First release










Fairly happy with the outcome given the fact that it is hard to come by suitable source images. Not as flexible as I would have liked so you might want to really emphasize details in your prompt regarding background and/or outfit or you'll be ending up with little variation. Also does a slightly better job with semi-realistic checkpoints and styles.
Recommended strength around 0.9 - CFG-Scale depends on the checkpoint.
Here is a general tip that might help you overcome some hurdles in terms of flexibility for LoRAs and checkpoints as well: use a more flexible checkpointer and sampler-combination for the base generation and switch to a more photorealistic-leaning combination (might even only switch the sampler, e.g. from an Euler-variant to UniPC) during high-res fix with a reasonably high denoising strength. So essentially use the base generation as a "sketch" and bring in the details on the highres/upscaling pass - which of course also works as two completey separate steps/workflows.
First release
1. 转载模型仅供学习与交流分享,其版权及最终解释权归原作者。
2. 模型原作者如需认领模型,请通过官方渠道联系海艺AI工作人员进行认证。我们致力于保护每一位创作者的权益。 点击去认领
