Complete audio of the full script — 73 s.
Same Script, Three Free Tiers: One Finished Clean, One Watermarked Itself, One Refused to Start
One script, word for word, in every tool: a 228-word group-lunch explainer written to run about ninety seconds, frozen before the sprint so no tool got an easier version. This is the avatar-and-voice battery — the S2 test from our How We Test protocol — run the way a cautious buyer runs it first: on the free tiers, before any card comes out. Total spend was $0, and the free tiers still managed to teach us more than most pricing pages do: one tool finished the job end to end, one finished it and then stamped itself into the corner while quietly re-directing the video, and one refused to start at all — because ninety seconds is longer than its free plan allows a video to be.
Plans tested: ElevenLabs Free ($0; 10,000 credits/month), Synthesia Basic ($0; 500 credits/month per the pricing page), HeyGen Free ($0; 3 videos/month, up to 1 minute each). All raw outputs and full logs are archived with this article’s evidence trail. How we handle affiliate links is on our Disclosure page; the affiliate links in this article are marked †.
Script estimated at 78.4 s — over the free plan’s 60 s per-video cap.
The result, in one table
| ElevenLabs Free | Synthesia Basic | HeyGen Free | |
|---|---|---|---|
| Outcome | ✅ Finished — complete audio of the full script | ✅ Finished — complete video of the full script | ❌ No video produced |
| Output | MP3 narration, 44.1 kHz / 128 kbps | MP4 video, 1920×1080 landscape | — |
| Actual duration | 73 s | 75 s | Script estimated at 78.4 s by HeyGen’s own editor — over the free plan’s 60 s per-video cap |
| Generations / retries | 2 generations (1 failed partial + 1 full) | 1 generation, 0 retries | 0 generations; 4 blocked attempts |
| Time to usable output | Full generation under ~1 min | ~9 min from prompt (~11 min to download) | Never — blocked pre-generation |
| Watermark | N/A (audio only) | Yes — black pill, bottom-right | No output to inspect (free output is watermarked per plan page) |
| Credits consumed | 1,232 of 10,000 (incl. 121 wasted — see ledger) | Unresolved — two counters disagree (see ledger) | 0 — nothing consumed, nothing charged |
| Free-tier commercial use | Non-commercial only (terms) | Terms quote not captured in this run — see fine print | Personal / non-commercial / internal evaluation only (terms §4) |
ElevenLabs: the only clean finish — and a 121-credit lesson in pasting
ElevenLabs took the script as narration (voice: “Roger”, model Eleven v4) and returned the whole thing — verified by playback through the final line — in under a minute of generation. Seventy-three seconds of finished audio for a script that announces “the next ninety seconds”: narration pace, not the brief, sets the real duration, in every tool here.
The one failure was ours to learn from and yours to avoid: paste the script with its paragraph breaks and the editor splits it into blocks and generates only the first one. That mistake cost 121 credits before the full single-block re-entry (1,111 credits) produced the usable take. Nothing about the tool’s marketing page will tell you that; it is exactly the kind of friction a free tier exists to reveal.
Synthesia: finished — with a watermark, an auto-director, and two credit counters that disagree
Synthesia’s new accounts are routed through Syren, its conversational video flow, rather than the classic editor. One generation, zero retries, about nine minutes to a usable 1080p video (stock avatar Gordon, Brian voice, plain studio background). The script was rendered verbatim.
Three things a buyer should know before the demo reel seduces them:
- The watermark is real. A black Synthesia pill sits in the bottom-right of the exported MP4, and the pricing page confirms Basic-plan videos carry it. The free output is a demo, not a deliverable.
- Syren directs, whether you asked or not. It added an opening title card and a summary card for each script section on its own. Some users will love the head start; anyone with a fixed format or brand template should know the tool has opinions.
- The credit accounting does not reconcile. The Syren conversation counter showed 131 credits used for this video. Settings > Billing showed the Basic plan at 0 of 500 credits used, unchanged. Both numbers are from the same account, minutes apart. Until Synthesia’s counters agree, nobody — us included — can honestly quote a per-video credit cost on the Basic plan, so we are not quoting one.
