Video models need far more compute than image models. Small models run on a 16 GB card, while large ones look noticeably better on 48 GB or more. Start with a short clip and measure the time.
Wan 2.1 1.3B needs 16 GB or more. Wan 14B and HunyuanVideo at good quality need 48 GB; smaller cards can offload, but generation is much slower.
ComfyUI: Image and video generation: FLUX, SDXL, Wan, HunyuanVideo. Nodes and ready workflows.
In short
Video generation needs a card with 16 GB or more of memory and disk space for the model weights. On gpu.nz such servers start from 11 ₽ per hour; generate a short test clip first and time it.
One hour of video generation: about 11 ₽
Estimate at the current per-GPU hourly price. Billed per minute, the price is fixed at launch.
Prices and availability come from the live catalog and update automatically.