A dubbed minute costs $0.33 to $9.00, and the watermark discount is gone

What moved between June 2025 and August 2026

Four dated events changed what a buyer can actually see on a price page.

26 June 2025. RWS announced it had acquired the IP behind Papercup's AI dubbing technology. As of today the consequence is visible in DNS: papercup.com and papercup.com/pricing both return HTTP 301 to an RWS AI dubbing and voiceover page carrying utm_campaign=papercup+migration. One of the four vendors most often named in dubbing comparisons no longer has a website, let alone a price.

28 May 2026. ElevenLabs announced Dubbing v2, describing a model that conditions directly on the original performance rather than working from a transcript, across 90+ languages. The post contains no benchmark, no listening-test result and no comparison figure.

2 August 2026. The EU AI Act's Article 50 transparency obligations began to apply, requiring providers of systems that generate synthetic audio, image, video or text to mark output in a machine-readable format. We covered what survives a dubbing mix in a separate note.

6 to 10 August 2026. ElevenLabs posted Dubbing v2 is now available via ElevenAPI, dated 6 August, and the changelog entry for 10 August 2026 records API availability. The two dates differ by four days and we did not resolve which one is the switch-over; treat the API as live in the second week of August.

The pricing consequence is the part nobody wrote up. Dubbing v1 sold a watermarked dub for less than a clean one. Dubbing v2, per the ElevenLabs documentation, "does not include a watermark toggle", and paid-tier dubs are not watermarked while free-tier dubs are watermarked automatically. The cheapest way to buy a marked dub from the most-quoted vendor in the category was removed in the same month the marking regime began to apply.

What does a minute actually cost?

Two vendors quote in minutes, one quotes in credits, and two quote nothing. Where a vendor sells an annual plan with an included minute allowance, the per-minute rate below is that plan's annual price divided by its annual minutes, on the assumption the allowance is fully consumed. That assumption is generous to the vendor: unused minutes raise the real rate.

Note on the TTS row: we assume roughly 150 spoken words per minute at about six characters per word including spaces, so about 900 characters per minute of speech. That is our assumption, not a vendor figure, and it moves with speaking rate and script.

Three things fall out of that table that are not on any vendor's page.

Rask's most expensive plan has the second-cheapest minute. Creator Pro at $936 a year for 1,200 minutes is $0.78 a minute. Business at $6,000 a year for 6,000 minutes is $1.00. The plan that costs 6.4 times more per month costs 28% more per minute. Same for HeyGen on credits: Business is $0.0993 per credit against Creator's $0.0483, a factor of 2.06 in the wrong direction. Buying up a tier in this category buys seats, support and limits, not unit economics.

The v1 to v2 jump is 4.4 times. $0.50 to $2.20 on the clean price, 6.7 times on the watermarked price. No published quality figure accompanies either model, so a buyer choosing between them is choosing between a documented price and an undocumented improvement.

HeyGen's own comparison page disagrees with HeyGen's pricing page. The comparison page, last updated 19 May 2026, states Creator is "$24/month" with "no per-minute or per-credit charges at any tier". The pricing page today shows Creator at $29 a month with a 600-credit monthly allowance. We did not resolve the difference and are reporting both.

Where $2.20 goes when the voice costs nine cents

The reference rows in Table 1 make the cost structure legible. ElevenLabs sells the two component operations of a dub separately and publicly: transcription at $0.22 an hour and speech synthesis at $0.10 per 1,000 characters. Put a minute of speech through both and the raw model cost is around nine cents.

The same subtraction run across the three published dubbing prices gives a wrapper of $0.24 on the $0.33 watermarked v1 job (72% of the price), $0.41 on the $0.50 clean v1 job (81%), and $2.11 on the $2.20 v2 job (96%). The share the buyer cannot inspect grew with every release.

That is the useful reframing for procurement. You are not buying synthesis, which is a commodity with a public unit price that has fallen for three years. You are buying the alignment, the timing, the speaker handling and whatever review sits inside the pipeline, and none of the five vendors itemises the last one. A vendor charging $2.20 and a vendor charging $0.78 may differ by 100% in human review or by 0%, and the price lists do not distinguish those cases.

What to do with this if you are buying dubbing

Price the plan, not the sticker. Divide annual plan cost by annual included minutes before comparing anything, then multiply by the lip-sync consumption factor you will actually use. Rask's enhanced lip-sync consumes three minutes of allowance per minute of video, which turns a $0.78 plan into a $2.34 plan without any line item changing. Then assume you will not consume the full allowance and recompute at 60% utilisation, which is where most annual minute plans land in our experience. At 60%, Rask Creator Pro standard is $1.30 a minute, not $0.78.

