I gave up on Blender two years ago. With Opus 5.5 I have a research-to-video pipeline and a first episode out.

I've had this concept for years: short, calm explainers about what's changing on the internet, built from paper, with every claim sourced. I tried learning animation and Blender a couple of years ago and gave up. This weekend I set the whole pipeline up with Claude Code, and now when I read something fascinating I can turn it into an episode. How it works: • Research first. Claude builds a claims ledger: every fact with its source, date and a note on any arithmetic. A second pass with a fresh context checks the claims, and a "devil's advocate" pass tries to knock the story down. Nothing without a source goes in the script. • Script and storyboard, with each line tagged to its claim ids. • Images: Claude writes the prompts, I generate the pictures myself (a paper diorama look, separate layers so things can move). • Animation is code. Every frame is drawn on a Skia canvas from the timeline and piped to ffmpeg, so a new voice take re-times the whole video. • Voice is TTS for now, music is generated, mixed and loudness-checked automatically. A QA script transcribes the result and compares it to the script. • The sources page for each episode is a static site Claude generates and pushes to GitHub, which auto-deploys. What I do: pick the topic, direct the look, generate the images, review every cut, and publish. What Claude does: the research system, the code, the renders, the checks. Happy to answer questions about the pipeline, and honest about where it still needs a human.