Supported strength: Breadth of model access at a low entry price.
Watch out for: Output quality belongs to the model chosen, not to the platform.

Not establishedVerdict withheld

Read research

Scene generator

Runway Gen-4.5

Gen-4.5, text-to-video, 720p web

Supported strength: Visual composition and scene control.
Watch out for: Object-action binding; native audio announced but not yet on the scoped web route.

Moderately supportedProvisional findings

Read research

Scene generator

PixVerse

V6 text-to-video and image-to-video

Supported strength: Mid-table blind-preference placement.
Watch out for: Action and causal prompts; on-screen text; dialogue audio.

Limited evidenceProvisional findings

Read research

Scene generator

Luma Dream Machine

Ray 3 (benchmarked); Ray 3.2 is current

Supported strength: Documented output specifications well beyond the rest of this set.
Watch out for: No audio; identity drift between generations; most findings are the vendor's own admissions.

Limited evidenceProvisional findings

Read research

Scene generator

Vidu

Q3 Pro, and the faster Q3 tier (Turbo in API docs, Flash on the pricing page)

Supported strength: Transparent published pricing and the largest arena sample in this set.
Watch out for: Action, state and causal prompts; physics plausibility; the Pro tier is unmeasured.

Limited evidenceProvisional findings

Read research

Scene generator

Google Flow

Flow application over the Veo 3.1 family

Supported strength: Consistent mid-table placement across three Veo 3.1 tiers.
Watch out for: Every quality finding is model-level; the Flow route is never specified. Consistency in complex motion is an admitted weakness..

Limited evidenceProvisional findings

Read research

Scene generator

Sora 2

Sora 2 and Sora 2 Pro — discontinued April 2026

Supported strength: Placed well on blind preference while it was available.
Watch out for: The product is gone; most criteria can no longer be established at all.

Not establishedVerdict withheld

Read research

AI presenter

HeyGen

Avatar IV photo and video looks

Supported strength: The most completely published pricing of any tool in this set.
Watch out for: No independent evaluation of avatar or lip-sync quality was accepted.

Limited evidenceVerdict withheld

Read research

AI presenter

Synthesia

Express 3 stock avatars

Supported strength: Manual scene and brand editing on the platform.
Watch out for: Express 3 stock avatar quality is unsupported by independent testing.

Not establishedVerdict withheld

Read research

Scene generator

Topaz Video AI

Desktop application and Video API model set

Supported strength: Unusually specific documentation of purpose-separated models.
Watch out for: Desktop licence pricing not established; no independent quality evidence.

Moderately supportedProvisional findings

Read research

Scene generator

InVideo AI

Agent One (April 2026) over third-party generation models

Supported strength: One credit balance across an editor and several models.
Watch out for: Quality is not attributable to InVideo: the models underneath change.

Limited evidenceProvisional findings

Read research

Scene generator

CapCut

Standard and Pro tiers

Supported strength: Documented 8K export on the free tier.
Watch out for: Pricing is region-dependent and not published as a figure.

Not establishedVerdict withheld

Read research

03 / SIDE BY SIDE

Find the tradeoff that matters.

Compare real costs ↗

Choose a category, then compare findings and entry prices. Coverage measures evidence, not quality.

