How AutoVidGen builds a video

You give a topic, a format and a one-line prompt. AutoVidGen writes a full script: a hook, chapters, and a closing call to action. The chapter breakdown is not cosmetic - it is the structure the rest of the pipeline reads. Each chapter becomes a scene, and each scene gets its own imagery and its own slice of narration.

Script writing runs as a background job. You can close the tab; it finishes without you and notifies you when it is done.

Every chapter gets an image generated for that chapter's content, in one of 11 animation styles - photoreal, anime, comic, watercolour, retro pixel, stick figure and more. Nothing is pulled from a stock library, so two videos on the same topic do not share visuals.

Generated images do not reliably respect an aspect-ratio instruction, so every image is checked and anything that came back in the wrong orientation is re-framed before it reaches assembly. Ken Burns pan and zoom is then applied so no frame sits still.

Where motion matters more than a still can carry, individual scenes can be upgraded to real AI video clips for 10 credits per scene.

The script is narrated by an AI voice you pick, at a speed you set. The recording is then transcribed back to get word-level timings - which is why subtitles land on the word rather than on the line.

Those timings drive 9 subtitle styles, from a plain classic caption to word-by-word pop-in and a karaoke fill, with key words highlighted in your colour. Most short-form video is watched muted, so this is not decoration.

Scenes, narration, subtitles, music and effects are composited into one MP4 - 1280×720 for landscape, 720×1280 for vertical. Music is mixed under the voice and levelled so the narration stays intelligible.

A thumbnail is generated alongside it, plus a title and description for the upload.

The finished video lands in a six-hour review window before it publishes. You can watch the final MP4, rewrite a chapter's script and re-record its narration, regenerate a scene image, or generate extra thumbnail variants - and then approve, or reject it entirely.

Those edits cost credits: 5 for an image, 3 for a chapter rewrite, 5 for a thumbnail variant. Every plan starts with 30. If you do nothing, the window expires and the video publishes on schedule - the default is fully hands-off.

Connect YouTube, TikTok and Instagram once by OAuth. A series then publishes on its own frequency, from twice a week up to twice a day, to whichever of the three you connected.

Nothing here needs a browser open. The generation, the render and the upload all run as server-side jobs on a schedule.