Reddit developer demonstrates voice-driven persistent canvas editing with LiveLoop
A Reddit developer said they built LiveLoop to let voice commands continuously modify a persistent visual canvas through direct OpenAI API calls, without Blender or MCP. The author reported that GPT-6.1 Luna, Terra, and Sol completed the recreated initial steps, while Luna sometimes failed to visibly update the canvas until a refresh or undo-and-redo action; the author said the cause remains under investigation. The demonstration used GPT-6.1 Sol, with ElevenLabs V4 Turbo replacing OpenAI Realtime WebRTC for voice input, and was recorded in one take apart from speeding up playback.
This is a firsthand community demonstration of a concrete AI workflow for persistently editing one visual project through voice and API calls, including observed live-patching limitations, but it is based on a single Reddit post rather than independent testing.
#LiveLoop#OpenAI#GPT-6.1 Luna#GPT-6.1 Terra#GPT-6.1 Sol#GPT-6 Astra#ElevenLabs V4 Turbo#Claude Haiku 5.5