add lora

[liemthanh-playgroundvina]

I want to add LORA: https://huggingface.co/latent-consistency/lcm-lora-sdv1-5 into the pipeline. How to do it??

[liemthanh-playgroundvina]

i'm using scripts at ./scripts/inference.py

[crystallee-ai]

In fact, we have fine-tuned the SD1.5 weights, so directly using the SD1.5 lcm-lora weights might not be suitable. If you insist on using them, you may consider fine-tuning them.

[liemthanh-playgroundvina]

Thanks for your assistance! Also, do you have other methods or suggestions you might have for increasing the interface's speed? I'm using 15steps that take 5 minutes for a 10s 512x512 video and it's too long to use.

[tonyabracadabra]

Thanks for your assistance! Also, do you have other methods or suggestions you might have for increasing the interface's speed? I'm using 15steps that take 5 minutes for a 10s 512x512 video and it's too long to use.

Is the result legit for 15 steps, what machine did you use for generation?

[liemthanh-playgroundvina]

Thanks for your assistance! Also, do you have other methods or suggestions you might have for increasing the interface's speed? I'm using 15steps that take 5 minutes for a 10s 512x512 video and it's too long to use.

Is the result legit for 15 steps, what machine did you use for generation?

I'm using 3090ti with cu118+py3.9, the result is quite good with a little noise in the background

[liemthanh-playgroundvina]

https://github.com/fudan-generative-vision/hallo/assets/128352834/644b242f-baca-48b2-b4dc-b3b86f9a46e3

[crystallee-ai]

We have received a lot of issues about inference time, and we have plans to optimize it, but not in the near future. Thanks!