AI Faceless Video Generator

Give it a subreddit, a story, or a topic and it returns a finished faceless short: an AI-written script, an AI voiceover, cinematic scene images with ken-burns motion, and word-synced captions burned in. It runs the whole pipeline in one go, so the output is a ready-to-post vertical video for TikTok, Reels, or YouTube Shorts. Built for people churning out faceless content who don't want to stitch five tools together by hand.

How it works

It pulls the source text (top Reddit post via RSS, or your own story/topic), rewrites it into a tight retention script, generates the voiceover with OpenAI TTS, transcribes that audio back to get per-word timings, generates one scene image per ~6 seconds with gpt-image-1, then animates the stills with ffmpeg ken-burns and burns the captions over the top.

Input

Nothing is strictly required, but you do need to give it a source (one of subreddit, story, or topic) and an OpenAI key. subreddit is prefilled with tifu so a run with just a key will produce something. Run it with no key and it returns one labelled SAMPLE row showing the output shape instead of a video — that row goes to the run's key-value store rather than the dataset, and is not charged.

Field	Required	Notes
subreddit	one of source	Pulls the top post with enough body text, e.g. tifu, AmItheAsshole, nosleep. Prefilled tifu.
story	one of source	Your own text to turn into a video. Overrides subreddit.
topic	one of source	A topic to write about from scratch. Overrides subreddit.
openaiApiKey	yes in practice	Bring your own key. Used for script, TTS, transcription, and images. Kept private.
tone	no	Script tone: dramatic (default), casual, funny, emotional, educational, hype.
voice	no	OpenAI TTS voice: onyx (default), alloy, echo, fable, nova, shimmer.
aspectRatio	no	9:16 (default), 16:9, or 1:1.
captionPreset	no	Burned caption style: karaoke (default), hormozi, beast, tiktok, clean.
targetSeconds	no	Approximate length. Default 50. The scene count scales with the final voiceover duration.
imageQuality	no	gpt-image-1 quality: low, medium, high (default). Drives most of your OpenAI cost.
style	no	Art direction for the scene images. Defaults to a cinematic, moody look.
characters	no	Recurring-character description to keep faces and outfits consistent across scenes.
sort / time	no	Reddit sort (top/hot/new/rising) and time window for top.

Output

One dataset item per run with the finished MP4 and the metadata around it. The video and the raw voiceover MP3 are written to the run's key-value store. Key fields: videoUrl and videoKey (the MP4), title, hook and altHooks (extra hook options to A/B), script, voice, scenes, aspectRatio, durationSeconds, and source (where the story came from).

Example

{
  "subreddit": "tifu",
  "tone": "dramatic",
  "voice": "onyx",
  "captionPreset": "hormozi",
  "aspectRatio": "9:16",
  "targetSeconds": 50,
  "openaiApiKey": "sk-..."
}

Pricing

$0.69 per finished video, plus a run-start fee of $0.0000125 per gigabyte of run memory — this actor runs at 4 GB, so $0.00005 a run. Flat rate: no volume tiers, no plan gates, no subscription.

That price takes effect on 26 August 2026. Runs before that date bill the previous $0.39 per delivered video; every run from 26 August 2026 onward bills $0.69. The run-start fee is unchanged.

Only a short that actually finished rendering costs anything:

• a keyless run writes one labelled SAMPLE row to the run's key-value store, not the dataset, and is free;
• a run that fails outright (bad subreddit, no usable post, an OpenAI or render error) is free;
• you are charged once, and only once, per delivered .mp4.

OpenAI usage (script, TTS, transcription, images) runs on your own key and is billed to you by OpenAI directly — none of it is resold through this actor, so what you pay here is the orchestration and the render, not a markup on model inference. imageQuality is the main lever on the OpenAI side.

Notes

Reddit serves the public RSS feed, which works for top posts but not comment threads, so this reads the post body only. imageQuality: high is the main lever on OpenAI cost and run time. If your story or topic is short, the script model pads to roughly targetSeconds, which can read thin; feed it more source text for better results.