Boris Dayma
Boris Dayma
The main issue with PNG is that the size is much bigger so it increases a lot egress fees. There is also webp but don't think it's supported everywhere and...
Oh interesting, thanks for reporting it
It's not possible in the current model but maybe in the future
You need a GPU with 16GB and you can run it yourself
It's not possible with this specific model but may be feasible on a next architecture
It's not the same. What we need is an image encoder. We're trying to train a better one at the moment.
Maybe @chenxwh has it. I didn't want to put it in this repo because I don't want to have to maintain it but it can be somewhere else.
You may have a wrong reference. Make sure to follow the inference notebook
Oh I didn't try it but I would expect that if you run the notebook twice with the same seed value you would get the same images, no?
Yeah that's why I have the raw notebook in the repo. In fact the colab just pulls the file from this repo. I personally really like VS Code.