YouTube Shorts series covers: one template, one edit per episode

To make a matching cover for every episode of a YouTube Shorts series, generate one template image with Ideogram 4.5 on Sume, then run one edit call per episode that changes only the episode number. The first image carries the look; every later call is an edit of that image, so the colors, type and layout stay put and only the digits move.

The trigger is YouTube's rollout of Shorts series, with seasons, episodes, custom thumbnails and sequential playback, which began on 2026-09-23 on web, mobile and TV (read 2026-10-06, Orthotropy's October 2026 platform roundup). A series wants covers that look like a set. Ideogram's edit mode fits that job: according to the MindStudio write-up of Ideogram 4.5, the update is about precise editing, such as fixing text and making targeted changes without regenerating the whole image. That is a reviewer's account, not a guarantee, so check your own output.

What Sume gives you for this

On Sume, Ideogram 4.5 is ideogram/ideogram-v4.5. Without input_references it generates from text. With references it edits the first image and treats up to four more as references, five in total. If you omit aspect_ratio on an edit, the result keeps the shape of the source, so you do not need to repeat the ratio. List price is $0.03, $0.06 or $0.22 per image by quality, and Sume bills that times 1.25.

Check YouTube's current spec for the thumbnail size, file type and size limit before you pick a ratio, because this post does not state them. Read the aspect_ratio values the catalog lists for the model with GET /v1/images/models and choose the one that matches YouTube's requirement.

The episode loop

The loop below takes the template URL, then asks for each episode in turn. Quote the old number and the new number so the model changes only that text. Always edit the template, not the previous episode, so a small error in one cover cannot carry into the next.

import os
import requests

URL = "https://api.sume.com/v1/images"
HEADERS = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}
TEMPLATE = os.environ["TEMPLATE_URL"]  # public https URL of episode 1

for n in range(2, 5):
    body = {
        "model": "ideogram/ideogram-v4.5",
        "prompt": f'Change the text "EP 1" to "EP {n}". Change nothing else.',
        "input_references": [{"type": "image_url", "image_url": {"url": TEMPLATE}}],
    }
    r = requests.post(URL, headers=HEADERS, json=body, timeout=60)
    print(n, r.status_code, r.json()["data"][0]["url"] if r.status_code == 200 else r.text[:200])

Checks before you upload

Look at every cover. A billed edit that returns the wrong digit is still a completed job, so budget for a retry on a few percent of episodes. If a cover comes back with changes outside the number, tighten the prompt rather than retrying the same words.

Custom Shorts thumbnails have conditions on the YouTube side, which another post covers in Custom Shorts thumbnail needs YPP and a computer. If you would rather use a frame from the video itself, see same-second thumbnail frames for every episode.

A short checklist per episode

Before the first upload, settle these in writing, because changing them later means regenerating the whole set.

• The template cover is approved by whoever owns the channel look, and its URL is saved.
• The number text is quoted the same way in every prompt, with the old value and the new one.
• The ratio comes from the catalog row and matches the current YouTube requirement.
• Each result is opened and read, not only counted as a success.
• A failed or wrong cover is re-edited from the template, never from another episode.
• The file type and size are checked against YouTube's limits before upload.

What this does not do

Sume does not upload to YouTube and does not create the series. It makes the image files. The template look is only as consistent as your first image, so settle the template before you spend on episodes, and keep the template URL and the quoted number text in your own notes so the run can be repeated for season two.