Shorts series: serial or non-serial for AI episodes
Shorts series can be serial (oldest first) or non-serial (newest first). Which suits AI episodes, and how Sume jobs differ for each, per YouTube Help.

Choose serial when each episode depends on the one before it, and non-serial when any episode can be the first a viewer sees. YouTube Help says a serial series lists episodes oldest to newest and a non-serial one lists them newest to oldest, and every episode must be portrait or square and at most 180 seconds. For AI episodes the choice decides how much continuity you have to pay for in each render.
What YouTube documents
| Setting | What YouTube says |
|---|---|
| Episode length | Maximum 180 seconds |
| Orientation | Portrait or square |
| Playlist | Must contain only Shorts |
| Serial | Meant to be watched in order, listed oldest to newest |
| Non-serial | Watchable in any order, listed newest to oldest |
| Seasons | You set a season number when adding videos; the default is Season 1 |
| Thumbnails | A custom thumbnail can be uploaded for each Short in the series |
| Discovery | A viewer who lands on an episode in the feed sees a watch-series button and can swipe to the previous or next episode |
Serial: pay for continuity
A serial story means episode 5 has to start where episode 4 stopped, and the same people have to look the same. Two Sume tools carry that. For a video model, pull the last frame of the previous episode with video frames and send it as the next job's first_frame in frame_images; the opening of episode 5 then matches the end of episode 4. For a Format, send previous_run_id on the next run so the agent continues the same conversation and keeps what it already made, as the Runs and results page describes.
Serial episodes therefore render one after another, or at least in dependency order. A bulk queue of 100 independent items does not fit well, because item 5 cannot wait for item 4's frame.
Non-serial: render in parallel
A non-serial series is a set of standalone episodes that share a bible: the same cast references, the same style text, the same length. Nothing needs the previous result, so a bulk queue with a high concurrency window is the natural fit, and a failed episode can be retried alone without touching the others. The viewer's entry point is any episode, so each one needs its own hook in the first seconds and no 'as you remember' lines.
How to decide
- If a viewer who skips episode 3 would be lost in episode 4, use serial.
- If you want to add episodes forever without a plot to maintain, use non-serial.
- If you are unsure, start non-serial. You can render later episodes with a continuing arc, but you cannot easily remove a dependency you already built into ten episodes.
- Whichever you choose, do not put the episode number in the video itself if you might reorder later; YouTube keeps the order, so the number can live in the title.
Sources
Related posts
More in Use cases
- Skincare routine clip at 3:4: Seedance 2.5, $6.98 for 12 seconds
Seedance 2.5 has no 4:5 ratio on Sume. Render 3:4 at 834x1112 for $6.98 (12 s, 720p) and crop to 4:5 for the feed. Prices at 480p and 1080p too.
- Slow-motion water splash prompt for Gemini Omni: what 'slow' does
Prompt a slow-motion splash in Gemini Omni: how to word the speed, why a 5-second clip is enough, and what 360p, 720p and 1080p cost on Sume.
- Sneaker drop teaser in 6 seconds: Gemini Omni timecodes and cost
A 6-second vertical sneaker teaser built from a product photo and two timecoded beats in Gemini Omni on Sume, with 720p and 1080p prices and a Veo comparison.
- Solar roof before and after clip: Wan 3.0 first and last frame, 7 s
Send the bare roof as first_frame and the finished array as last_frame to wan-3.0. A 7-second clip costs $0.88 at 720p on Sume. Rules for matching the photos.
Written by Sume