Generating a Seinfeld Short with AI: A New Attempt at Continuous Video
nathandreamfast · reddit · 2026-08-04
A developer conducted a long video generation experiment, creating a 3-minute Seinfeld-style short. The script was entirely written and plotted by the GLM 5.2 model.
Technical Implementation:
- Workflow: A Python script uses the ComfyUI API to chain scenes by feeding the final frame of the previous video as the starting frame for the next one.
- Post-processing: FFmpeg is used for fade transitions and combining the clips.
- Efficiency: Generating this 3-minute clip took about 18 minutes.
The author notes that while speech consistency and scene generation require further refinement, this proves that creating full-length TV episodes or movies using current toolchains is technically feasible with enough patience.
More from Multimodal
- Research: Frozen Pixel-Space Diffusion Models Can Self-Guide — nanyang-technological-university-singapore · 2026-08-04
- Black Forest Labs Launches FLUX 3: A Unified Multimodal Model for Image, Video, and Audio — dl_weekly · 2026-08-04
- MiniMax H3 Released: Compact Transformer with Text-to-Video Generation — beatlepol · 2026-08-04
- Seedance 2.5 Short Film 'The Paineaters': One-Take Style — Internal_Ad_4036 · 2026-08-04
- MiniMax H3 Generated Video: Breaking Gooner — Secure-Message-8378 · 2026-08-04
- Wan 3.0 Leak: Native 1080P and 30-Second Video Generation Coming — bdsqlsz · 2026-08-04