What 10 seconds of talking avatar video costs on Sume
Sume bills a 10-second avatar video at $1.84, $2.45 or $5.50 by tier. How that compares with Hedra's published 10-second video model prices, and what differs.
A 10-second talking avatar video costs $1.84 on Sume's standard tier, $2.45 on plus (the default) and $5.50 on max, with no product image. Add a product_image and it is $1.94, $2.58 and $5.80. Sume bills avatar video per second at a fixed rate per tier, so the cost of any length is the rate times the estimated seconds, between the 4-second minimum and the 60-second maximum.
The rates come from the catalog rate card shipped with the Sume API (sume_avatar_video_fixed_tier_rates_2026_07_02); confirm them live with GET /v1/catalog before you build a budget on them.
Sume rates for a 10-second clip
The per-second rates multiply cleanly.
| Tier | Per second, no product | 10 s, no product | Per second, with product | 10 s, with product |
|---|---|---|---|---|
standard | $0.184 | $1.84 | $0.194 | $1.94 |
plus (default) | $0.245 | $2.45 | $0.258 | $2.58 |
max | $0.55 | $5.50 | $0.58 | $5.80 |
The Hedra table is a different product
Hedra published a price comparison on 2026-09-25 for one fixed job: a 10-second, 1080p clip with native audio generated from a start frame, using rates from its own API catalog. Its table runs from $0.60 for LTX-2.3 Fast to $7.91 for Seedance 2.5, with Kling V3 Pro at $1.68 and Wan 3.0 Standard at $2.00.
Read that table as a price list for general video models, not for talking avatars. A start-frame video model animates a scene and can generate sound; it does not take your script and keep one presenter's face and voice consistent. A rendered avatar clip from Sume is script in, presenter out, at 720p. Putting $2.45 next to $1.68 and calling it a like-for-like comparison would be wrong, because the 1080p general clip and the 720p presenter clip are different deliverables.
The overlap is real only in one place: if your 10 seconds is B-roll or product motion with no talking face, a general video model is the right line item, and Sume's models page lists those separately from avatar video.
Working out your own number
Per-second billing also means a rewrite that trims two seconds saves two seconds of cost, which is rarely true of plans that sell a fixed bundle of minutes.
- Count the words and speak them at a normal pace; Sume estimates duration from the script and rejects estimates outside 4 to 60 seconds.
- Pick the tier.
standardis the fastest path,plusis balanced,maxis highest quality and slower. - Add the product-image rate only if you pass
product_image. - Preview first when the framing is uncertain: preview stills are tier-independent, and
generate-videocan set the final tier.
Limits
Avatar video is 720p today, up to 60 seconds per job, and one avatar per video. Hedra's rates are as published on the date above and may change.
Sources
Related posts
More in Pricing
- 30-second AI video in one take vs stitched clips on Wan 3.0
Wan 3.0 accepts 2 to 30 seconds per request. A 30-second take costs $1.88 at 480p, $3.75 at 720p and $7.50 at 1080p on Sume; stitched clips cost the same.
- AI image budget guard in Python: stop when usage.cost passes a limit
Each Sume image response reports usage.cost in USD. Add it up in a loop and stop a bulk run at a cap you set, so a stuck prompt cannot drain credits.
- AI video analysis API cost: free probe and stills, $0.01 a minute
Sume video inspect probes a clip and samples up to 24 stills free; only the optional transcript bills, $0.01 per audio minute from a hint of up to 600 seconds.
- AI video credits in dollars: InVideo, Kapwing, a USD wallet
InVideo credits cost $0.015 to $0.05 each and Kapwing Pro credits $0.016 to $0.024. Sume balances are in USD, so a price reads the same on every call.
Written by Sume