ReviewUpdated 28 Oct 2026
Vizard AI Model v1 vs v2: Which Is Better?
Same video through Vizard AI's Model v1 and v2 on the free plan. Retested 28 Oct 2026: v1 now returns 5 clips and joins moments, as v2 did. v2 costs 30% more.
⌕ How we tested this
Hands-on free-plan testing 2026-08-25 — same 10:01 1080p source run through Model v1 and Model v2 with identical settings, credit costs recorded before and after each run, output clips downloaded and inspected with ffprobe, captions scored against an official reference transcript. Model v1 retested 2026-10-28 on the same source file (same settings, English); v2 not re-run.
⚠️ Vizard Model v1 or v2 — Which Should You Use?
Start with v1. It costs about 30% fewer credits per minute of video, and since Vizard changed v1 its output is much closer to v2's than it was.
In August 2026, on an identical 10-minute video, v1 returned 10 clips and v2 returned 5. Retested on 28 October 2026, v1 returned 5 clips from the same file, and it joined separate moments into one clip, which only v2 did in August. We have not re-run v2 since August.
Both models are available on the free plan — v2 is not paywalled. In August, captions and export quality were identical between them, so the extra credits bought clip selection and nothing else.
This page answers one question: when you see that "Model v1 / Model v2" dropdown on Vizard's upload screen, which one should you pick? If you want the full tool walkthrough instead, read our Vizard AI review. Here we stay narrow — same video, same settings, both models, and the numbers that came back.
Update, 28 October 2026: Model v1 has changed. We ran the same 10:01 file through Model v1 again, with the same settings and English set. It cost the same 10 credits, taken at import. It returned 5 clips (35 to 86 seconds, 284 seconds in total) instead of August's 10, and two of them were built from separate moments of the source: one from 01:51, 02:46, 02:53 and 03:13, another from 05:15, 05:23 and 05:54. In August only v2 did that. Downloads were still 720×1280 with the watermark in every frame.
We did not re-run Model v2, so its numbers below are from August. The rest of this page is the August comparison as we ran it.
What We Tested
We ran Vizard exactly as a creator on the free plan would, with no paid features unlocked.
The setup:
- A fresh free-plan account with 60 monthly credits
- One source file: a 10:01, 1920x1080, 30fps video, uploaded directly (not pasted as a link)
- Identical settings on both runs — 9:16 ratio, "Any length" clips, Default template, Add emojis on, Highlight keywords on, B-rolls off, Remove silences off, Auto-censor off
- No subtitle file attached, so captions had to come from Vizard's own transcription
- Credits recorded before and after every step
- Output clips downloaded and inspected with
ffprobe
The source was a NASA Artemis II mission broadcast — public domain, two presenters at a desk, continuous speech throughout, no burned-in captions. We use the same file across every tool we test so results stay comparable.
One thing we could not control: the v2 run had spoken language set explicitly to English, while the v1 run was left on the default. For unambiguous English audio we do not think this changed anything, but it means this is not a perfectly controlled A/B and we would rather say so.
What Each Model Costs
Vizard prices by the minute of source video, and v2 carries a premium:
| Source length | Model v1 | Model v2 |
|---|---|---|
| 10:01 | 10 credits | 13 credits |
| 22:48 | 22 credits | 28 credits |
| Effective rate | ~1.0 credit/min | ~1.3 credits/min |
The model selector appears only at import. For this 22:48 source, Model v1 was quoted 22 credits and Model v2 28 — roughly a 30% premium.
On a 60-credit free month that works out to roughly 60 minutes of source video on v1, or about 46 minutes on v2.
Model v2 selected on a 10:01 1080p upload: 13 credits, against 10 for the same file on v1.
Watch the billing order. Vizard charges the base per-minute rate the moment you import a video — before you have configured anything or generated a single clip. The v2 premium is then charged separately when clips are generated. If you paste a link, change your mind at the settings screen and walk away, those base credits are already gone.
