Zhen Li comments

Results 197 comments of


                                            Zhen Li

How app. py performs parallel computing through multiple GPUs

Using this [link](https://github.com/TencentARC/PhotoMaker/discussions/114#discussioncomment-8348999) as a solution.

Does it support the way to generate images from images?

Hello, the current version can already support it, but you need to change the inference code.

Using two different face images in the same picture

Two faces in one image or one blending face?

Using two different face images in the same picture

Although the code we provide requires the user to enter one trigger word only, maybe you can try multiple trigger words with the corresponding photomaker embeddings to achieve this function....

Using two different face images in the same picture

Yes, you're right. Please feel free to contact us if there are any updates or questions.

Are lora residual weights critical to faithfully keep ID in the generated image?

Hi, based on our past experiments, LoRA residuals in our PhotoMaker are important.

Why is it so slow?

What is your equipment and environment. How long does it take approximately?

loss explosion when training on custom Dataset

Yes, the simple solution is resuming from a non-crashed checkpoint. stayhungry1 ***@***.***> 于2024年4月9日周二 15:06写道： > Is the loss increase from the DCN layer in the training > > — >...

Log the training and testing configuration on ubuntu 18.04 CUDA 10.2

👍 Thank you

CUDA OOM for ~1-2 min long videos

Thank you for your interest in our work. Processing 1000 frames at once is far beyond the GPU memory requirements. The best approach is to divide the 1000-frame video into...