Tech • AI • Robotics • Game
I Told Claude to "Make My Video Better" 100 Times (It Went Too Far)
6/10TL;DR
A creator is using Claude Code with open-source tools to turn one raw recording into five short-form videos, cover images and cross-platform posts, cutting editing time while testing multiple versions for reach.
KEY POINTS
Five videos from one recording
The workflow starts with raw footage or a voiceover, then Claude Code generates five separate short-form edits in one run. The creator said the system now handles production that previously had to be edited manually, allowing one recording session to produce multiple posts and alternate cuts for testing.
Strong early distribution results
One AI-generated reel cited in the demonstration had passed 50,000 views, while the broader method was credited with helping drive nearly 2 million views in 30 days. A notable share of viewers were non-followers, suggesting the system is designed to optimize discovery rather than rely on an existing audience.
Template-based editing with a custom character
The process depends on feeding Claude an example of the desired format. From that template, the system recreated the visual style, including a pixel-art avatar built from existing photos, then matched scenes to a supplied voiceover to produce multiple finished versions automatically.
Posting automation across platforms
After rendering, the setup can publish directly to Instagram, TikTok, YouTube and Facebook, complete with captions and cover images. Extra versions can be routed to Instagram trial reels, a format aimed at non-followers, letting creators test slight variations without flooding their main feed.
Selective re-rendering saves time
A key part of the setup is the ability to re-render only one “beat” or scene instead of exporting the entire video again. If a graphic, map or transition looks wrong, Claude can replace just that segment, reducing revision time from a full re-export to a targeted fix.
Open-source stack behind the workflow
Several repositories power the system. Hyperframes adds motion graphics through code rather than a separate AI video model. Hypit creates templates from existing viral videos, helping users avoid starting from scratch. RVB Lip Sync aligns speech with animated or faceless characters. Higgsfield AI is used for image generation, including character sheets.
How Claude analyzes video references
To build templates from online videos, the system uses yt-dlp to download media and FFmpeg to inspect visuals frame by frame. That lets Claude extract layout, pacing and shot structure from an existing example and recreate a similar editing pattern with new assets and narration.
Research drives topic selection
The editing system is paired with a custom dashboard that tracks top-performing shorts, long-form videos, news, Reddit posts and GitHub repos over the past month. Rather than automating content blindly, the process begins by identifying formats, hooks and topics already proving effective online.
AI augmentation, not full automation
The approach emphasizes AI-assisted production over full replacement of human judgment. Topic choice, hooks, calls to action and brand positioning still require human input, while Claude handles repetitive editing, formatting and distribution. The result is framed as a way to expand output by 20 times rather than remove strategy from the loop.
Content can be repurposed beyond video
The same source material can be turned into multiple additional assets, including carousel posts. That means one sit-down recording can generate a batch of short videos plus several static or swipeable posts, extending the value of a single idea across formats and platforms.
CONCLUSION
The workflow shows how Claude Code can function as a production assistant, editor and publishing tool in one system. Its main advantage is not fully autonomous creativity, but faster testing, wider distribution and more efficient reuse of proven content formats.
Ask a question