3-minute AI presenter Reel: three 60-second avatar jobs, cost by tier

A talking-avatar job on Sume accepts a script that estimates between 4 and 60 seconds, so a 3-minute presenter Reel is three jobs of up to 60 seconds. At the no-product per-second rates, 180 seconds of avatar video costs $33.12 on Standard, $44.10 on Plus and $99.00 on Max, before assembly: $0.01 per audio detach (three of them) and the $0.30 Timeline render.

The arithmetic

Rates come from the catalog rate card for sume/avatar-1.0/talking-video (read 2026-10-03): Standard $0.184 per second, Plus $0.245 and Max $0.55, with product-image variants slightly higher. Output is 720p, and the default aspect ratio is 9:16. Standard is the fastest tier and Max the highest quality and slower; Plus is the default.

Plan the split

Metricool (read 2026-10-03) reports a 3-minute Reel length, and Instagram's guide (read 2026-10-03) lists 60 seconds to 3 minutes for longer storytelling. Split the script at sentence ends into three parts of about equal estimated length, so no part exceeds 60 seconds; a part that estimates over 60 is rejected, and one under 4 is billed at 4.

import re

def split_script(text, parts=3):
    sents = re.split(r"(?<=[.!?])\s+", text.strip())
    target = len(text) / parts
    out, cur = [], ""
    for s in sents:
        if cur and len(cur) + len(s) > target and len(out) < parts - 1:
            out.append(cur.strip())
            cur = ""
        cur += s + " "
    out.append(cur.strip())
    return out

for p in split_script("One. Two is here. Three follows. Four ends. " * 6):
    print(len(p))

Join the three clips

Put the three MP4s on consecutive slots of Timeline 1.0. The render takes its sound from one audio spine, so detach each clip's voice ($0.01 each) and join the three files with audio.parts[] in the render, or into one file with timeline audio ($0.01). Use a hard cut at each join, since the presenter returns to a similar pose, or add a short fade (up to 1 second). Preview the first part with the preview flow before you commit to three paid renders.

What Sume does not do

Sume does not keep gesture continuity between separate jobs and does not provide a single 180-second presenter job. Confirm the live prices with GET /v1/catalog before a batch; the figures here are dated.