Pretrained models

[loretoparisi]

Thanks for this project, it's simply amazing. Any plans to share the pre-trained model(s)? That would be super helpful to compare it against CLIP, DALL-E, VQGAN, and these models ensemble combinations.

Thanks a lot

[chris-aeviator]

also Code :eyes:

[johnpaulbin]

@loretoparisi They already did some small comparisons with the original OpenAI Dall-E model in the paper:

With clip and VQGan, those specifically aren't designed to generate images, but I'm sure you've seen a google colab somewhere that combines those. I recommend instead checking out Guided-Diffusion and CLIP, it generates much more fine details than CLIP + VQGan could ever do.

[johnpaulbin]

but yes, their pre-trained models would be nice to have.

[loretoparisi]

@loretoparisi They already did some small comparisons with the original OpenAI Dall-E model in the paper:

With clip and VQGan, those specifically aren't designed to generate images, but I'm sure you've seen a google colab somewhere that combines those. I recommend instead checking out Guided-Diffusion and CLIP, it generates much more fine details than CLIP + VQGan could ever do.

Thank you! For CLIP + Guided Diffusion I'm using the 512x512 patch with unconditional ImageNet diffusion here.

Didn't test it intensively actually on free Colab with K80 is pretty much slow, at least a V100 is needed to have a reasonable 10 minutes computing time...

Yes having a Colab with pretrained is the omly way this models cam be compared by ml practitioners.

[chenfei-wu]

The model and code will be made public in the near future. But recently we need time to clean up the code to make it simple and easy to reproduce. In addition, we also need to follow the company's public process. Thank you all for your understanding😊.

[loboere]

please add colab

[DexiangHong]

The model and code will be made public in the near future. But recently we need time to clean up the code to make it simple and easy to reproduce. In addition, we also need to follow the company's public process. Thank you all for your understanding😊.

Looking forward to the code!

[Sierkinhane]

Looking forward to the code TOO !😊

[josephrocca]

For the sake of those subscribed to this thread for updates from the authors, it would be nice if there were fewer +1/"me too" type comments. Everyone is clearly excited for the code and pretrained models! :)

[dcsrgh]

What a magical work. Waiting the code.

[You-SB]

666666

[YashVardhan-AI]

i want codeeeeee

[myname1111]

When can we expect the code to be released in the worst case scenario?

[YashVardhan-AI]

i would like the code as soon as possible i honestly wanna see how good their claims are

[afiaka87]

https://github.com/lucidrains/nuwa-pytorch

[YashVardhan-AI]

https://github.com/lucidrains/nuwa-pytorch

yes lol we get another guy(he is amazing btw) release the code before the official code it released nice

[VictorSanh]

+1 and bump: would love to see public models and code of NUWA-infinity! on the hugging face model hub (https://huggingface.co/) for instance :)

[loretoparisi]

+1 for pre-trained models on HuggingFace

[Maxlinn]

still waiting, any progress?

[myname1111]

still waiting, any progress?

Just. Give. Up Text to image: https://github.com/CompVis/stable-diffusion Text to video: https://github.com/THUDM/CogVideo

[Maxlinn]

still waiting, any progress?

Just. Give. Up Text to image: https://github.com/CompVis/stable-diffusion Text to video: https://github.com/THUDM/CogVideo

those repos seem interesting, thank you!