BC-250 AI Studio adds a deterministic proposal engine for LLM automation
The Vulkan-based local image generation platform for AMD's BC-250 board adds a deterministic proposal engine for LLM automation and prompt composer docs.
The BC-250 AI Studio platform for the AMD BC-250 board adds a deterministic proposal engine for LLM automation, alongside documentation updates for the prompt composer.
The project provides a Vulkan-backed local image generation setup with a FastAPI backend, SQLite queue, and a React-Vite UI. The recent commit implements automation logic while keeping the default configuration on a fake engine and manual policy.
To run it locally for development on CPU:
uv sync --extra dev --locked
export STUDIO_API_KEY="$(python3 -c 'import secrets; print(secrets.token_urlsafe(32))')"
export STUDIO_ADMIN_KEY="$(python3 -c 'import secrets; print(secrets.token_urlsafe(32))')"
export STUDIO_POLICY=image_priority
./scripts/dev.sh
Real GPU usage on the cut-down PS5 APU requires an LLM admission gate, service allowlists, and proper polkit verification before switching away from the mock engine.