FL Bernini-R Environment Check
Run this before you gamble 117 GB on Bernini-R
- report_json
The whole Bernini-R pack hinges on one fragile thing: the official ByteDance Bernini source, vendored inside the node folder, importing cleanly inside your ComfyUI Python environment. When that chain breaks, every Generate fails before a single frame is drawn. FL Bernini-R Environment Check exists so you find out in five seconds instead of after a 117 GB download.
It's a read-only diagnostic. No inputs, no model load, no VRAM. Drop it in your graph, queue it, and it returns a report_json string covering:
- Python version, whether torch sees CUDA, and your GPU name
- Installed versions of torch, diffusers, transformers, accelerate, huggingface_hub, safetensors, decord, imageio, imageio-ffmpeg, and ftfy
- Whether the pack's vendored Bernini code (
bernini.pipeline,bernini.cli,bernini.data_utils) imports without error
The report also renders as a text widget right on the node, so you don't even need to wire report_json anywhere. It's flagged as an output node, but honestly the widget is the point.
What to actually look for
Three lines tell you almost everything. torch_cuda_available: false means your torch build can't see the GPU - fix that before anything else, because the Loader hard-fails on device: cuda. Any package showing null means it's missing or unreadable; pip install it. And if any bernini.* import reports an error string instead of true, that's the vendored import chain breaking, usually because a dependency version doesn't match this ComfyUI install.
The report deserves one more read because it lists exact versions. Bernini is a Wan-2.2-derived renderer, so the diffusers/transformers combination matters more than it does for a plain SDXL node. If the Loader or Generate later throws something cryptic, this JSON is your first and best clue - it's the difference between debugging and guessing.
Install
ComfyUI Manager: search for ComfyUI-FL-BerniniR and install. Manual:
cd ComfyUI/custom_nodes
git clone https://github.com/filliptm/ComfyUI-FL-BerniniR.git
cd ComfyUI-FL-BerniniR
pip install -r requirements.txt
Restart ComfyUI afterward. This is a standalone pack - same author as ComfyUI_FL-Trainer, but it needs no other FL pack installed.
Run this node before you bother downloading the model at all. If the environment's broken, no amount of disk space fixes it. If it's clean, you just saved yourself an afternoon of debugging a 117 GB mistake.
Inputs (0)
No inputs
Outputs (1)
| Name | Type | Description |
|---|---|---|
| report_json | STRING | — |