AI ad production cost per 1,000 impressions: when a clip is too costly
Turn a Sume clip price into a production CPM: a 15-second Avatar Plus ad with captions is $3.875, so 100,000 views carry about 3.9 cents per thousand.

The production cost of one ad per 1,000 impressions is the clip cost divided by impressions, times 1,000. A 15-second Sume Avatar Plus clip with one caption job costs $3.875, which is $3.875 per 1,000 impressions if only 1,000 people see it and about $0.039 per 1,000 if 100,000 do.
The point is not that AI video is cheap. It is that production cost is a fixed amount spread over reach, so it matters for the creative that gets almost no delivery and nearly vanishes for the one that gets a lot. Every media figure below is a variable you supply; this post uses none from an ad platform.
What does one finished ad cost on Sume?
The inputs are Sume's published rates from the API pricing page and the Avatar videos and Video captions docs. Avatar Video bills per second by quality tier, and a standalone caption job is $0.20 for a video up to 60 seconds. The avatar window is 4 to 60 seconds.
| Quality tier | Rate per second | 15 seconds | Plus one caption job |
|---|---|---|---|
| Standard | $0.184 | $2.76 | $2.96 |
| Plus (default) | $0.245 | $3.675 | $3.875 |
| Max | $0.55 | $8.25 | $8.45 |
How does the cost fall as reach grows?
Using the Plus figure of $3.875 per ad, production cost per 1,000 impressions is:
| Impressions | Production cost per 1,000 |
|---|---|
| 1,000 | $3.875 |
| 10,000 | $0.3875 |
| 100,000 | $0.0388 |
| 1,000,000 | $0.0039 |
When does production cost matter against media cost?
Compare production cost per 1,000 with your own media CPM. If production should be no more than 10% of media, you need clip cost x 1000 / (0.10 x CPM) impressions. With a made-up CPM of $10, chosen only for the arithmetic, that is 3,875 impressions for the $3.875 ad. Below that, the clip costs more than a tenth of what you spent to show it.
That is the real use of the number: an ad test with 1,500 impressions per variant is dominated by production cost, so make variants cheaper (see the tool prices below), not fewer.
def production_cpm(clip_cost: float, impressions: int) -> float:
"""Production cost per 1,000 impressions."""
return clip_cost / impressions * 1000
def impressions_for_share(clip_cost: float, media_cpm: float, share: float) -> int:
"""Impressions at which production is `share` of media spend."""
return round(clip_cost * 1000 / (share * media_cpm))
cost = 15 * 0.245 + 0.20 # 15 s Avatar Plus, no product, plus one caption job
for n in (1_000, 10_000, 100_000, 1_000_000):
print(n, round(production_cpm(cost, n), 4))
print(impressions_for_share(cost, media_cpm=10.0, share=0.10))What about retakes?
A delivered ad is not the only ad you paid for. If you generate three takes and keep one, the cost of the kept ad is three times the clip, or $11.63 at Plus. Failed jobs release their hold in Sume, but a take you simply reject is a completed, captured job. The retake maths is in cost per keeper.
Use your own keeper rate, not an average from a blog. Count two numbers over a week: takes generated and takes shipped.
Where does Sume not help here?
Sume does not report impressions, spend or CPM, and it does not publish ads. The wallet side stops at the job. Join the job ids from /v1/usage to the delivery numbers from the ad platform in your own sheet.
Sources
Related posts
More in Pricing
- AI lip sync API cost per second: H3 Max 480p, 768p, 1080p vs Fabric
On Sume, H3 Max lip-sync is $0.0625, $0.10 or $0.20 per audio second by resolution; VEED Fabric is $0.10 or $0.1875. A 14.8 s clip costs $0.94 to $3.00.
- AI video cost per deliverable: a rate sheet for finance approval
A cost-per-deliverable sheet built from Sume's published rates: 15-second avatar ad, captioned cut, 6-image set, voiced script, music bed, with a retake factor.
- AI virtual try-on cost per image: GPT Image 2.5 on Sume
What one try-on still costs on Sume: the documented token rates, why quality auto reserves the max price, and a script that reads the live endpoint pricing.
- AI voice agent cost per minute: speech-to-text, LLM and TTS stack
A dated cost sheet for the October 2026 voice stack: MAI-Transcribe-2-Streaming, Mercury Voice, MAI-Voice-2.1-Flash, and Sume's file STT and TTS, per minute.
Written by Sume