Best AI Video Generators in 2026: Text-to-Video Models and What a Second Costs

On 6 November 2026, a second of Gen-4.5 video on Runway's Standard plan cost about 23 cents, the model topping the blind-vote board cost $0.20 a second for full HD, and Sora had shut down.

Hokai chart of pay-by-second video prices, from Kling's API price table read on 6 November 2026 and Hokai's model records.

This guide is about generating new footage from a prompt. Three neighbors do other jobs: AI video editors for cutting, captioning and clipping footage you already shot, AI for YouTube videos for a channel's whole workflow, and free AI for turning a photo into a video for animating one still.

Which generator fits which job

Those numbers matter because the models people now search for are sold by the second through an API, and the subscription tools have started reselling them.

The best AI video generator depends on the job. For a prompt-to-video clip, the strongest current names are Wan 3.0, MiniMax H3 and Seedance 2.5, the top three of them on the leaderboard below, plus Kling 4.0 for long takes of up to 30 seconds.

For a talking presenter, it is HeyGen or Synthesia. For a whole pipeline without a crew, from script to voice to edit, see how to create AI video without a production team. For footage you already shot, it is CapCut or Descript. Not sure which job you have? Answer a few questions in Smart Match and get a shortlist matched to your use.

What changed since 10 October 2026

This guide was first published on 10 October 2026 and re-judged on 6 November 2026, against the Artificial Analysis leaderboard and the vendor pricing pages read that day. The subscription and avatar picks hold. Two prompt-to-video picks changed: MiniMax H3 comes in on its arena score and price, and Gemini Omni 1.1 Flash moves down. Kling 4.0 replaces Kling 3.0 as the Kling pick. Six things moved in the market.

• Sora is gone. OpenAI closed the Sora app on 26 April 2026 and scheduled the developer API to end on 24 September 2026. Reference sources report the API as discontinued on that date. We could not open OpenAI's own notice to confirm after the fact, so treat Sora as unavailable and do not build on it. Our Sora shutdown guide has the timeline, and where Sora users went covers the move to Kling.
• Wan 3.0 leads, and MiniMax H3 overtakes Seedance 2.5. On Artificial Analysis's text-to-video arena with audio, read on 6 November 2026, Wan 3.0 scored 1172, MiniMax H3 1160 and Seedance 2.5 1155. We compare Wan and Seedance in Wan 3.0 vs Seedance 2.5.
• Gemini Omni 1.1 Flash moves down. It placed 10th on the same arena at 1129, below MiniMax H3, which costs about the same per second. It stays a cheap 720p option, and our Gemini Omni Flash guide covers what it is for.
• Runway now resells Kling 4.0. Its pricing page lists Kling 4.0 at 25 percent off inside Runway credits until 30 November 2026.
• Open weights arrived. MiniMax H3 and LTX-2.5 both ship weights you can run yourself.

Kling 4.0 takes Kling 3.0's place in the lineup

Kling 4.0 was announced on 30 September 2026. It generates 3 to 30 seconds in one pass where Kling 3.0 stopped at 15, accepts up to 10 keyframe images, and takes up to 15 reference assets through Omni Reference. A cheaper Kling 4.0 Flash covers 720p work up to 20 seconds; it reached Ultra Yearly subscribers first, on 28 September, two days before the 4.0 announcement.

Kuaishou lists 4K and HDR as coming soon, and its API price table shows 720P and 1080P only. Kling 4.0 was not on Artificial Analysis's arena when we read it on 6 November, so it is picked on clip length and reference control, not on a measured win. Kling 3.0 stays on sale at a lower price per second.

We weighed twelve challengers on 6 November: five for prompt-to-video, three for avatar video and four for editing, drawn from the arena, Hokai's newest listings and current roundups. Only MiniMax H3 beat a current pick on dated evidence. FLUX 3 Max, released in October, placed 4th at 1156, inside Seedance 2.5's margin of error, so it is one to watch rather than a pick yet.

