AI character series on Shorts: avoid the same situation each time

Keep the character, and change the problem and the result each episode. YouTube's inauthentic content policy (read 2026-10-03) lists characters in identical situations with the same outcomes across videos as not allowed, and allows series with distinct storylines, focus or concepts per episode. A recurring AI character is therefore fine; a recurring AI character who does the same thing and gets the same ending in every Short is the pattern the page describes.

Sume gives you a way to hold the look steady while the writing changes. The writing is the part YouTube cares about.

What can stay the same and what must change?

The policy page explicitly allows a consistent intro and outro with varied core content, so branding, a character's look and a fixed opening are safe to repeat. What has to differ is the core of the video. The Shorts originality update that YouTube's Creator Liaison announced on October 1, 2026 is, as reported by Relevant Audience, aimed at re-uploads and template-based bulk changes, so a series built by swapping one detail on a fixed scene also sits close to what was named.

How do you keep a character consistent in Sume?

For a talking presenter, Sume's Avatar 1.0 turns a ready avatar and a script into a talking video, so one avatar can deliver a different script each episode. Avatar 1.0 is English-only in current code. For a character in a scene, Video generation accepts reference images, and the model catalog at GET /v1/videos/models lists which models take input_references or first and last frame images.

The script below keeps a shared base description and varies only the episode's problem and cause. It submits three jobs and prints their ids; poll each at its polling_url as in the docs.

import os
import requests

key = os.environ["SUME_API_KEY"]
base = "A vertical 9:16 clip. Same baker, same kitchen. "
episodes = [
    "Problem: the loaf spread flat. Cause shown: overproofed dough poked by a finger.",
    "Problem: a gummy crumb. Cause shown: sliced while hot, steam escaping.",
    "Problem: pale crust. Cause shown: oven door opened early, temperature dip.",
]
for i, ep in enumerate(episodes, 1):
    r = requests.post(
        "https://api.sume.com/v1/videos",
        headers={"Authorization": f"Bearer {key}"},
        json={"model": "sume/auto", "prompt": base + ep,
              "aspect_ratio": "9:16", "duration": 8},
    )
    print(i, r.status_code, r.json().get("id"))

Notice what the code does not do: it does not change one noun in a fixed prompt. Each episode states a different problem and a different cause, which is the difference between a series and a template. If you cannot write three distinct episode briefs without repeating yourself, the concept has more room in one Short than in a series.

How do you plan the episodes so they differ?

Plan on paper first. Write one line per episode in three columns: the problem, what causes it, and what the viewer can do differently afterward. If two rows have the same answer in any column, merge them or rewrite one. Three genuinely different episodes are better for a channel than ten that are one idea with a changed prop.

Then decide what the character is for. A presenter who explains is different from a character who is the subject of a story. Explainers vary by topic; stories vary by event and consequence. Either way, the thing the viewer takes away should be new each time, and that is what the policy's wording about distinct storylines, focus or concepts per episode points to.

Does the character need an AI label?

It depends on realism. YouTube's disclosure page (read 2026-10-03) asks for disclosure when content makes a real person appear to say or do something they did not, or shows a realistic scene that did not occur. It lists non-realistic content, such as someone riding a unicorn through a fantastical world, as not needing one. A clearly stylized or animated character generally falls on the non-realistic side; a photoreal presenter does not. The page says disclosure does not restrict reach or monetization eligibility, so labeling is not a cost to the series.

• Write an episode brief with a different problem, cause and result before generating anything.
• Reuse the look, voice and opening; vary the core.
• Read episode endings side by side; identical ones are the warning sign.
• Label realistic AI characters as YouTube's page asks.
• Keep series short enough that each episode earns its place.