The August Result: 10 Clips vs 5
Same video, same settings, both models, 25 August 2026 (v1 now returns 5 clips and splices, see the update above):
| Model v1 | Model v2 | |
|---|---|---|
| Clips returned | 10 | 5 |
| Average clip length | ~38 seconds | ~59 seconds |
| Clip construction | One continuous run of the source | Several segments spliced together |
| Scoring | Single virality score | Virality plus Hook / Flow / Insight / Spread |
| Titles | Descriptive | Editorial — capitals, emoji |
| Export resolution | 720p | 720p |
| Watermark | Yes | Yes |
In August, Vizard's own description — v1 is "faster, more clips", v2 gives "fewer clips, more complete" — matched what we got: exactly half the clips, noticeably longer. In our October retest v1 returned 5 clips too.
August 2026: Model v1 returned 10 clips from the 10-minute source, each a single continuous segment, each scored out of 10. In October it returned 5.
v2 splices, and now v1 does too
In August this was the difference between the two models, and it is not in Vizard's changelog.
In August, every v1 clip was a single continuous run of the source. Start point, end point, done. In our 28 October retest, v1 joined separate moments too (see the update above).
v2 assembles clips from multiple non-adjacent moments. Its top clip was built from four separate points in the source — 03:55, then 05:15, 05:23 and 05:54 — stitched into one narrative.
That is a genuine editorial decision, not a longer trim. It means v2 can build a story arc that never existed contiguously in your footage. It also means v2 clips can cut between moments in ways you did not film, so if your video depends on continuity, check v2's output before publishing it.
Model v2 returned 5 clips from the identical source, with longer runtimes and an extra Hook / Flow / Insight / Spread breakdown under each score.
Longer source, more clips — but not proportionally
We also ran a 22:48 version of the same broadcast through v2. It returned 9 clips — up from 5 on the 10-minute cut, but for 2.3x the source length and 2.8x the credits.
What it does when several people are on screen
The 22:48 broadcast contains a press conference — four people answering questions from a room of reporters, shot as a single wide 16:9 frame. That is structurally the same problem as a multi-person podcast or a panel interview, and it is the case most likely to break a vertical reframe.
v2's top clip from that run was pulled from 18:57, inside the media Q&A, and scored 10.0. Its stated reason: "the most compelling clip in the batch — a raw, vulnerable moment of human connection."
It did not crop down to one speaker. It built a three-panel split screen: one person across the top, two side by side underneath, each a separate crop of that single wide shot.
Left: the source, one wide shot. Right: Vizard's 9:16 output from the same moment — three panels, each cropped from that one frame. The grid does not exist in the source.
The layout is fixed. We sampled frames across the clip and the panel structure never changed — it does not cut or switch to whoever is talking. So this is composition, not speaker tracking. Everyone stays on screen the whole time, whether they are speaking or not.
That treatment is specific to footage with people in it. A clip from the same run with no faces — an exterior shot of the spacecraft — was not cropped at all. It was letterboxed, 16:9 content sitting inside the 9:16 frame with black bars above and below.
Same run, same model, same export settings — footage with no people in it is letterboxed rather than reframed.
What that means if your content is conversation: expect everyone in shot to stay in shot, arranged into panels, for the whole clip. If you wanted the frame to follow whoever is speaking, Vizard did not do that here.
One limit worth stating plainly: our source holds a wide shot and never cuts between individual close-ups, so the model was never offered the choice of following a speaker. Whether it tracks speakers on footage that does cut between them is untested. This is also one clip from one run — and we have not tested a podcast, only footage that shares its shape.
The Scores Are Decorative
v2 shows a headline virality score out of 10 plus four sub-scores. It looks rigorous. We would not lean on it.
The top-scoring clip in our v2 run was rated 10.0 — while the model's own written rationale for that clip said it "lacks a strong hook or high-stakes information, making it the weakest in terms of shareability." Its sub-scores were Hook 7, Flow 7, Insight 8, Spread 9. None of that adds up to a 10.
Another clip scored 9.4 while being described as "somewhat dry" with no "wow moment".
There is also a range problem. Every v2 clip landed between 9.2 and 10.0, so the score cannot separate a strong clip from a weak one. v1's scores at least spread from 7.5 to 9.0.
