AI image slideshow Shorts on YouTube: what narrative to add

A run of AI images with music and no story is the pattern YouTube's policy names. YouTube's inauthentic content policy (read 2026-10-03) lists image slideshows with minimal narrative or commentary as not allowed for monetization. Add a spoken or written argument that carries from the first image to the last, and let the stills illustrate it rather than replace it.

The same October 2026 originality update that cut reach for re-uploads is reported to include template-based changes among the things that do not count as original. Slideshows are not mentioned in that coverage, so treat the policy page as the source and the news as context.

What counts as narrative here?

The page does not define it, but its allowed examples point the way: a series with distinct storylines, focus or concepts per episode, and a similar format with material variation that delivers creative or educational value. In practice a Short has narrative when something is claimed, shown and resolved, and the images change what you understand as they go.

How do you build one with Sume?

Start from the script. Decide the three or four beats, then make each visual for its beat. For stills, Sume's Image API makes images from prompts; for motion, Video generation takes a text prompt or first and last frame images and returns a short clip, and the Timeline joins pieces and audio into one render. Moving a still into a few seconds of motion is usually the quickest way to leave the slideshow pattern behind.

YouTube's Shorts page (read 2026-10-03) says Shorts run up to 3 minutes, vertical, with a maximum resolution of 1080p, so there is room for a real argument.

How do you add on-screen narrative lines?

Video captions can burn authored text without speech recognition: pass cues with text, start and end, and no ASR runs. That makes it usable for a silent sequence, where a speech-based caption job would fail as caption_no_speech. The video must be a public HTTPS URL. This request burns two authored lines:

curl -X POST https://api.sume.com/v1/video-captions \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "video_url": "https://example.com/clips/batch-3.mp4",
    "style": "slam",
    "cues": [
      { "text": "Batch 3 collapsed", "start": 0.0, "end": 2.0 },
      { "text": "Here is the crumb at 4 hours", "start": 2.0, "end": 5.0 }
    ]
  }'

What is a quick test before you publish?

Mute the Short and read only the on-screen text. If the lines tell a viewer what happened and why, there is a narrative. If they only name what is in each picture, there is not. Then unmute and ask the opposite question: if the picture were missing, would the voice still carry the point? A Short passes when each channel, text, voice and image, adds something the others do not.

Also check variety across your channel. A series of Shorts with identical pacing, identical caption style and different stock images is a template, and the inauthentic content page lists generic or unoriginal templates for AI content as a problem. Vary the argument, not just the pictures.

Length is a lever you control. YouTube's page allows Shorts up to 3 minutes, but a story that takes 25 seconds should take 25 seconds. Padding a thin idea with more stills makes the slideshow pattern worse, not better.

Caption text you write is only narrative if it says something. A line that names the image is a label; a line that states what changed between images is a story. Burned captions are not an originality fix by themselves, and YouTube lists caption creation among edits that need no AI disclosure, which is a disclosure matter, not an originality one.

• Write the script first; make images for each line, not the other way round.
• Use generated clips for the moments where something moves or changes.
• Give every Short its own subject and conclusion.
• Disclose realistic AI scenes as YouTube's disclosure page asks.
• Do not rely on a slideshow template with swapped images.