rainsoulsrx

Hi, I use your model to run reference-guided generation. However, I give a strong 3D style reference, while the result seems to be an oil panting just like this: ...

finetune?

4

Hi, great work, I tested the model on some cases and find performance is not good in portrait image while good in common scene. I guess the reason maybe train...

About noise optimize in image2styleGAN++

6

Hi, thank you for your good work, I have a question, In paper image2styleGAN++, the author mentioned that they both optimize w and n(noise) , but in your code, I...

你好，在论文中关于训练数据，写的是直接crop乘512*512的尺寸进行训练，请问是否需要提前做reszie等处理呢？如果不做的话，原始图像分辨率很大，如果直接crop出来512的图像区域，是不是很容易会crop到没有语义的背景区域呢？这种无语义的背景区域是否应该参与训练呢，另外crop之后的图像怎么和原始的llava标注信息对应上呢？

Hi, in the paper, you use LLAVA to get more detailed caption, but SDXL only can take into 77 token. How to deal with the caption more than 77, just...

Colab？

1

您好，请问会计划放出colab测试吗。想看更多的case效果如何

Hi, cool work! In this picture, how can we get Jt? I guess it is the result of per frame editing by image to image editing with prompt "colorful painting"?...

what about the speed

1

Is control-lora more quickly than original control-net?

使用更大规模的数据训练，会有更好的效果吗？

17

你好，如题，请问有没有做过这样的实验，使用更大量级的训练集训练的话，会有更加显性的结果吗？和后面开源的[supir](https://github.com/Fanghua-Yu/SUPIR)相比，想知道supir的效果更多的是数据的贡献还是整体架构的贡献

overfitting on small dataset?

2

Hi, I use 500 human face images(only contain 6 id) to train this model, but when I test the trained data, the result is strange, I test using A people,...