Not enough VRAM for a stable production pipeline. Rent an RTX PRO 6000 by the hour and run everything.
We have not tested AMD or Apple Silicon. ComfyUI runs on them in some setups, but our workflows are validated on NVIDIA only. We are not going to guess.
Rent the exact machine
Same template we use to build and test every workflow. ComfyUI preinstalled, nothing to configure. Pay by the hour, stop when you're done.
Nothing is quantized. No fp8, no GGUF, no block swap. The workflows run the models at full precision because that is where the quality is. That costs VRAM, and we would rather tell you the real number than sell you a pack that OOMs on the first run.
How fast it actually is
One minute of finished video takes roughly two to three minutes of generation on an RTX PRO 6000. Raw generation, no upscale pass.