Summary
Nate Herk demonstrates what may be the most complete end-to-end AI video production workflow yet shown publicly: he gives GPT-6 Astra a single open-ended prompt and the model autonomously plans, researches, gathers B-roll, writes a script, directs HeyGen Avatar V5 with an ElevenLabs voice clone, edits in HyperFrames, syncs sound effects and music, verifies the output frame-by-frame, and delivers a finished YouTube-ready video — all without human intervention beyond the initial prompt.
The video is particularly notable because the AI narrator (Astra itself, speaking through Herk’s avatar and voice clone) explains its own production process in real time, describing how it opened source posts, captured screenshots using computer use, organized footage, split narration into segments, and built edits around audio cues. Herk reveals the full prompt he used — notably open-ended, without step-by-step instructions — and discloses the approximate API cost at roughly $60 on fast mode.
The broader showcase embedded within the video also highlights third-party creators building with GPT-6 Astra: a 3,000-object Blender model from a train drawing, a Minecraft-style game described as one-shot, a Call of Duty-style shooter iterated through natural language between matches, and a Blender-compatible Manhattan world built street by street in Unreal Engine. Together, the examples illustrate how Astra’s computer use and long-horizon task capabilities are enabling a new class of autonomous creative production workflows.
📺 Source: Nate Herk | AI Automation · Published September 04, 2026
🏷️ Format: Showcase
