Just tested WAN2.2 Rapid-AllInOne – literally one file to run video generation, and even low VRAM setups can handle it.

The most annoying thing about messing with the original WAN 2.2 is having to download a ton of model files, and then your VRAM just keeps blowing up. This Rapid-AllInOne by Phr00t stuffs WAN 2.2, the accelerator, CLIP, and VAE all into a single safetensors file—load one file and you’re done.

After actually using it, what stood out most is that you only need 4 sampling steps and CFG set to 1. It handles both image-to-video and text-to-video generation, and supports modes like last frame and first-last frame. It’s also compatible with both 2.1 and 2.2 LoRAs. Just drag the workflow into ComfyUI, and if you see any red nodes, hit Manager to install the missing ones—pretty standard stuff.

Switching modes is controlled by bypass nodes: for I2V, bypass the end frame; for T2V, bypass the start frame too and set WanVaceToVideo strength to 0. Even low-spec machines can get results, but don’t expect to max out image quality. If something messes up, it’s usually because the input image resolution is too low or the prompt is too vague.

4 steps and you’re done, they really went all-in on this speed boost.

Low VRAM gang rejoice, finally no more staring at OOM errors every day.

Single file is honestly so nice. Used to take me half a day just to download models.

For T2V you gotta bypass the start frame too, I didn’t pay attention the first time and it totally blew up.

24GB can run it, but the dev recommends 48GB. The middle tier is the most awkward spot.

Yeah, the quality definitely won’t be as good as the full-fat original version, but it’s all about that speed.

Phr00t’s latest integration really hit the spot, solved a major pain point.

I’ve been using the first-and-last-frame mode for transitions, and it works pretty well.