Ask what the wrapper contains, in writing. The 96% of the Dubbing v2 price that is not synthesis is the whole product. The three questions that separate vendors are: is a human native speaker reviewing any output, on what sampling fraction, and is the translation step a general MT system or one tuned per locale. None of these are on any price page.

If you need a machine-readable mark, check before you migrate. The v1 to v2 migration removes the watermark toggle from the category leader's flagship model. If your Article 50 assessment concluded that your output needs marking, moving to v2 changes your compliance posture, and it is not a line item anyone will flag for you. Our note on whether the mark survives a dubbing mix covers what the mark is worth once it exists.

Treat the absence of a quality figure as the finding. Five vendors, zero published evaluation numbers, and a 27-fold price spread. In any market where price varies that much and quality is unmeasured, the price is carrying information about packaging rather than about output. Run your own listening test on your own content in your own top three locales before signing anything, because nobody is going to hand you a number.

The consolidation is real and it removes options. Papercup was the vendor most often recommended for human-in-the-loop dubbing, and its domain is now a redirect. That is one fewer independent price point in a category that already publishes almost nothing. This is our reading rather than a company statement: expect the enterprise tier of this market to move further toward quote-only pricing, not less.

Check it yourself

Every number in Table 1 is reproducible in about five minutes, and the two redirect findings take one command each.

# 1. the Papercup redirect and the missing Deepdub price page
for u in https://www.papercup.com/pricing https://www.papercup.com/ \
         https://deepdub.ai/pricing https://deepdub.ai/; do
  printf '%s -> ' "$u"
  curl -s -o /dev/null -w '%{http_code} %{redirect_url}\n' -A "Mozilla/5.0" "$u"
done
# 2026-09-10 output:
#   https://www.papercup.com/pricing -> 301 https://www.rws.com/localization/services/
#     translation-services/video-and-audio-translation/ai-dubbing-and-vo?utm_campaign=papercup+migration
#   https://deepdub.ai/pricing       -> 404

# 2. the price pages, read them yourself and date-stamp what you see
#    elevenlabs.io/pricing/api   elevenlabs.io/docs/overview/capabilities/dubbing
#    rask.ai/pricing             heygen.com/pricing

# 3. normalize any plan to dollars per minute
python3 - <<'PY'
plans = {  # name: (annual_price_usd, annual_minutes)
  "Rask Creator":     (396,   300),
  "Rask Creator Pro": (936,  1200),
  "Rask Business":   (6000,  6000),
}
for name, (price, mins) in plans.items():
    base = price / mins
    print(f"{name:18s} standard ${base:.2f}/min  enhanced-lipsync ${base*3:.2f}/min"
          f"  at 60% utilisation ${base/0.6:.2f}/min")
PY

# 4. the raw model cost of a minute, for the wrapper calculation
#    STT: $0.22/hr / 60                        = $0.0037/min
#    TTS: $0.10/1000 chars * ~900 chars/min    = $0.090/min
#    wrapper share of a $2.20 dub              = 1 - 0.094/2.20 = 95.7%

Three numbers are worth recording per vendor per quarter: the per-minute rate you actually paid including unused allowance, the fraction of delivered minutes your reviewers sent back, and whether the output still carries a machine-readable mark after your delivery encode. The third one changes without notice, as the v1 to v2 migration shows.

What would prove this wrong

The claim under test is that AI dubbing in 2026 is priced without any published quality signal, and that price differences across vendors therefore carry no information about output quality. It is wrong if, by 1 June 2027, at least two of the five vendors named here publish a per-minute price alongside a reproducible quality number for at least ten target languages, naming the evaluation set and the rater population, in a document a buyer can link to. Today the count is zero of five. A vendor-run listening test with an unnamed rater pool and no released audio would not settle it.

A second prediction, marked as judgement rather than finding: the $0.33 watermarked tier will disappear from the ElevenLabs price list before the v1 model is retired, because it is the only line item in the category that prices provenance marking as a discount rather than a feature, and that is an awkward position to hold while Article 50 enforcement builds. If it survives past 1 June 2027, that reading was wrong.

FAQ

How much does AI video localization cost in 2026?

Between $0.33 and $9.00 per minute of source media from published price lists read on 10 September 2026. ElevenLabs lists Dubbing v1 at $0.33 watermarked and $0.50 clean, and Dubbing v2 at $2.20. Rask AI's annual plans compute to $1.32, $0.78 and $1.00 per minute on Creator, Creator Pro and Business, tripling with enhanced lip-sync and reaching $9.00 on an overage minute. HeyGen publishes credits with no minute rate. Deepdub and Papercup publish no price.

Does ElevenLabs still offer a watermarked dubbing discount?

