Troubleshooting

Hugging Face token errors

If the CLI reports that HF_TOKEN is missing:

• copy .env.example to .env
• set HF_TOKEN
• confirm your account has accepted the gated model access request

Memory pressure on 6GB GPUs

The saved experiments were run on an RTX 3060 6GB setup. The main practical findings were:

• max-side 256 is the safest stable starting point
• balanced / 320 / 4 steps can fail on heavier inputs
• hard does not solve memory pressure; it mainly changes edge post-processing behavior

CPU offload caveat

--cpu-offload was not reliable on the tested Windows setup because the stack hit backend/kernel failures during decoding. Treat it as experimental in this repository.

Pages and docs

To build the docs locally:

Set-Location .\docs
npm install
npm run docs:build