Three products share one label, and that is the first thing to fix

Ask what footage you are starting with, and the category splits cleanly. Three buckets, three different jobs. If you have no footage and want a prompt to become a moving image, you need a generative model or an app built on one: Runway, Kling AI, Luma Labs Dream Machine, Pika or Vidu.

If you have a script and need a person to deliver it without a camera or an actor, you need an avatar tool: HeyGen or Synthesia. If you already have a recording, a podcast or raw clips and want AI to caption, cut, dub or repurpose them, you need an editor such as CapCut, Descript or VEED.

None of these three groups substitutes for the other two. A marketing lead who needs a talking-head product demo will not solve that with Runway, however good Gen-4.5 looks in a demo reel. A podcaster who needs faster editing will not solve that with Kling AI. The question "which AI video generator should I buy" only has an answer once you know which of the three jobs you have.

The criteria that separate these tools

Four questions narrow a long list to a short one, and they work whether or not you read the rest of this guide.

1. What are you starting with: nothing, a script or footage? This is the bucket question above, and it removes most of the category at once.
2. What does a second, or a minute, cost at your volume? Every subscription tool prices in credits, and the credit-to-dollar rate differs by vendor. The sections below do the conversion.
3. Does the free tier remove watermarks, or only let you preview? A free tool that cannot ship a usable, unbranded file is not free for commercial work.
4. Do you need output every month, or once? For a single campaign, paying by the second through an API or a top-up avoids a plan you will cancel after one use.

These questions matter more than any ranked list because you can test them against your own project before you pay. A ranking tells you what reviewers liked. The questions tell you whether the tool solves the job in front of you.

What the leaderboard says on 6 November 2026

Artificial Analysis ranks video models by Elo from blind pairwise votes, with one older model fixed at 1000 as the reference point. Scores within about 15 points of each other overlap at its 95 percent interval, so treat neighbours as ties.

The board is a preference vote on short clips with audio, not a test of long takes, character consistency or editing control, which is why Kling 4.0's 30-second clips and reference inputs still earn it a place without a score.

Subscription video tools: what a second of finished footage costs

Runway prices its Gen-4.5 model at 12 credits per second: its pricing page says 625 credits equal 52 seconds. Each plan is cheaper billed annually than monthly, and the table uses the annual rate. Max is described as the heavy-use plan, with credits that roll over for one month.

Kling AI charges 6 credits per second for 720p without audio, up to 12 credits per second for 1080p with native audio. Kling's plan page sends a signed-out visitor to its homepage, so on 6 November we could not re-read it; its plan figures come from Hokai's Kling record, last reviewed on 23 September 2026. The Standard price with 660 credits is a first-month promotion that renews at a higher rate.

Read the table as a ceiling on the gap, not a fixed saving. Kling Standard is about 30 percent cheaper than Runway Standard once the promotion ends, and the two meet near the top: Kling Premier at its renewal price costs about 12 cents a second, against roughly 10 cents on Runway Max. Runway's monthly-billed prices are higher still, at about 29 cents a second on Standard.

Luma and the smaller names

Luma Labs Dream Machine no longer sells a budget plan, and it has no free plan. It now resells other labs' models inside its credits, so the same plan buys different amounts of video depending on the model.

At the Plus rate, Luma's own Ray3.2 costs about 24 cents a second at 1080p, Kling 3.0 about 15 cents with audio, Kling 4.0 about 24 cents and Seedance 2.5 about 57 cents.

Two smaller names are worth testing before you commit to a bigger plan. Pika and Vidu each list an $8 entry tier in Hokai's records, reviewed in September, and both have a free tier. PixVerse and Higgsfield are two more services worth a look; Higgsfield lists access to all its models on its Pro plan.

Pay-by-second models: the new top end

The newest models are sold by the second rather than by subscription. You pay a vendor or a reseller for exactly the clip you generate, which suits a project with a fixed number of shots.

