 0-0.3
CFG: 1
Description
trained on biglust 1.7
90 images, florence captioning, bucket sizes:
bucket 0: resolution (704, 1216), count: 1
bucket 1: resolution (704, 1280), count: 1
bucket 2: resolution (704, 1408), count: 1
bucket 3: resolution (768, 1216), count: 2
bucket 4: resolution (768, 1280), count: 1
bucket 5: resolution (768, 1344), count: 6
bucket 6: resolution (832, 1088), count: 25
bucket 7: resolution (832, 1152), count: 3
bucket 8: resolution (832, 1216), count: 6
bucket 9: resolution (896, 1024), count: 1
bucket 10: resolution (896, 1088), count: 22
bucket 11: resolution (896, 1152), count: 1
bucket 12: resolution (960, 960), count: 1
bucket 13: resolution (960, 1024), count: 1
bucket 14: resolution (960, 1088), count: 1
bucket 15: resolution (1024, 1024), count: 16
bucket 16: resolution (1216, 832), count: 1
FAQ
Comments (4)
Woohoo finally got download to work! Thx gonna test it out now
This is really well done! Excellent job training, excellent likeness. If you're taking any requests for similar SDXL models Id love to see an Olivia Ponton at this quality.
amazing work 😊
it really does compare to flux when using the DMD2 lora. play around with the lora strength, CFG scale and image steps to get different results . for example if you want to increase the CFG scale lower the lora strength . i tend to go by 3x. so if CFG and lora are set at 1 I'll increase the config to 3, reduce the lora strength to 0.33 and increase the steps by 3x. results will vary of course, but you'll find a few keepers out of a few runs .
thanks