New here? You are in the right place — this page is where accounts start,
in Japan and overseas. If you already use our shared ComfyUI, support lives at
aicu.jp/comfy.
How it works
Open it. A ComfyUI instance starts in your browser. Nothing to install.
Pick a workflow. Verified combinations of models and custom nodes, ready to run.
Generate. On an NVIDIA RTX 4000 Ada with 20 GB of VRAM.
What we run
Video
MiniMax H3 (Ref2VA / FL2VA), with and without the 4-step Turbo LoRA
Image
SDXL, FLUX, and the models the Japanese creative-AI scene actually uses
GPU
NVIDIA RTX 4000 Ada, 20 GB VRAM
Licensing
Checked per model. Only redistributable weights go into our public mirror
Why we publish our numbers
We are building ComfyLTS, a long-term-support
ComfyUI package. Upstream moves every day, and what worked yesterday can break today. So we measure —
timings, VRAM, what the model actually does with a prompt — and we publish it, including the runs
that failed.