Type any topic and get a finished, narrated short documentary in about 30 seconds — a fully automated text-to-video pipeline.
Last verified · 2026-07-07 · by Moe Ameen
InstantVideos.org was a single-prompt documentary generator: you typed a topic, and about 30 seconds later you had a finished, narrated short video. It launched as a Show HN experiment on July 6, 2026 from an independent builder posting as "pw" who built it with Claude (the same builder behind the "By Claude" YouTube and TikTok channels), and it was intentionally minimal — no accounts, no editor, no settings beyond the format. Shortly after launch the creator wound it down, calling it "a short experiment" that "had its moment," and pointed visitors to a lighter sibling, InstantImages.org, which turns a prompt into a still image in roughly three seconds.
The interesting part is the pipeline, because it is a clean template for fully automated video. Under the hood InstantVideos chained four commodity pieces: GLM-5.2 (the fast tier, served via Fireworks) wrote the script and the per-scene image prompts; Nano Banana 2 Lite generated the images; gpt-4o-mini-tts narrated the script; and ffmpeg stitched the stills into a video with a Ken Burns zoom over each frame. The output is a narrated slideshow documentary — AI-generated still images with slow pans and a voiceover — not live-action or generative motion footage.
The economics were the headline claim. According to the creator, a short-form video cost about 25 cents to produce, and nearly 90% of that was the images at roughly 3.336 cents each. Hitting the ~30-second render time meant throwing hardware at ffmpeg — the demo ran on a 64 vCPU cloud instance. The pipeline also leaned on AI still images rather than a generative-motion model, which is why the output is a narrated slideshow rather than live video — motion footage priced by the second would have dwarfed the rest of the budget.
Because it is discontinued, treat InstantVideos as a reference architecture rather than a tool you can sign up for today. What it demonstrated — that a topic-to-narrated-documentary pipeline can be assembled from off-the-shelf models for pennies — is the durable takeaway. The specs above are reconciled against the creator's own Show HN post and the site's shutdown notice; the successor InstantImages.org (text-to-image in ~3 seconds) remained live.
InstantVideos handed you one thing: a finished narrated MP4 — a topic, a script, AI stills, and a voiceover, panned together. That is a strong starting asset and a terrible finished post, because a raw documentary clip is not sized, captioned, or on-brand for any specific feed. Kompozy is the layer that takes that MP4 the rest of the way. Drop it in and Clipped Shorts cuts the long-form version into vertical highlights; the captions engine burns in word-synced subtitles so the muted first second still lands; and HyperFrames reframes and stacks a branded hook so the same documentary reads natively on TikTok, Reels, Shorts, and a LinkedIn feed instead of looking like an anonymous slideshow.
The bigger gap InstantVideos left is that one topic produced exactly one output with zero brand identity. Kompozy turns that same topic into a content unit: the video plus native Text Posts, a Blog Article, an Email Newsletter, Quote Graphics of the standout lines, and a brand-exact Carousel — every piece governed by your Persona Brief and banned-word filters so the volume reads as your brand, not generic model output. And it generates the formats InstantVideos never could — Persona Shorts and HeyGen avatar video with a face-locked recurring identity so your documentaries have a consistent host. Then Kompozy schedules and publishes the whole set across nine social platforms plus blog and email from one queue, with Autopilot and a per-post review pipeline. InstantVideos proved the raw generation is cheap; Kompozy is the on-brand, multi-format, actually-published half that it deliberately skipped.
No. The creator described it as a short experiment and wound it down shortly after its July 6, 2026 Show HN launch, pointing visitors to InstantImages.org (text-to-image in about three seconds) instead. Treat InstantVideos as a reference pipeline rather than a live product.
It chained commodity models: GLM-5.2 fast (via Fireworks) wrote the script and image prompts, Nano Banana 2 Lite generated the images, gpt-4o-mini-tts narrated, and ffmpeg compiled the stills with a Ken Burns zoom. The creator reported a short-form video cost about 25 cents, roughly 90% of it images at ~3.3 cents each.
A narrated slideshow — AI-generated still images with slow Ken Burns pans over each one, set to a synthetic voiceover. It was not live-action or generative motion footage; the "documentary" was stills plus narration.
For the video itself, any text-to-video or narrated-slideshow tool fills the slot. But if your real goal is publishing, a content engine like Kompozy generates video, captions and reframes it per platform, fans the topic into blogs, carousels, and newsletters under your brand voice, and schedules it across nine platforms — the finishing work InstantVideos never did.