Hedra live avatars at $0.05 a minute vs 60 seconds of Sume lip sync

Hedra lists real-time live avatars at $0.05 a minute. A recorded 60 second line on Sume costs $6.00 on H3 Max 768p. The arithmetic and what each buys.

4 min readSume
All posts

A minute of Hedra's live avatar is listed at $0.05, while a minute of recorded lip sync on Sume H3 Max at 768p costs about $6.00 (four segments, each billed as 15 seconds at $0.10 a second). The two numbers are not like for like: one is a real-time conversation that responds as someone speaks, the other is a finished video file you can cut, caption and publish. This post lays out the arithmetic so you can decide which problem you have.

The numbers

Hedra's page, dated June 1, 2026, lists real-time live avatars at $0.05 per minute with sub-100ms response. Sume has no real-time avatar route, so every Sume figure is for a rendered clip.

The H3 Max lip-sync route only accepts clips of 5 to 14.8 seconds. A 60-second line therefore becomes four segments. Each is billed on ceil(duration_seconds), so a 14.8 second segment bills as 15 seconds. At 768p the Sume price is the fal list of $0.08 times a 1.25 margin, which is $0.10 per second. Fabric 1.0 charges the provider list of $0.15 a second at 720p plus the same margin, which is $0.1875 per second.

One minute, two products (read 2026-10-08)
OptionBasisOne minute
Hedra live avatar$0.05 per minute as listed, real time$0.05
Sume H3 Max 768p4 segments x 15 billed seconds x $0.10$6.00
Sume H3 Max 480p60 billed seconds x $0.0625$3.75
Sume H3 Max 1080p60 billed seconds x $0.20$12.00
Sume Fabric 1.0 720p60 seconds x $0.1875 (derived from the docs rate)$11.25

Check the math yourself

This script reproduces the Sume rows. It makes no network calls.

import math

RATE = {"480p": 0.0625, "768p": 0.10, "1080p": 0.20}

def cost(durations, res):
    return sum(math.ceil(d) for d in durations) * RATE[res]

segments = [14.8, 14.8, 14.8, 14.8]
for res in RATE:
    print(res, round(cost(segments, res), 2))

Reading the rounding

The rounding is the part people miss. Every segment is rounded up to a whole second before pricing, so splitting a minute into many short lines costs more than one long line would. Four segments of 14.8 seconds total 59.2 seconds of speech but bill 60 seconds. Sixteen segments of 3.7 seconds are not allowed at all, because the route refuses anything under 5 seconds. If your script has short lines, merge them into one audio file of at least 5 seconds first. Because the route is billed by reserve at admission, the number you see in the admission preview (POST /v1/generation/admission-preview) is the number that is held, and a failed job is refunded.

When live is the right tool

If a visitor needs to talk to a face and hear an answer in under a second, you want a live avatar and you should price it from the vendor's own page. If you need the same short message to be correct, reviewable and reusable across channels, a recorded clip is cheaper to check: you read the script, approve the file and publish it. Sume gives you the second kind, with the job reserved at admission, captured on completion and refunded on failure.

What Sume does not do

Sume does not stream audio into an avatar and does not answer in real time. I did not find a Sume price that is comparable to $0.05 a minute, and I would not claim one. Also remember that the Hedra figure is a page dated June; confirm it is current before you budget.

Sources

Related posts

More in Comparisons

All Comparisons posts

Written by Sume