// HOW-TO · SHORT-FORM

How to make YouTube Shorts with ChatGPT (2026): script in ChatGPT, render with a real generator, publish vertical

ChatGPT can't render video. The honest 2026 workflow for YouTube Shorts: script and plan in ChatGPT, render in a real generator, then caption and publish.

Last verified · 2026-11-04 · by Moe Ameen

The honest starting point for a "ChatGPT video generator for YouTube Shorts" workflow is a letdown that saves you an hour: ChatGPT does not render video. It is a language model with an image generator attached, so it writes the script, plans the vertical frame, and drafts the render prompts — but there is no button inside ChatGPT that turns a prompt into a moving 9:16 clip. OpenAI's own video model, Sora, was the thing that did that, and it was wound down in 2026 (the consumer app and website closed April 26, 2026, with the developer API scheduled to end that September), so there is no native ChatGPT video generation to fall back on. Any 2026 tutorial that shows a "make a Short" button inside ChatGPT is either stale or quietly handing the render to a different tool.

That still leaves ChatGPT as the fastest front half of a Shorts pipeline — the part before anything renders. YouTube Shorts are vertical (9:16), up to three minutes, and won or lost in the first two seconds, and ChatGPT is genuinely good at the hook, the spoken script, the shot list, and the prompts you feed a real generator. This guide walks the whole thing end to end for Shorts specifically: set the boundary, brief ChatGPT with the format, get a hook-first script, turn it into a vertical shot list, pick a render route and generate the prompts, render the clip, then caption, reframe, and upload it as a Short. For the general version across all video, see [how to make videos with ChatGPT](/how-to/how-to-make-videos-with-chatgpt); for a long-video source, [how to make a YouTube Short from a long video](/how-to/make-a-youtube-short-from-a-long-video).

The steps

Set the boundary: ChatGPT writes the Short, a render tool makes it. Be clear before you start so you don't prompt for a clip that will never arrive. ChatGPT handles script, structure, shot list, render prompts, and a thumbnail concept; it cannot output a moving vertical video, and with Sora discontinued there is no OpenAI consumer video model to lean on. Treat ChatGPT as the writer's room and storyboard, then hand the render to a tool built for it.
Brief ChatGPT with the Shorts format, not just the topic. The output is only as specific as the brief. Tell ChatGPT it is writing a YouTube Short: vertical 9:16, a hook that lands in the first two seconds, one single takeaway, and a target runtime of about 20–40 seconds (Shorts can run to three minutes, but shorter holds retention better). Give it the audience, the angle, and the one action you want a viewer to take. A one-line prompt gets you generic filler; the constraints are what make it usable.
Get a hook-first spoken script. Ask for the script as short spoken lines — under about 15 words each — with the strongest, most specific line first. The opening is the whole ballgame on Shorts: no "hey guys, welcome back," just the payoff or the tension up front. Have ChatGPT mark the hook line explicitly and keep the whole thing to your target length so you aren't cutting half of it at the edit. Our [ChatGPT video-script guide](/how-to/use-chatgpt-to-write-video-scripts) has a reusable prompt and the word-count-to-runtime math.
Turn the script into a vertical shot list. Ask ChatGPT to break the script into beats and tell you what fills the 9:16 frame at each one — the on-screen action or b-roll, the caption text, and any title card — plus where the subject or text sits so it stays clear of the Shorts UI that crowds the bottom and right edge. A storyboard table (beat, spoken line, visual, on-screen text, seconds) turns a wall of narration into something you can actually shoot or generate.
Pick a render route and have ChatGPT write the prompts. Three routes cover most Shorts, and each wants a different prompt shape, so name your tool to ChatGPT and ask for prompts in that format. For a faceless talking-head Short, an avatar generator like [HeyGen](/ai-tools/heygen) needs a clean spoken script. For cinematic or scene-driven b-roll, a text-to-video model like [Veo 3](/ai-tools/veo-3), [Kling](/ai-tools/kling-ai), or [Runway](/ai-tools/runway) needs a visual scene description per shot. If your source is an existing long video, you skip generation entirely and clip it instead.
Render the vertical clip in a real generator. This is the hand-off ChatGPT can't do. Render each beat from the plan, and set the output to 9:16 — generate vertical from the start rather than shooting 16:9 and cropping, which strands your subject. Avatar tools voice the script directly; text-to-video models build the scenes from ChatGPT's prompts. Expect to regenerate a shot or two: the render does exactly what the prompt says, so a vague prompt gets a vague clip.
Caption, reframe to 9:16, and cut it tight. A raw render is not a postable Short. Burn in captions (most Shorts are watched on mute and the words carry the hook), confirm the final frame is a true 9:16 with nothing important under the caption bar or the right-side buttons, and cut every dead second so the pace never sags. This post-production is entirely outside ChatGPT — it wrote the words, but sizing, subtitles, and the final cut are on you or your editor.
Upload as a Short, then reuse the script. Upload the vertical file; YouTube classifies a square or vertical video of three minutes or less as a Short automatically, so a stray horizontal bar or over-length cut quietly demotes it to a regular upload. Write a title and description that repeat the hook in words, and disclose AI where YouTube requires it for realistic synthetic content. Then don't stop at one: the same script ChatGPT wrote is also a TikTok, a Reel, a carousel, and a text post, so one scripting session should feed the week, not a single clip.

