YouTube Shorts series: a weekly AI presenter episode pipeline

A weekly AI presenter series is one avatar, one script per episode, a draft render at standard, a first-frame check, and a final at max. YouTube's Shorts series feature, with seasons, episodes, custom thumbnails and sequential playback, began rolling out on 2026-09-23 across web, mobile and TV, per a platform-updates roundup (Orthotropy, read 2026-10-06). Series reward a consistent face and a steady cadence, which is what a reusable avatar gives you.

This post covers the pipeline on Sume. It makes no claim about how YouTube ranks or recommends series; the source above lists only what the feature is and when it began rolling out.

What stays fixed and what changes each week?

How do you keep a week cheap?

Render the first frame before the full video. A preview call returns stills for approval, and you can regenerate only the stills until the composition is right. When it is, call generate on the preview id; the optional quality there overrides only the final render tier, and preview stills are tier-independent, so approving then upgrading or downgrading does not need a new preview. Structural changes such as a new script, avatar or scene still need a new preview.

Use one Idempotency-Key per episode and per attempt. A key that encodes the season and episode, plus a revision counter, means a retried request cannot create a duplicate render, and a different payload under the same key is rejected rather than silently reusing the first.

What can go wrong on a Shorts-length episode?

The draft-then-final post goes deeper on tier choice, and the first-frame post walks the preview flow. Once the episode is final, upload it through your normal YouTube process; Sume does not publish to YouTube, and you should check the platform's own series rules for how to group uploads.

• Over 60 seconds. Sume rejects an estimated duration longer than 60 seconds, and inline captions have the same ceiling. Cut the script, or split it into two episodes.
• Caption mismatch. If the caption stage fails, the job can still succeed with a clean video and captions.status failed; check it before you upload.
• Korean episodes. Latin-style caption presets are rejected for Hangul scripts; use a Hangul style instead.
• A drifting look. Reuse the same avatar handle and scene reference; do not recreate the avatar mid-season.

What does a four-week plan look like?

In week one, create the avatar and render a draft at standard for the pilot episode, check the first frame, and settle the scene. In week two, move to the production tier and render episode one from the approved preview. In weeks three and four, repeat with new scripts and the same handle, keeping a log of each Idempotency-Key and the clip it produced. That log is your audit trail if a render needs redoing.

Keep scripts about 120 to 150 words for a 45 to 60 second clip at a normal pace, and keep the first sentence a clear hook. The documented limit is 60 seconds per avatar job, so plan the series around that unit and split anything bigger into parts.

Price one episode before you commit to the cadence. The Sume catalog read on 2026-10-06 lists avatar creation at a flat $0.95 per avatar, paid once for the season, and Avatar Video at $0.184 a second on standard, $0.245 on plus and $0.55 on max without a product image. For a 45-second episode that is $8.28 on standard and $24.75 on max, using the per-second rate alone; the catalog notes that clips are planned in chunks, so confirm the exact figure with the estimate on the preview. Because the preview stills are tier-independent, a standard draft followed by a max final of the same 45-second script comes to about $33.03, which is the number to weigh against a max-only week plus one rework.