Published research only

8 tools · 8 researched in this category

Scroll the table to see all columns

AI video tools compared on evidence, coverage, findings and entry price
Tool	Evidence	Coverage	Strength / limitation	From	Free tier	Details
Kling AIVIDEO 3.0 Standard, text-to-video	Moderately supported	91%	Lighting and cinematic camera motionFast anatomy, precise physics and multi-shot continuity	—	Yes	Read research ↗
HiggsfieldPlatform layer over 15+ models (current); a July 2026 promotion listed 7 video, 11 image and 5 audio	Not establishedverdict withheld	27%	Breadth of model access at a low entry priceOutput quality belongs to the model chosen, not to the platform	—	—	Read research ↗
Runway Gen-4.5Gen-4.5, text-to-video, 720p web Model & plan notes Gen-4.5 needs the Standard plan or higher. 2–10 second clips at 720p.	Moderately supported	91%	Visual composition and scene controlObject-action binding; native audio announced but not yet on the scoped web route	$12/mo	Yes	Read research ↗
PixVerseV6 text-to-video and image-to-video	Limited evidence	73%	Mid-table blind-preference placementAction and causal prompts; on-screen text; dialogue audio	$10/mo	—	Read research ↗
Luma Dream MachineWe earn nothingRay 3 (benchmarked); Ray 3.2 is current	Limited evidence	64%	Documented output specifications well beyond the rest of this setNo audio; identity drift between generations; most findings are the vendor's own admissions	—	—	Read research ↗
ViduWe earn nothingQ3 Pro, and the faster Q3 tier (Turbo in API docs, Flash on the pricing page)	Limited evidence	73%	Transparent published pricing and the largest arena sample in this setAction, state and causal prompts; physics plausibility; the Pro tier is unmeasured	—	—	Read research ↗
Google FlowWe earn nothingFlow application over the Veo 3.1 family Model & plan notes Bundled with Google AI plans rather than sold separately.	Limited evidence	73%	Consistent mid-table placement across three Veo 3.1 tiersEvery quality finding is model-level; the Flow route is never specified. Consistency in complex motion is an admitted weakness.	—	—	Read research ↗
Sora 2We earn nothingSora 2 and Sora 2 Pro — discontinued April 2026 Model & plan notes Kept as a research record. There is no way to buy or use this product now, and we do not link to it.	Not establishedverdict withheld	36%	Placed well on blind preference while it was availableThe product is gone; most criteria can no longer be established at all	—	—	Read research ↗

Tools that make video from a text prompt or a still image.

How to read the evidence and prices

Confidence describes how strongly the sources support a finding. Coverage is the share of our criteria those sources answer, not a quality rating. “Not researched” means no profile has been published. A dash marks an unverified price or allowance. Free tiers may exclude the researched model. Full research method ↗

A CLOSER LOOK / THE DETAILS MATTER

See what to look for.

Depth, fine detail and physical contact: three visual cues to follow when you watch an AI-generated clip. Start with these illustrations.

01 / Space & framingAI illustration

Notice the depth.

When watching video In a moving shot, follow the edges: nearby objects should shift differently from distant ones. Watch for bending architecture or a changing cup shape.

How this image was made

Generator: OpenAI image generator
Created: 2026-09-12
Format: AI-generated still illustration

Create a premium editorial photographic illustration for Tested Clips, an independent AI video research website. Landscape 16:9, refined art direction, warm ivory and paper tones (#F4F2EB), near-black olive shadows (#151713), restrained chartreuse (#D8FF4F) and coral (#FF725E). Beautiful tactile materials, credible natural lighting, carefully composed, realistic detail. No text, no logos, no UI, no border, no watermark. Scene: a small matte chartreuse ceramic cup on an ivory stone plinth in a quiet architectural room. A dark olive doorway close to the lens frames the left edge, cup in the middle distance right of center, cream vertical columns far behind. Strong readable foreground/midground/background separation, angled late-afternoon light, precise long shadows. Concept: depth, camera framing and parallax. Sophisticated architectural photography, 50mm lens look.

02 / Faces & fine detailAI illustration

Look past the first impression.

When watching video In video, check whether these details stay consistent as a face turns. A convincing portrait alone cannot establish character consistency.

How this image was made

Generator: OpenAI image generator
Created: 2026-09-12
Format: AI-generated still illustration

Create a premium editorial photographic illustration for Tested Clips, an independent AI video research website. Landscape 16:9, refined art direction, warm ivory and paper tones (#F4F2EB), near-black olive shadows (#151713), restrained chartreuse (#D8FF4F) and coral (#FF725E). Beautiful tactile materials, credible natural lighting, carefully composed, realistic detail. No text, no logos, no UI, no border, no watermark. Scene: medium close-up portrait of a fictional adult woman in her thirties, short dark bob haircut, brown eyes, subtle natural freckles, plain muted coral cotton shirt. Three-quarter view, calm neutral expression, cream studio background, gentle window side light, natural skin texture, clearly visible eyes and hairline. Concept: facial detail and character identity. Tasteful editorial portrait, 85mm lens look; no beauty retouch or exaggerated fashion pose.