HeyGen: a 78-second script meets a 60-second wall
HeyGen never generated a frame. Its editor estimates this script at 78.4 seconds; the Free plan caps each video at one minute. We tried every legitimate route: the single-scene editor, AI Studio, AI Studio with the voice speed at its 1.5× maximum (the cap check ignores playback speed — it measures the script, not the render), and scene-splitting (the cap is on total duration). A fourth attempt opened nothing but an upgrade modal. Nothing was consumed — the account still shows 0 of 3 free videos used — and the full script sits saved as a draft in the account, ready for the paid pass.
The finding is not “HeyGen is bad.” The finding is narrower and more useful: HeyGen’s free tier is a one-minute sampler. Any real explainer script — ninety seconds is short by explainer standards — cannot be evaluated on it at all. If your content runs longer than a minute, HeyGen’s free tier tells you nothing except where the paywall is.
The credit ledger (everything we spent, including the waste)
| Tool | Before | After | Burned | Note |
|---|---|---|---|---|
| ElevenLabs | 10,000 | 8,768 | 1,232 | 1,111 for the usable take + 121 wasted on the paragraph-split partial generation |
| Synthesia | 500 (Billing) | 500 (Billing) | 0 per Billing / 131 per Syren | Two counters, two answers — unresolved, reported as-is |
| HeyGen | 3 videos/mo | 3 videos/mo | 0 | Blocked before generation; “you weren’t charged” |
Cash per usable finished minute is $0.00 across the board, which is another way of saying the free tiers price nothing: what a finished minute actually costs is a paid-plan question, and it is not yet tested for any of these three tools.
The fine print: what “free” lets you do with the output
- ElevenLabs (Free): the Terms of Service limit free-tier use to non-commercial purposes; the Commercial License is a feature of the paid Starter plan. The finished audio above cannot legally front a business.
- HeyGen (Free): Terms of Service §4 (updated 2026-10-16) licenses free-plan output solely for personal, non-commercial, and internal evaluation purposes — it may not be sold, sublicensed, redistributed, or monetized.
- Synthesia (Basic): the verified facts from this run are the watermark on the export and the pricing page’s statement that Basic videos carry it. We did not capture a Basic-tier commercial-terms quote during the run, so we are making no claim about Synthesia’s free-tier commercial terms until that quote is archived (our protocol’s §6 publish gate).
Pattern worth remembering: on the two free tiers whose terms we did capture, “free” means free to evaluate, not free to use in your business.
Scores — deliberately not published for this battery
Our protocol scores six weighted dimensions. This battery cannot honestly fill them: HeyGen produced no output to score, the script-change editing test was not run on any tool (it is scheduled with the paid pass), and the language pass (ES/ZH) was not run on free tiers. What the logs do support is a completion verdict for the free tiers, and that is what this article publishes:
- ElevenLabs Free — completes the S2 job (voice only), with the paste-blocks caveat. The only one of the three we’d call evaluation-complete.
- Synthesia† Basic — completes the S2 job as a demo: real video, real watermark, an auto-director you didn’t hire, and credit accounting you cannot yet audit.
- HeyGen Free — cannot attempt the S2 job. A 1-minute cap fails a 90-second brief before quality is even a question.
Not yet tested — read this before quoting us
- All paid tiers of all three tools: NOT TESTED. HeyGen Creator ($29/mo shown in-product), Synthesia Starter, and ElevenLabs’ paid entry tier have not been run, paid for, or scored. Any claim about their output quality, duration limits, watermark removal, or cost per finished minute would be a vendor claim, and we don’t publish those as results.
- The script-change re-render test (change one sentence, measure the cost of fixing it) — deferred to the paid pass.
- The ES/ZH language pass and the S3 source-video generation (the Sprint B winner records the ~10-minute source for the repurposing battery) — both wait on the paid pass, because no free tier here can hold a 10-minute video.
Changelog
- v1 — October 29, 2026: First publication of Sprint B free-tier results (runs 2026-10-28). Paid Sprint B deferred by editorial decision the same day; this article covers free tiers only and will be superseded by the paid head-to-head when it runs.