Spaces:
Runtime error
Not as expected - Sinister Video Output
Prompted "A promotional video with voiceover for the business, Go Green Home Supply, referencing the website gogreenhomesupply.com for brand story, why sustainable building materials are important, showcase images of sustainably built homes as part of the promotional video, video should be 50 to 59 seconds in length, professional, include the Go Green Home Supply logo as a watermark, and be compelling"
Output video was sinister, off-topic, and irrelevant.
This v1 model is trained on 400k videos, and has curently has proplems following prompts. it works sometimes, but not always. (2 seconds only)
To be honest - this is one of the worst v that I saw. That's not even close to sora. -rep to the ts
I mean it wouldn't have audio to be fair - I am not sure why audio was asked for in the prompt. But any prompt I put in gets me an angry looking chubby white dude in a martial art outfit standing in a school gym. I asked for nature scenes etc. I would have preferred the cool demon guy. I get the same video on any prompt.
Admittedly, I have no idea what I'm doing regarding prompts, and didn't know the limits of this tool. I'm just trying to learn a little bit and understand by playing around and asking to see what I can come up with. Either way - the tool is either broken or off the rails.
I guess there are some errors with this implementation, maybe it is because there are no GPU resources so it did not actually run the inference stage. Few weeks ago, I create this and it tooks 200s to see the output and the output looks fine. But now, it only takes 66s and the output are always the same so I guess it did not really generate from my input.
The open-sora model is a project that continues to develop. And I optimized this demo for everyone to test quickly. If you want to get high output, you can use the colab notebook file.
Notebook: https://github.com/camenduru/Open-Sora-jupyter
Replicate: https://replicate.com/camenduru/open-sora