A developer has demonstrated the creation of short-form video documentaries in approximately 30 seconds using an AI pipeline. The process leverages multiple AI models for script generation, image creation, and narration.

The pipeline utilizes GLM-5.2 fast via Fireworks for script and image prompt generation, Nano Banana 2 Lite for image synthesis, and gpt-4o-mini-tts for voiceovers. The final assembly and application of a Ken Burns zoom effect are handled by ffmpeg. A significant speedup in video compilation was achieved by running the process on a 64 vCPU EC2 instance.

This demonstration highlights the accelerating pace of AI-driven video generation. While the computational cost for the EC2 instance is substantial, the per-video cost averages around 25 cents, with image generation accounting for nearly 90% of that expense. This suggests on-demand AI video creation is rapidly approaching mainstream viability within specific content domains.