AI Video Automation Lessons: Why Horror Shorts with Claude Code and Google Flow Stalled

On October 17, 2026, digital creator Ramz (@ramztd) shared an open post-mortem on X detailing an automated short-form AI video experiment that combined coding assistant Claude Code with Google Flow Veo for ghost and horror lore content, breaking down the reasons behind its lack of traction.

Image source: X @ramztd

While urban legends and paranormal topics appeared to be an inexhaustible, evergreen niche naturally suited for static visuals, real-world distribution plateaued quickly. Across TikTok, typical video impressions stagnated around 100 views, with the channel's top-performing release peaking at just 400 views. Deeming the channel unviable in its current form, the creator halted production and analyzed the failure across three core structural factors.

Pipeline Architecture and Iterative Format Experiments

The published pipeline decoupled generative tooling across logic scripting and video synthesis.

• Claude Code as the Primary Driver: Claude Code orchestrated the end-to-end production scripts and overall automation workflow.
• Selective Google Flow Veo Video Generation: Rather than generating every shot dynamically, Google Flow Veo was deployed selectively only for specific cuts requiring actual motion.
• Runtime Duration Adjustments: The creator tested multiple pacing formats, starting with 44-second concepts, expanding to 99 seconds, and eventually adjusting to 66 seconds to identify an optimal balance.
• Visual Staging Evolution: Transitioning away from purely static 2D image slideshows, generative video clips produced with Google Flow were blended in to alleviate visual monotony.

The Three Decisive Pitfalls Identified in Self-Diagnosis

Despite systematic pipeline automation and repeated formatting adjustments, the decisive bottleneck remained fundamental: the resulting videos failed to entertain. The creator identified three key issues in retrospective self-analysis:

• Weak First-Scene Hook: The opening frames failed to establish immediate immersion or effectively hook viewer attention from the start.
• Lack of Narrative Forward Momentum: Although individual cuts and scenes connected sequentially, the videos lacked narrative tension or a compelling reason compelling viewers to keep watching until the end.
• Absence of Authentic Chills and Emotional Stimulation: Despite centering on horror lore, the presentations were simply not frightening and failed to generate the suspense or emotional charge essential to the genre.

Performance Metrics and Practical Takeaways for AI Shorts

Operational data and insights from running the automated series provide valuable takeaways for short-form AI creators:

• Audience Metrics Ceiling: The channel's strongest release—an early video exploring the Japanese folklore giant skeleton spirit Gashadokuro—achieved only 400 TikTok views, while standard uploads consistently stalled near 100 views.
• The Threshold of Genre Execution: Horror suffers when executed half-heartedly. The takeaway was clear: either lean fully into genuinely terrifying, high-impact staging, or steer the creative angle in an entirely different direction.
• Strategic Pivot Across Platforms: Production on this concept was discontinued across YouTube, TikTok, and Instagram, with plans to refocus on a completely new subject.
• Story and Subject Over Pure Automation: As echoed in community responses, genre success depends more than half on compelling narrative discovery and directing choices rather than the efficiency of the automation pipeline itself.

Original source

• Ramz (@ramztd) Official X Post: AI Horror Video Automation Post-Mortem and Lessons Learned