03 / Shadows & surfacesAI illustration

A coral wooden sphere and a dark olive block rest on a cream tabletop with visible contact shadows. — The shadows help place the sphere and block on the tabletop.

Find the point of contact.

When watching video In video, follow the contact shadow as an object moves. Look for floating, sliding or objects passing through each other.

How this image was made

Generator: OpenAI image generator
Created: 2026-09-12
Format: AI-generated still illustration

Create a premium editorial photographic illustration for Tested Clips, an independent AI video research website. Landscape 16:9, refined art direction, warm ivory and paper tones (#F4F2EB), near-black olive shadows (#151713), restrained chartreuse (#D8FF4F) and coral (#FF725E). Beautiful tactile materials, credible natural lighting, carefully composed, realistic detail. No text, no logos, no UI, no border, no watermark. Scene: a small matte coral wooden sphere resting on an ivory tabletop beside a fixed dark olive rectangular wooden block, a narrow visible gap between the two. Close-up low three-quarter camera angle with sphere left of block. Clear realistic contact shadow beneath sphere, soft light from upper left, subtle wood grain, physically grounded objects, spare architectural background. Concept: object boundaries, shadow and surface contact. Sophisticated still-life photograph.

These are AI-generated still illustrations made with OpenAI’s image generator, not outputs from the video tools we review. They illustrate what to examine; they do not demonstrate motion, establish consistency or contribute to our evidence ratings. Read our research method ↗

04 / OUR STANDARD

A finding you can follow.

Read the full method ↗

Match the model.

Research names the exact model and scope. Evidence about an older version is not quietly treated as current.

Weigh the sources.

We consider version match, relevance and recency. Every finding has a confidence label and cited sources.

Show the gaps.

Missing evidence stays missing. We withhold conclusions when the research cannot support them.

05 / BEFORE YOU SUBSCRIBE

A better first decision.

Start with your task. Check the model. Understand what you are paying for.

Explore the pricing guide ↗

Which kind of AI video tool do I need?

Choose a scene generator for footage from a prompt or image, an avatar tool for a speaking presenter, an enhancer for existing footage, or an editor to assemble a finished video. Compare tools that do the same job.

Does a free tier include the model in your research?

Not necessarily. A listed free tier describes the platform, not access to every model. Check the exact model, resolution, watermark and usage allowance before relying on a free plan.

How should I compare credits and subscription prices?

Start with the model and clip duration you need. Compare the cost of that generation, then allow for attempts you may discard. Our pricing guide normalises published prices to five seconds; it does not measure the cost of a successful result.

Do you test these tools yourselves?

Our findings come from cited public evidence, not controlled tests we ran. We identify the model researched, weigh the sources and publish what remains uncertain. The illustrations on this page are separate from that research.

What does evidence coverage mean?

Coverage is the share of our research criteria that the available evidence answers. It is not a quality score. A withheld verdict means the evidence does not meet our recommendation threshold.

IN THE OPEN

Independent, by design.

Some links earn us a commission at no extra cost to you. Findings are based on published evidence; commission rates do not determine what we publish. Tools that pay us nothing are included under the same standard.

Run by the team behind Video Editing Service. Found a missing source or changed price? Tell us.

AI video tools.Less hype.More evidence.

What are you making?

Start with the evidence.

Start with what we know.

Find the tradeoff that matters.

See what to look for.

Notice the depth.

Look past the first impression.

Find the point of contact.

A finding you can follow.

Match the model.

Weigh the sources.

Show the gaps.

A better first decision.

Independent, by design.

AI video tools.
Less hype.
More evidence.