These figures come from Hokai's model records, last reviewed between 13 September and 26 October 2026, except two groups. The three Kling lines are Kling's own API price table, which we read on 6 November, with units converted at about $0.154 each, the rate on its $500 prepaid package. The Seedance 2.5 line is derived from Luma's own credit table, re-read on 6 November. Wan 3.0's Qwen Cloud page showed 30 percent off all three resolution tiers on 26 October; the chart uses the list price.

• Kling 4.0 Flash$0.09 Leader
• Gemini Omni 1.1 Flash$0.10
• LTX-2.5 Pro$0.12
• MiniMax H3$0.13
• Kling 3.0$0.18
• Wan 3.0$0.20
• Kling 4.0$0.23

Resolutions differ: Kling 4.0 Flash, Gemini Omni and LTX-2.5 Pro are 720p; the others are 1080p or 2K.

ByteDance's own API price for Seedance 2.5 was not confirmed when we checked, so the Luma figure is a reseller rate, not the vendor's.

How Kling 4.0 is billed

Kling 4.0 bills a clip's total duration in units: 1.0 a second at 720p and 1.5 at 1080p when no input video is supplied, and 0.6 and 0.9 when you give it a source video, with the billed length counting input plus output.

The prepaid packages start at $500 for 3,250 units and expire after 180 days with no rollover, so a one-off project pays for credit it cannot finish using. The price table lists no separate audio charge for 4.0 and no 4K price.

Against Kling 3.0 at 1080p with audio, the new model costs about a quarter more per second in exchange for clips twice as long. Our Kling 3.0 record keeps the older generation's figures for comparison.

Wan, Seedance and Gemini Omni in detail

Wan 3.0, from Alibaba Cloud, bills by resolution, from 480p up to 1080p, and a full 30-second clip at the top tier comes to about $6. Seedance 2.5, from ByteDance, takes up to 50 reference inputs in one request and costs more per second through resellers. CapCut's Pro plan advertises Seedance 2.5 from $0.12 a second in the first month, though that is a promotional rate. For a side-by-side of the two, read Wan 3.0 vs Seedance 2.5.

Gemini Omni 1.1 Flash is priced in tokens: $17.50 per million output tokens, converted at 5,792 tokens per second of 720p video, which comes to about $0.10 a second. Google lists a single standard tier with no free option. Google's Veo 3.1 remains a line to look at for cinematic shots, though it placed 22nd on the arena; Luma's credit table on 6 November put it at about 42 cents a second at 1080p with audio.

Open weights you can run yourself

MiniMax H3 and LTX-2.5 are the self-hosting options. Both publish open weights, so a team with its own GPUs can run them for the cost of compute. LTX-2.5 is free for commercial use under the LTX-2.x Community License for companies under $10 million in annual revenue. LTX Studio is the hosted app from the same lab for people who do not want to run anything.

Is there a free AI video generator?

Yes, but "free" means a trial or a daily allowance, not unlimited commercial output. Here is what each service offered when checked.

• Runway: 125 one-time credits, which is a trial and not a monthly allowance.

• Kling AI: about 66 daily credits and no payment card, per Hokai's September record.

• Luma: no free plan on its current pricing page.

• HeyGen: 3 videos a month, each up to 1 minute and watermarked, on the free plan.

• Synthesia: 500 free credits a month, about 10 minutes of video, with a Synthesia watermark.

• CapCut: a free tier with basic editing and free templates.

• Seedance: limited free daily credits. Our guide to using Seedance for free lists where.

The practical test is the watermark. If you need to publish the result for a client or an ad, budget for the first paid plan, not the free one. For character animation and dance clips of the kind Viggle makes, our Viggle alternatives guide covers free and no-watermark options.

Is there an AI video generator app?

