Type | |
Stats | 1,202 280 |
Reviews | (37) |
Published | Mar 14, 2023 |
Base Model | |
Training | Steps: 10 Epochs: 40 |
Trigger Words | <lora:aoyem_v5_10rp_40ep_1e-6-000030:0.8:outall> |
Hash | AutoV2 9EC33D58CA |
V4: Update by Locon trained: Ao Yem - Vietnamese Dress 2 | Stable Diffusion LoCon | Civitai
V3:
Must use LORA block Weight at OUTD or OUTALL: <lora:AoYem_V10_4rp_9ep_1e-4:1.5:OUTD>
<lora:AoYem_V10_4rp_9ep_1e-4:1.5:OUTALL>
You can change the weight from 0.2 to 1.7. Weight 1.5 & 1.7 are good for both OUTD or OUTALL.
Must use DPM++ 2S a Karras as Sampler.
CFG scale: from 5-7.
Higher native 512x512 resolution will generate extra people even it is mentioned in negative prompt. So that you must you Hires fix if you want higher resolution. Use Restore faces for better face, also.
Hires fix with Desoising: 0.55 - 0.6, Upscaler: R-ESRGAN 4x+
V2:
Must use LORA block Weight
Below are best lora weight with sampler:
Use <lora:AoYem_V5_10rp_40ep_1e-6-000030:0.8:OUTALL> with Sampler: DPM++ 2M Karras | CFG Scale: 6
Use <lora:AoYem_V5_10rp_40ep_1e-6-000030:0.8:OUTD> with Sampler: DPM++ 2S a Karras | CFG Scale: 5
You can try with another CFG Scale and Sampler for best fit with you.