Common gotchas

ChatGPT cannot render a video, full stop. A "generate Short" button inside ChatGPT is either an out-of-date Sora integration or a third-party plugin doing the render elsewhere.
Sora is not a fallback. The consumer app closed April 26, 2026 and the API was scheduled to end that September — do not build a 2026 Shorts pipeline on it.
Aspect ratio decides whether it is even a Short. A horizontal or letterboxed clip gets standard distribution regardless of length; generate 9:16 from the start, do not crop 16:9.
The hook is the first two seconds, not the first ten. A ChatGPT script that opens with a greeting or setup loses the swipe before the point lands — lead with the payoff.
Captions are not optional on Shorts. Most views are muted, so an uncaptioned clip hides the hook ChatGPT wrote; burn them in, do not rely on the script staying audible.
Fact-check every specific ChatGPT writes in. It fabricates confident stats and product details exactly where you need them right, and those go on camera if you miss them.
Over three minutes and it stops being a Short. Keep the cut inside the Shorts window, and far shorter than that for retention.

Legal note

YouTube requires creators to disclose realistic altered or synthetic content — including AI-generated video, a synthetic voice, or a face that isn't the uploader's — through its disclosure control, and it labels some AI content automatically. A ChatGPT-scripted Short rendered with a real face delivering real information usually does not trip it, but an AI avatar, a cloned voice, or a realistic AI scene can. Rules and the realism line change, so confirm YouTube's current AI-disclosure and likeness policies before relying on this, and disclose when in doubt.

Where Kompozy fits

This workflow is a relay race: ChatGPT writes the script, a second tool renders the clip, a third burns in captions, a fourth reframes to 9:16, and a fifth uploads and schedules it — and you re-explain your brand at every handoff. The render step, the one you went looking for a "ChatGPT video generator" to do, turns out to be the single thing ChatGPT can't do at all. [Kompozy](/) closes the gap between the script and the posted Short: you give it the same idea you'd have briefed ChatGPT with, and it writes the script under a [Persona Brief](/glossary/persona-brief) that already holds your voice, audience, and banned words — then renders the vertical Short itself.

For Shorts specifically, that lands three ways. A [Persona Short](/glossary/persona-shorts) becomes a captioned talking-head clip delivered by a face-locked avatar — the faceless-narration route, rendered and subtitled in one pass instead of HeyGen plus a caption tool plus a reframe tool. A Listicle Video stacks title and body cards over a vertical clip for a punchy five-point Short. And if your raw material is a long video, [Clipped Shorts](/glossary/clipped-short) cut it into several vertical hooks with burned-in captions, each already sized 9:16. The caption and reframe steps that were manual in the relay are native here, and the 9:16 target that strands a cropped 16:9 render is the default.

Then Kompozy publishes — to YouTube plus the rest of the eight social platforms, blog, and email behind a per-post review gate where you sharpen a hook or fix a stat before it ships — and fans the same script into a carousel, a text post, a blog section, and a newsletter, so one scripting session becomes a week of Shorts instead of a single clip. Honest boundary: if you make a Short now and then and like directing each tool by hand, ChatGPT plus a render app gives you full manual control for less, and for a single photoreal hero shot a model like Kling or Veo still beats an avatar. Kompozy earns its place when the job is a recurring Shorts cadence rendered, captioned, on-brand, and published without you stitching the chain together every time — Starter ($199/mo for 5,500 credits) for a solo creator on a steady cadence, Pro ($499/mo for 18,000 credits) for high-volume multi-format output, Enterprise via Contact Sales.

Frequently asked questions

Can ChatGPT make YouTube Shorts?

Not on its own. ChatGPT writes the script, plans the vertical shots, and drafts the render prompts, but it does not render video — there is no native video generation in ChatGPT in 2026, and OpenAI's Sora was discontinued (app closed April 26, 2026). To make the actual Short you take ChatGPT's script to a real video generator, then caption, size to 9:16, and upload it yourself.

What is the best tool to turn a ChatGPT script into a YouTube Short?

It depends on the format. For a faceless talking-head Short, an avatar generator like HeyGen voices the script into a presenter clip. For cinematic b-roll, a text-to-video model like Veo, Kling, or Runway renders from ChatGPT-written prompts. A content engine like Kompozy renders the script as a captioned vertical Short and publishes it to YouTube in one pass, skipping the separate caption and reframe steps.

How long should a ChatGPT-made YouTube Short be?

A Short can run up to three minutes, but about 20–40 seconds usually holds retention best, so brief ChatGPT to write to that length rather than cutting a long script down later. Keep the whole thing one takeaway with the hook in the first two seconds; length past the point the payoff lands costs you the swipe.

Do I need to disclose that a Short was made with AI?

Sometimes. YouTube requires disclosure of realistic altered or synthetic content — an AI avatar, a cloned voice, or a convincing AI scene — and labels some AI content automatically. A real person delivering a ChatGPT-written script usually does not need it; a synthetic presenter or voice can. Check YouTube's current AI-disclosure policy and disclose when the content could be mistaken for real.

Can I batch a week of Shorts from one ChatGPT session?

Yes, and it is the point. Ask ChatGPT for several Short scripts on one theme in a single session, each with its own hook and one takeaway, then render and publish them on a cadence. The same scripts also convert to TikToks, Reels, carousels, and text posts, so a single scripting session can feed every platform rather than producing one clip.