Several of these run on a phone. Runway lists iPhone and Android apps on its pricing page, and HeyGen now offers a plugin inside ChatGPT. CapCut is the editor most people already know on mobile. Kling sells its memberships through a web app. If you work mostly from a phone, the real constraint is the credits, not the app, so the plan prices above are the ones to compare.

Avatar video: HeyGen's flat pricing beats Synthesia's credit pool

HeyGen charges 20 credits per minute for its premium Avatar IV and V models, a rate its record carries from the September review. Synthesia sells a shared credit pool: its pricing page on 6 November said Starter's credits make up to about 12 extra minutes a month and Pro's about 60, on top of a free 10 minutes. First months are discounted to $14 and $49.

That puts Synthesia at about 1.5 to 2.5 times HeyGen's per-minute price. HeyGen also lists a Business plan at $149 a month with seats at $20 each, and its Creator plan now includes its own text-to-video model, which placed 6th on the arena. Synthesia's Enterprise plan adds SSO, SCORM export and brand kits, while HeyGen's plans are built around output volume.

A five-person startup producing weekly product updates gets more minutes for less money from HeyGen. A regulated company standardizing training content may be paying for Synthesia's governance features, not just its video.

AI editors: where the free tier stays free, and where it does not

CapCut prices differ by country. Its pricing page, opened from Indonesia, listed Pro at IDR 188,000 a month with 500 monthly AI credits and 1 TB of storage, and an Ultra tier with 15,370 credits. Amounts in US dollars may differ, so check the page for your country. The free tier covers templates and basic editing.

Descript is the stronger pick for anyone editing by transcript, which describes most podcast-to-video and interview repurposing.

Hobbyist adds watermark-free 1080p export, and Creator adds 4K.

Neither tool is trying to win the other's job. CapCut suits short-form social cuts and templates. Descript suits long-form transcript editing, where cutting a word cuts the matching frame. inVideo and VEED sit between the two for teams that want templates plus AI generation.

Who should skip this category entirely

Not everyone shopping the "AI video generator" label needs a subscription. A team that needs one polished 90-second explainer for a single launch is often better served by a freelance editor, or by paying per second through an API, than by a recurring plan it will forget to cancel.

A solo creator whose bottleneck is finding a repeatable hook, not production quality, will get more from studying what performs in their niche than from any tool here. And a team that already pays for an editor, a colorist and motion graphics will not replace that pipeline with these products. The honest use for this category is closing gaps in a workflow, not replacing a team that exists.

A fourth group has a legal rather than a creative requirement. Training content that needs indemnification language, or footage that goes through a compliance review, rules out many smaller tools until each vendor's terms are checked. No per-second calculation here replaces that review.

The case against optimizing for cost per second

The obvious objection to all the math above is that cost per second stops deciding once quality varies enough between vendors. A brand shooting a hero ad will not pick the cheaper model to save a few cents if the pricier one holds a character steady across a ten-shot sequence and the cheaper one does not. That trade-off is real, and we have not run a blind test that settles it.

The math still matters for the larger group producing volume: social ads, product variants, translated versions of one clip, or a weekly cadence that repeats dozens of times a month. At that scale a 30 to 50 percent gap in per-second cost becomes a real budget line, and it is the number most rankings never compute.

Vendors are also repricing. Luma's cheapest tier is gone, Runway shows a monthly and an annual rate, and Synthesia rebuilt its plans around credits. Re-check any price on the vendor's page before you pay.

What we re-checked, and what we could not

On 6 November 2026 we read Artificial Analysis's text-to-video arena, Kling's API price table, and the pricing pages for Runway, Luma, HeyGen, Synthesia, Descript and CapCut. Most plan prices were unchanged from 10 October; Synthesia now states minutes per plan, and one reseller rate fell.

Kling's plan page did not open for a signed-out reader, so its consumer figures rest on Hokai's record, last reviewed on 23 September. OpenAI's Sora notice was blocked by a bot check, so the Sora dates rest on the published schedule. Per-second prices for Wan 3.0, Gemini Omni 1.1 Flash, MiniMax H3 and LTX-2.5 come from Hokai's model records, not from a fresh read of each vendor page.