Read the written rationale, ignore the number. The prose is genuinely useful and often quite honest about a clip's weaknesses. The score is not.
What The Premium Does Not Buy
Two things were identical across both models, and both matter more than clip count for most creators.
Captions. We scored Vizard's transcription against an official reference transcript of the same footage. On conversational speech it managed 2.75% word error rate over a 109-word sample — genuinely good. On technical, jargon-heavy speech it hit 10.0% over a 100-word sample. Every error was a proper noun or domain term: "optical comm" became "optical calm" repeatedly, "phased array" became "phase ray", and astronaut Christina Koch's surname came out as "Cook".
Those exact same errors appeared in both the v1 and v2 output. The clipping model does not touch transcription. If captions are your problem, changing model will not fix it — and note that the free plan shows you the transcript on screen but offers no subtitle file to export.
The per-clip export menu on the free plan: 720p is selectable, while 1080p, 2K and 4K each carry a paid badge — regardless of which model produced the clip.
Export quality. Both models exported at 720x1280 from a 1920x1080 source, both with the "made with Vizard.ai" watermark burned into the video file. On the free plan, 1080p, 2K and 4K all sit behind the paywall regardless of which model you choose.
The Trap: Your Model Choice Is Account-Wide
This one cost us credits, so learn it for free.
The model dropdown appears only on the import screen. Once a video is imported, open it later and there is no model selector anywhere — which makes it look like the model is locked to that project.
It is not. The model is an account-level setting applied at generation time. We imported a video at the v1 price, later switched the account to v2 for an unrelated test, then came back and generated the first video — and it processed on v2 and charged the premium.
Before you generate an old draft, check which model your account is currently set to. Otherwise you will be charged a premium you did not choose for a video you imported days earlier.
Which Should You Use?
Use Model v1 if you are on the free plan and want the most minutes per credit, you are testing whether Vizard suits your footage at all, or you plan to trim clips yourself anyway. Note that v1 now joins separate moments too, so check clips for continuity on either model.
Use Model v2 if you want clips closer to publishable without manual work, your source is loose and conversational so a spliced narrative genuinely helps, or you are producing a small number of pieces and quality matters more than volume. Just budget for roughly 30% fewer minutes per month.
For most small creators evaluating Vizard on the free plan, start with v1: it costs about 30% less per minute. Since October it returns a similar number of clips to v2 and joins moments as v2 does, so the case for paying v2's premium is weaker than it was in August. We have not re-tested v2 to compare the two directly again.
Testing Vizard for the first time? Start on the free plan with Model v1 — it costs fewer credits per minute, so you can judge whether the tool suits your footage before you commit to anything.
FAQ
Is Model v2 available on the free plan?
Yes. Both v1 and v2 are selectable on the free plan with visible prices. v2 is not locked behind an upgrade — it simply costs more credits per minute.
How many credits does Model v2 use?
About 1.3 credits per minute of source video, against roughly 1.0 for v1. A 10-minute video cost 13 credits on v2 and 10 on v1 in our testing.
Does Model v2 give better captions?
No. Caption accuracy was identical in our tests — the same transcription errors appeared in both v1 and v2 output. The clipping model does not affect transcription.
Does Model v2 export at higher quality?
No. Both models exported at 720p on the free plan from a 1080p source, both watermarked. Export resolution is a plan limit, not a model feature.
Can I switch an existing project from v1 to v2?
There is no per-project model selector after import. The model is an account-level setting applied when clips are generated, so changing your account setting and then generating an older draft will process it on the new model — and charge the difference.
Why did Model v2 give me fewer clips?
That is the intended behaviour. Vizard describes v2 as returning "fewer clips, more complete". In our test it returned exactly half as many as v1 from the same source, with each clip about 1.8x longer.
Related Pages
- Vizard AI review — the full walkthrough
- Vizard AI watermark — when it appears and what removal costs
- Vizard AI vs OpusClip — if you are still choosing a tool
- Quso.ai vs Vizard AI — the other free-plan comparison