How it works

Turn blog posts into videos, automatically.

Point a workflow at your blog's RSS feed. Every new post becomes a short vertical video with generated visuals, a voice-over and captions — rendered on your own machine and published without you touching it. The workflow is nine visible steps you can edit at any point, not a black box.

This is one of the things people most often try to build and most often give up on, because the hard part is not any single step — it is keeping the timing between them in sync. Below is what the workflow actually looks like.

The nine steps

1. New in a feed. Give it the address — usually ending in /rss, /feed or .xml. Both RSS and Atom work; the two formats look the same from outside but are named differently inside, and reading only one of them is why half of all feed integrations silently return nothing. You get title, url, summary, published and an id.
2. Skip what was already handled. Without this, a workflow that checks every fifteen minutes republishes the same post forever. The memory lives with the workflow, so closing the app does not reset it.
3. AI analysis. The post becomes a scene list. Give the step a JSON schema so the shape is guaranteed rather than hoped for — a fixed number of scenes, each with the line to speak and a description of what should be on screen.
4. Split a list. One record per scene. Everything after this point runs once per scene automatically.
5. Generate image. One visual per scene, in 9:16. Keep any text out of the prompt — image models cannot set type, and asking for a caption returns a letter-shaped smear. The caption comes later, set properly.
6. Speech. One voice line per scene. The length of this audio is what decides how long the scene stays up, which is why the timing works without you managing it.
7. Collect into one. The scenes come back together into a single record. This is the step people forget; without it the render only ever sees the last scene.
8. Video from scenes. Renders one vertical MP4. Each scene stays up exactly as long as its own audio, with an optional camera move and fade.
9. Check the video, then publish. Aspect ratio, duration and audio are verified before anything leaves the machine. Then it goes to YouTube with title, description, tags and thumbnail — or the upload page opens with the caption already in your clipboard for TikTok, Instagram and X.

What it costs to run

Credits are spent only on the managed AI actions — the analysis, the images, the voice. Editing the canvas, rendering and publishing do not consume credits. A short video with a handful of generated scenes is a small fraction of the monthly allowance; plans start at 3,000 credits for €11.99 and top-ups start at €4.99.

Things that go wrong, and why

The same post gets published twice

The deduplication step is missing, or it is placed after the rendering instead of directly behind the feed. Put it second, so a post you have already handled never costs you a render.

Every scene shows the same image

The split step is missing, so everything downstream ran once on a single record instead of once per scene.

The video is much shorter than the audio

The scenes were collected but the audio was not, so the render had one voice line for the whole timeline. Collect both, or let the timeline read the ready-made scene list.

The scheduled run stops working after a week

That is not the workflow. If it publishes to YouTube, the sign-in expires after seven days until the Google verification for the app is complete.

Beyond blogs

The same shape works for anything that publishes a feed: a podcast, a news site, a changelog, a subreddit. Swap the feed address and adjust what the AI step is asked to write. For a YouTube channel there is a dedicated trigger that reads the channel's public feed without an API key.