How Humva scores AI avatar video tools
Every tool is scored with one published formula: five weighted criteria, each banded from documented and dated evidence, combined into a total out of 10. Anyone can recompute a score from the inputs printed on its review.
The formula and its five criteria
total = sum(criterion score 0-10 x weight) / 100, shown to one decimal.
Not a hands-on rating. The desk has not run HeyGen.
- 30%
Cost per finished minute
Entry paid plan billed monthly: (a) the lowest documented per-minute cost (lowest-rate engine) and (b) the cost on the dearest documented avatar engine on the same plan.
- 20%
Plan limits and output rights
8 checks on the entry paid plan, score = checks met / 8 x 10: no watermark; commercial use of output; export at 1080p or better; a single video can run 10 minutes or more; a voice clone included; 100+ languages; extra credits purchasable on that plan; plan credits usable through the vendor's own integrations..
- 20%
Terms, likeness and data
5 checks, score = checks met x 2: paid users own their output; no training license over uploads, or one that ends with the account; a documented training opt-out; a consent check on every avatar built from a real person; a stated deletion deadline..
- 15%
Pricing transparency and stability
5 checks, score = checks met x 2: publishes a credit-to-minute rate for each avatar engine; publishes rollover and expiry rules; publishes its refund rule; one published rate applies to every account; no plan-structure change in the last 12 months..
- 15%
Independent user sentiment
Trustpilot TrustScore x 2, minus 1 if the one-star share is 10% or more, minus 2 if it is 20% or more.
Each criterion, band by band
- Score = 0.6 x band(a) + 0.4 x band(b). Bands: <= $0.50 -> 10 | <= $1.00 -> 8 | <= $2.00 -> 6 | <= $4.00 -> 4 | <= $8.00 -> 2 | above $8.00 or undocumented -> 0
- 8 checks on the entry paid plan, score = checks met / 8 x 10: no watermark; commercial use of output; export at 1080p or better; a single video can run 10 minutes or more; a voice clone included; 100+ languages; extra credits purchasable on that plan; plan credits usable through the vendor's own integrations.
- 5 checks, score = checks met x 2: paid users own their output; no training license over uploads, or one that ends with the account; a documented training opt-out; a consent check on every avatar built from a real person; a stated deletion deadline.
- 5 checks, score = checks met x 2: publishes a credit-to-minute rate for each avatar engine; publishes rollover and expiry rules; publishes its refund rule; one published rate applies to every account; no plan-structure change in the last 12 months.
- Trustpilot TrustScore x 2, minus 1 if the one-star share is 10% or more, minus 2 if it is 20% or more. Trustpilot is used because it is the one public aggregate that lists every reviewed tool (G2 and Capterra returned bot walls on 2026-10-14). Invitation status is reported beside the score, not inside it.
A worked example: HeyGen, 6.6
HeyGen on the Humva Scorecard
6.6/ 10
Evidence dated
From evidence to a total
What the score never reads
- The desk has not rendered the tools itself, and vendor lip-sync metrics do not track human judgement, so quality evidence from named testers is reported beside the score, never inside it.
- No. A partner program, a commission rate or an advertiser relationship is never an input; the weights and bands are the same for every tool.
- Re-score when any input changes; prices are re-read every 90 days (refresh_prices.py --check); the score carries its evidence date beside it.
Recompute it yourself
Every input is printed on the review
Take any review's five sub-scores, multiply by the weights and divide by 100. If your number differs from ours, tell the desk.
Humva Scorecard, v1
total = sum(criterion score 0-10 x weight) / 100, shown to one decimal
- 30% Cost per finished minute
- 20% Plan limits and output rights
- 20% Terms, likeness and data
- 15% Pricing transparency and stability
- 15% Independent user sentiment