Make a Video by Meta mentions (9)

• [P] Do you want to join a motley crew who are scaling/retraining AnimateDiff for open source? AD trainer code just released!

  - Meta's Make-A-Video had a very similar approach to this. What could we learn from them? Source: 10 months ago

• AI Generated Beer Commercial - Synthetic Summer. No real people featured

  I suspect that the model used for this video wasn't either of the tools listed in the text-to-video section there, however- it looks like the studio that made it probably got access to a model that isn't publicly-available yet, like Meta's Make-A-Video or Runway. Source: about 1 year ago

• Used ControlNet + EBSynth to get rid of some of the flickering and export a more consistent animation.

  This is what FAIR has been doing with https://makeavideo.studio/ . They have a stable diffusion model trained on image pairs in sequence. If this was adopted by the community more broadly as a basis you'd be able to resolve things like flickering across frames much more robustly. Source: over 1 year ago

• ControlNet Video Test (Video To Video With Pose Transfer)

  Yeah, I feel from looking at https://makeavideo.studio/ and https://imagen.research.google/video/ that it's getting there. Source: over 1 year ago

• Question – will video models be general or fine-tuned?

  We're seeing early examples like Google's Imagen and Meta's Make a Video – which look to be general, and might make sense when creating clips. Source: over 1 year ago

uberflip mentions (0)