ali-vilab

/

text-to-video-ms-1.7b

like 482

Maybe it is because the model was trained with shutterstock images. All public available images from Shutterstock having this watermark. They should use a different data set. I think it is only a matter of time till a new version will arise without that watermark. I am curious if someone can "untrain" the AI model for that watermark. Maybe with control net?

Maybe it is because the model was trained with shutterstock images. All public available images from Shutterstock having this watermark. They should use a different data set. I think it is only a matter of time till a new version will arise without that watermark. I am curious if someone can "untrain" the AI model for that watermark. Maybe with control net?

Good thinking, maybe I'll look into that!

Can confirm I was able to remove watermarks by custom training a model using: https://github.com/ExponentialML/Text-To-Video-Finetuning
Also have found you can reduce some of the watermarking with negative prompting. Less successful than the finetune, but it does work.

Can confirm I was able to remove watermarks by custom training a model using: https://github.com/ExponentialML/Text-To-Video-Finetuning
Also have found you can reduce some of the watermarking with negative prompting. Less successful than the finetune, but it does work.

Thanks man!

Can confirm I was able to remove watermarks by custom training a model using: https://github.com/ExponentialML/Text-To-Video-Finetuning
Also have found you can reduce some of the watermarking with negative prompting. Less successful than the finetune, but it does work.

Hello @dvs could you give me some advice about how to finetune a T2V model on a custom dataset with aprox 1300 text - video paiers of sign language? What model would you use and where i can find come code of finetunning a video model? I am familliar with how a transformer work in nlp. Is is the same here? I can pass the text as input and the video as target and run the diffuser for n epochs?
I would appreciate if you could offer me some advices or maybe some tutorials.