Research profile · Agent One (April 2026) over third-party generation models
InVideo AI, researched.
One credit balance across an editor and several models. Provisional findings. Explore the findings, limitations and 3 cited sources below.
InVideo AI is supported as an assembly layer: the agent plans, iterates and holds project context across hundreds of generations. What it cannot do is fix what the model underneath gets wrong — the one professional case we have reports motion discontinuity, recurring artefacts and a scale error that survived every prompt. The verdict rests on a single studio's account and is provisional.
- Researched scope
- Agent One (April 2026) over third-party generation models
- Best supported for
- Assembling a finished edit when model choice is left open
- Principal limitation
- Quality belongs to the model underneath; evidence is one production case
- Accepted sources
- 3 · 6 of 9 criteria supported
Best for, and not for
Best supported use
Assembling a finished edit when model choice is left open. The strongest finding is one credit balance across an editor and several models.
Where the evidence warns
Quality belongs to the model underneath; evidence is one production case.
What the evidence supports, criterion by criterion
6 of 9 applicable criteria have enough accepted evidence to say something. The rest are listed below them, unscored.
| Criterion | Evidence | What we found |
|---|---|---|
| Temporal stability and artifact control | Limited evidence | On a real 30-second commercial, joins between generated clips showed visible motion discontinuity; the agent revised its plan from six short clips to two 15-second clips joined during a hover to hide it.A single production case by one studio. The underlying model was Seedance 2.0, so the defects belong partly to the model rather than to InVideo's application layer. |
| Prompt or script adherence and control | Limited evidence | Across 15 story beats and close to 300 generations, the studio reports useful iterative reasoning from the agent alongside recurring artefacts, odd human behaviour and a persistent object-scale failure it could not prompt away.One project, one team, weeks of use. Candid and specific, but not a controlled measure, and the model underneath was Seedance 2.0. |
| Workflow and editability | Limited evidence | Autopilot mode offers Basic, Pro and Ultra generation quality, with one credit balance spent across an editor and a set of third-party video models.Feature documentation only; the underlying models change, so quality is not attributable to InVideo itself. |
| Output specifications and delivery | Limited evidence | 720p is documented as standard on the free plan, with 1080p and 4K gated behind payment. Free tier allows 10 minutes of video per month.Resolution ceilings are documented per tier; per-model output specifications are not. |
| Pricing and value | Limited evidence | Published tiers: Plus $20/month, Max $100, Generative $200, Elite $1,000. Video models are billed per second of output, so a Kling 2.6 Standard clip may cost around 0.5 credits while a Veo 3.1 4K clip with audio exceeds 10.The spread between tiers is wide and the per-second model cost sits on top, so effective cost depends heavily on which model is chosen. |
| Character and object consistency | Limited evidence | Project context carried across hundreds of iterations, but continuity in the final video could still drift and the underlying model kept its own consistency limits.Single case; consistency failures cannot be attributed solely to InVideo when the generation was done by a third-party model. |
Not established
Researched, but the available public evidence does not support a conclusion. These are gaps, not low scores.
- Visual fidelity and aesthetics — Researched; no source meeting our weighting bar addressed this for the scoped version.
- Text rendering — The final bounded professional-review pass on 13 September 2026 found no exact-current-version result for text rendering.
- Native audio and lip sync — The final bounded professional-review pass on 13 September 2026 found no exact-current-version result for audio.
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 InVideo AI pricing scenarios and how they were normalised. Prices verified 2026-09-13 and change without notice.
Alternatives
Same category, so genuinely comparable. We do not rank across categories.
- CapCut — Free general-purpose editing with AI helpers, strongest for short-form social. Not established. We earn nothing
Sources
Every finding above traces to one of these. Version match records how closely each source matches Agent One (April 2026) over third-party generation models.
| Source | Type | Version match | Accessed |
|---|---|---|---|
| InVideo | Official documentation | Exact | 2026-09-12 |
| InVideo | Official documentation | Exact | 2026-09-12 |
| Lambda Films (London production company) | Professional review with method | Exact | 2026-09-13 |