ComfyUI Tora Text to Video Workflow

Stability AI dropped Stable Diffusion 3.5 on October 22, and ComfyUI already has it running. I gave both versions a spin — the Large and Large Turbo — and here’s what stuck out.

• SD3.5 Large: 8B params, handles up to 1MP images.
• Large Turbo: Same backbone but distilled for speed. Does decent output in just 4 steps.

There’s also a Medium version coming October 29, tuned for consumer GPUs at 2.6B params. If you’re on older hardware, that one might be your move.

You can grab the Large and Turbo models straight from Hugging Face (Large, Turbo).

What’s Different This Time?

The big win is flexibility. You can fine-tune without wrestling the model, and it handles diverse skin tones and features without needing a novel-length prompt. Styles? 3D, paintings, line art — it’s all there.

The Medium version is the sleeper hit, though. It’s not out yet, but Stability claims it balances speed and quality better than other mid-sized models. We’ll see.

Licensing: No Gotchas

Stability’s keeping it simple:

• Free for non-commercial use (personal projects, research).
• Free for commercial if you’re under $1M/year.
• Over $1M? Talk to them about an enterprise license.

ComfyUI Already Supports It

No surprise, ComfyUI added same-day support. Just drop the model into your ComfyUI/models/checkpoints folder and use the “missing model” feature to auto-load it. No coding needed.

For workflows, check out ComfyUI’s advanced refine workflow. It’s a solid starting point.