CogStudio x CogVideo Image-to-Video (Research Test Videos)
Added 2024-09-23 11:52:46 +0000 UTCFor this example, we tested out using CogVideo through CogStudio on an RTX 4090. The 6 second generated video clips were made with one input image that was previously generated in Flux using Forge UI. The average render time takes 14 minutes per clip which isn't terribly slow, but not speedy either. The results are impressive for running locally but probably will benefit from upgrading to a ComfyUI workflow to make finer adjustments and control generations more.
Generation Settings:
Prompt: a high-action movie scene featuring two characters riding a motorcycle, set against a desert backdrop. The man in the front is confidently driving the motorcycle. The woman behind him holds onto a large gun and him while wearing sunglasses, looking fierce and focused. Explosions are happening in the background, with fireballs and clouds of smoke rising into the air. A car speeds on the road behind them, and a helicopter hovers in the distance.
Strength: 0.8
Inference Steps: 50
Guidance Scale: 6
dtype: bfloat16
Seed: -1 (random)
Since this test came out so well though, we'll be releasing an easy installer and guide soon on using CogVideo so stay tuned.
