python -m fastvideo.entrypoints.openai.api_server parsed the hand-written FastVideoArgs argparse flags, an inference input path next to the typed config that fastvideo serve reads. It is removed; start the server with fastvideo serve --config SERVE_CONFIG and dotted overrides such as --server.port 8000. The GPU integration test starts the server that way, with the same model, GPU count, and DiT offload. FastVideoArgs.add_cli_args stays for the legacy training and preprocessing scripts. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012M91wnVFmPEvJ9h39r5BH7