Only on the older model. The API price list shows Dubbing v1 at $0.33 per minute watermarked against $0.50 clean, a 51.5% premium to remove the mark. The documentation states Dubbing v2 does not include a watermark toggle and that paid-tier dubs are not watermarked, while free-tier dubs are watermarked automatically. Dubbing v2 reached the API on 10 August 2026, eight days after Article 50 transparency obligations began to apply on 2 August 2026.

Why is Dubbing v2 more expensive than Dubbing v1?

No reason and no quality figure have been published. The observable differences are the price ($2.20 against $0.33 to $0.50), a model that conditions on the original performance rather than a transcript, 90+ languages with dialect variants such as en-AU, es-MX and fr-CA, a 3 GB API upload limit against v1's 1 GB and 45 minutes, and the fact that Dubbing Studio remains v1-only and is described as in maintenance mode.

Which AI dubbing vendors publish their prices?

Three of five, and only two in minutes. ElevenLabs publishes per-minute API pricing. Rask AI publishes plans, included minutes and a $3 overage from which a minute rate can be computed. HeyGen publishes plans and credit allowances but no credit-per-minute rate. Deepdub's pricing URL returned HTTP 404 on 10 September 2026 and papercup.com returned HTTP 301 to an RWS page, following RWS's acquisition of Papercup's IP announced 26 June 2025.

Is AI dubbing cheaper than human dubbing?

On published data, that comparison cannot be made. AI vendors publish per-minute prices; studio and agency dubbing is quoted per project and the ranges that circulate online come from vendor marketing pages rather than rate cards. We could not find a primary, citable human dubbing rate card, so we are not printing one. The honest statement is that AI dubbing has published prices and human dubbing does not, which is a difference in transparency before it is a difference in cost.

Sources

1. ElevenLabs API pricing, read 10 September 2026. Dubbing v1 $0.33 per minute watermarked and $0.50 without; Dubbing Studio $0.50; Dubbing v2 $2.20; Scribe v2 $0.22 per hour and Scribe v2 Realtime $0.39 per hour; text to speech v3 and v2 Multilingual $0.10 per 1,000 characters, Flash and Turbo $0.05.
2. ElevenLabs dubbing documentation, read 10 September 2026. "Dubbing v2 does not include a watermark toggle"; free-tier dubs watermarked automatically and paid-tier dubs not; 90+ languages with dialect variants; v2 3 GB API upload limit against v1's 1 GB and 45 minutes; Dubbing Studio v1-only and in maintenance mode; v2 transcript editing via API on Enterprise plans only.
3. ElevenLabs, Introducing Dubbing v2, 28 May 2026, and Dubbing v2 is now available via ElevenAPI, 6 August 2026, with the changelog entry dated 10 August 2026. Model description, 90+ languages, no evaluation figures in either post.
4. Rask AI pricing, read 10 September 2026, annual billing. Creator $33/mo billed $396/yr for 300 min/yr; Creator Pro $78/mo billed $936/yr for 1,200 min/yr; Business $500/mo billed $6,000/yr for 6,000 min/yr; $3 per extra minute on Creator Pro and Business; standard lip-sync 1 minute of video equals 1 minute of allowance, enhanced lip-sync equals 3, described as beta and subject to change.
5. HeyGen pricing, read 10 September 2026. Free $0 with 3 videos per month; Creator $29/mo with 600 credits; Pro $49/mo with 1,000 credits; Business $149/mo with 1,500 credits plus $20 per additional seat; Enterprise by quote. Credit consumption stated to vary with model, duration and complexity. Compare HeyGen's own comparison page, last updated 19 May 2026, which states a $24 Creator plan and "no per-minute or per-credit charges at any tier".
6. RWS, RWS acquires Papercup's IP to power next-gen AI dubbing for enterprise clients, 26 June 2025, Business Wire release of the same date. Acquisition of the technology IP, human-in-the-loop positioning, 1,800 in-house linguists and a network of over 40,000 language experts.
7. HTTP status observations made by BLOMEGA on 10 September 2026 with curl: papercup.com and papercup.com/pricing return 301 to the RWS AI dubbing and voiceover page with utm_campaign=papercup+migration; deepdub.ai/pricing returns 404 while deepdub.ai/ returns 200. The commands are in the "Check it yourself" section.
8. European Union, AI Act Article 50, transparency obligations for providers and deployers of certain AI systems, applicable from 2 August 2026, and the European Commission's FAQ on Article 50 transparency obligations.

Related BLOMEGA guides: Article 50 applies to your dub, and the watermark it asks for dies in your mix · Localization is the new default · Your multilingual LLM judge prefers the machine translation · Consented AI training data providers.