Type | |
Stats | 329 0 |
Reviews | (27) |
Published | Feb 23, 2024 |
Base Model | |
Usage Tips | Clip Skip: 2 |
Hash | AutoV2 109AA347C1 |
3D + 2D + hairstyles + clothing + photography
Images should have magazine proportions i added this dimensions to all training dataset:
512x768
768x1152
>>>post more portraits so i can keep training further versions :)
**fixed faces
**added realisitic portraits
**added artist use prompt ((works by ... )) and ad your favorite artist name,
**simple prompts work well, also complex prompts with higher CFG scale work well
be specific: Photo Realistic or Cartoon or Anime:
(photo realistic:1.4)
(cartoon:1.4)
(anime:1.4)
also added poses use:
(dynamic pose:1.4)
{photo shoot pose:1.4)
{fashion pose:1.4)
- Clip skip: 2
- CFG scale: 5 - 11 gets best results
- Sampler: All DPM++ samplers will get best results
- Sampling Steps : 30+
- Hires Fix, Restore Faces
- ControlNet
etc.
I tested it with al sorts of settings and works fine,
USE SIMPLE NEGATIVES > i found that not using negatives at all gives cool results at lower CFG Scales like 5.5 get cool stuff, let it do its thing a bit, or you can also go strict 9-12
(worst quality, low quality:1.4), FastNegativeV2,