L5 Generation Component (Standalone Testing Only)

For the lecture, use ../instructor_install.txt and its combined installer and
launcher. They manage memory when switching between Qwen and generation.

Requirements: Apple Silicon, macOS 14+, 16 GB RAM, 12 GB free disk and uv
(or Python 3.12/3.13). Internet is needed for the first setup.

From the extracted release folder:
  cd lecture.deckroom/demos/generation
  bash install_demo.command
  bash start_demo.command

Open http://127.0.0.1:7897. Choose speech or image generation and enter a prompt.
Stop any separate Qwen helper before generating on a 16 GB Mac.
Press Ctrl+C in Terminal to stop the server.

Models: MagpieTTS v2607 and FLUX.2 klein 4B (4-bit).
Inference runs locally, with network access disabled. Cached models are required
for offline use; missing files require rerunning setup online.

Setup and results: ~/.local/share/ailt_l5_generation
Model downloads: ~/.cache/huggingface/hub (or your configured cache)
Errors display the local log path. Keep caches when updating the lecture.
