Text-To-Video-Finetuning VS text-to-video-synthesis-colab

I used this repo for the finetuning: https://github.com/ExponentialML/Text-To-Video-Finetuning

Has anyone successfully done this? I walked through the steps and did not find what I wanted so wanting to know if anyone has a tutorial about fine-tuning Modelscope with https://github.com/ExponentialML/Text-To-Video-Finetuning

Well, actually today all I have is ModelScope txt2video, SadTalker and an understanding of how this technology works, but pretty soon I'll have this https://github.com/ExponentialML/Text-To-Video-Finetuning/pull/27 too. Then whatever advancements things like https://ai.facebook.com/blog/dino-v2-computer-vision-self-supervised-learning/ unlock will filter down to me too and so on it will go. My understanding of the tech will continue to deepen as I continue to retrain from traditional software engineering to machine learning. Things like Culitho (https://www.anandtech.com/show/18792/nvidias-culitho-to-speed-up-computational-lithography-for-2nm-and-beyond) and AlphaTensor (https://www.deepmind.com/blog/discovering-novel-algorithms-with-alphatensor) will continue to make compute faster and more affordable driving the cost of training/inference down and massively increasing the accessibility. Increasingly more functions will continue to be approximated closer and closer (https://www.youtube.com/watch?v=0QczhVg5HaI).

Made by strangeman3107 via https://github.com/ExponentialML/Text-To-Video-Finetuning. The original Diffusers weights https://huggingface.co/datasets/strangeman3107/animov-0.1

Just as one of Deforum Discord's server members linked me it, I was so inspired that I quickly wrote the Diffusers->pth (ModelScope original format) conversion script

Yes, there's a Diffusers based repo https://github.com/ExponentialML/Text-To-Video-Finetuning.

Github

A bit late on the spaghetti train, but this was made using a the colab link from Camenduru's text-to-video-synthesis-colab. I didn't do anything fancy to generate the video, just Michael Scott eating spaghetti. The only part that needed more help was when the model didn't understand the character's names, but in general using the actors' names with "at the office" seemed to solve the issue.