Research profile · V6 text-to-video and image-to-video
PixVerse, researched.
Mid-table blind-preference placement. Provisional findings. Explore the findings, limitations and 6 cited sources below.
PixVerse V6 is supported for fast short-form work with a published per-second rate and, unusually for this set, independent evidence on motion, text and audio. That evidence is thin and points the same way: motion is mid-table and clean at the sample, instruction following collapses on action and cause, text rendering drops words, and native audio syncs effects but not reliably dialogue. Eight of eleven criteria carry a source; most carry one.
- Researched scope
- V6 text-to-video and image-to-video
- Best supported for
- Fast short-form clips where a mid-table result at a low per-second rate is the point
- Principal limitation
- Action and causal prompts; on-screen text; dialogue audio
- Accepted sources
- 6 · 8 of 11 criteria supported
Best for, and not for
Best supported use
Fast short-form clips where a mid-table result at a low per-second rate is the point. The strongest finding is mid-table blind-preference placement.
Where the evidence warns
Action and causal prompts; on-screen text; dialogue audio.
What the evidence supports, criterion by criterion
8 of 11 applicable criteria have enough accepted evidence to say something. The rest are listed below them, unscored.
| Criterion | Evidence | What we found |
|---|---|---|
| Visual fidelity and aesthetics | Limited evidence | V6 sits mid-table on blind pairwise preference — Elo 1077 from 9,949 votes, ahead of its own V5.6 at 950.Arena Elo is aggregate human preference, not a measure of any single quality attribute. |
| Temporal stability and artifact control | Limited evidence | VGIF-Score places V6 mid-table on motion smoothness at 44.84 on a 0-100 scale, in a field running from 26.28 to 55.76 across fourteen models. A structured video comparison at 1080p and 15 seconds reports clean image-to-video motion with no morphing where Veo 3.1 showed artefacts on the same prompt.The benchmark score is an automatic rubric judgment, not human-rated. The video review is one creator on roughly six prompts. Two independent sources agree in direction; neither is strong alone. |
| Prompt or script adherence and control | Conflicting evidence | Instruction following holds for what is in the shot and collapses for what happens in it. VGIF-Score scores V6 at 75.4 on entity and 81.2 on location, against 25.9 on action, 15.5 on state and 4.2 on causal reasoning across 223 prompts.The pattern is the finding; the absolute figures are not. The paper's own composite does not reconcile with its repository normalisation, the table's scale is unlabelled, and scoring is by automatic judge rather than human annotators. |
| Workflow and editability | Moderately supported | V6 documents five generation routes — text, image, first-and-last-frame, extension and reference-to-video — with multi-clip on the text and image routes, plus a platform-level Modify tool for swapping subjects, adding or removing elements and restyling.A capability listing, not a measure of how reliably any of it works. Modify is a platform feature not tied to V6 specifically. |
| Output specifications and delivery | Moderately supported | Official V6 docs specify 1-15 second clips at 360p, 540p, 720p or 1080p, with text-to-video, image-to-video, first-and-last-frame transition and video extension.Documentation establishes what is offered, not how well it performs. |
| Pricing and value | Moderately supported | Published per-second credit rates: 9 credits/s at 720p without audio, 12 with; 18 credits/s at 1080p without audio, 23 with. Starter pack quoted at $1 for five 5-second 720p clips.Rates are official and current; value depends on retry rate, which we have no accepted evidence for. |
| Text rendering | Limited evidence | On a bilingual signage prompt, V6 rendered 'Grand' and dropped 'Hotel', and the Chinese characters came out incomplete. Veo 3.1 handled the same text better on the same prompt.A single stated test by one creator, verdict spoken on the recording. The first evidence of any kind on this criterion after official documentation was found to say nothing. |
| Native audio and lip sync | Limited evidence | Native audio was tested with the switch on at 1080p: a requested bottle-cap sound synced correctly, while in a separate clip a character's voice was missing from the track. The reviewer's own comparison found Veo 3.1 stronger on dialogue.One creator, qualitative, settings stated. Sound-effect sync and dialogue reliability are different things and the sample supports only the former. |
Not established
Researched, but the available public evidence does not support a conclusion. These are gaps, not low scores.
- Camera motion — Searched the VBench and VBench-2.0 leaderboards on 13 September 2026 by full model-column inspection. No version matching our scope appears in them.
- Character and object consistency — Searched the VBench and VBench-2.0 leaderboards on 13 September 2026 by full model-column inspection. No version matching our scope appears in them.
- Physics and material realism — Searched the VBench and VBench-2.0 leaderboards on 13 September 2026 by full model-column inspection. No version matching our scope appears in them.
Pricing
Published list price, normalised to a five-second generation. Assumes full use of the plan allocation, before tax, no retakes — so real cost per clip you keep is higher.
| Plan | Cost / 5s | Resolution | Audio |
|---|
All PixVerse pricing scenarios and how they were normalised. Prices verified 2026-09-14 and change without notice.
Alternatives
Same category, so genuinely comparable. We do not rank across categories.
- Kling AI — Complex physical movement and holding a character's face steady across a shot. Moderately supported.
- Higgsfield — Camera-move control and stylised cinematic shots, with access to several models in one place. Not established.
- Runway Gen-4.5 — Creative and editorial work — generative editing, motion brush and in-context controls. Moderately supported.
Sources
Every finding above traces to one of these. Version match records how closely each source matches V6 text-to-video and image-to-video.
| Source | Type | Version match | Accessed |
|---|---|---|---|
| PixVerse | Official documentation | Exact | 2026-09-12 |
| PixVerse | Official documentation | Exact | 2026-09-12 |
| Artificial Analysis | Independent benchmark | Exact | 2026-09-12 |
| Songyu Xu et al. (PRCV 2026) | Academic evaluation | Exact | 2026-09-13 |
| PixVerse | Official documentation | Family | 2026-09-13 |
| Skill Destiny (YouTube, 6.5K subscribers) | Professional review with method | Exact | 2026-09-14 |