Runway and Kling remain the two subscription tools worth pricing first. What changed is the model ranking: Wan 3.0, MiniMax H3 and Seedance 2.5 lead on blind votes, Kling 4.0 is the long-take pick, and Sora is no longer in the comparison.

Frequently asked questions

What is the best AI video generator in 2026?

It depends on the job. For prompt-to-video, Wan 3.0, MiniMax H3 and Seedance 2.5 led Artificial Analysis's arena on 6 November 2026, and Kling 4.0 handles 30-second takes; for a talking presenter, HeyGen or Synthesia; for editing existing footage, CapCut or Descript. OpenAI's Sora is no longer available.

How much does a second of AI video cost?

On Runway's Standard plan, about 23 cents a second of Gen-4.5 billed annually, falling to about 10 cents on Max. Kling AI works out to about 16 cents on Standard after its first-month promotion, per its September record. Pay-by-second models run from about $0.09 to $0.23 a second at 720p to 1080p.

Is there a free AI video generator?

Several offer a trial or daily allowance. Runway gives 125 one-time credits, Kling AI about 66 free credits a day, HeyGen 3 watermarked videos a month up to 1 minute, and Synthesia 10 minutes with a watermark. Luma has no free plan. None replaces a paid plan for commercial work.

Is there an AI video generator app?

Runway has iPhone and Android apps, CapCut is widely used on mobile, and HeyGen has a plugin inside ChatGPT. Kling sells memberships through a web app. Credits, not the app, decide what a clip costs, so compare plan prices rather than platforms.

Is OpenAI's Sora still available?

No. OpenAI closed the Sora app on 26 April 2026 and scheduled the API to end on 24 September 2026. Reference sources report it as discontinued on that date. Kling AI, Runway, Wan 3.0 and Veo 3.1 are the usual replacements.

Covered in this guide

• runway: AI to simulate the world through generative video and world models
• alibaba-cloud: Alibaba Cloud, founded 2009, is China's top cloud provider (35% share) and builds the Qwen AI models, which passed 1 billion Hugging Face downloads in 2026.
• bytedance: ByteDance is a Beijing-based consumer-internet group whose short-video apps and Seed AI models power the Doubao assistant.
• capcut: AI-powered video editor for creating trending content across all platforms
• descript: AI-powered video & audio editing as easy as editing text
• FLUX 3: FLUX 3 is a video, image, and audio model from Black Forest Labs, launched July 2026, built on a shared Self-Flow architecture.
• gemini-omni-1.1-flash: Gemini Omni 1.1 Flash (Google, GA 2026) generates and edits AI video through chat, adding native synchronized audio and multi-pass scene extension.
• heygen: Fast, simple AI video generator that creates professional videos with realistic avatars in 175+ languages.
• higgsfield: AI video and image studio, used in 238 countries, that runs Kling, Veo, Seedance and Sora plus its own camera and character tools on one credit plan.
• invideoio: Create videos without limits. Turn any idea into videos with AI.
• kling-3.0: Kling 3.0 is Kuaishou Technology's proprietary AI video model, built for native 4K output with synchronized multilingual audio in one generation pass.
• Kling 4.0: Kling 4.0 is Kuaishou's AI video model for long single-take clips with multi-keyframe control, Omni Reference and stereo audio, announced in September 2026.
• kling: Kuaishou's text-to-video platform with native 4K, lip-synced audio in 5 languages, and free daily credits. Used by 60M+ creators; plans from $6.99/month.
• ltx-studio: Browser-based AI video studio from Lightricks that turns scripts into storyboards, consistent shots and timeline edits, with partner models like Veo and Kling.
• ltx-2.5: Lightricks' open-weights video and audio model that renders synchronized 4K HDR clips from text, image or video input, self-hostable and fine-tunable.
• luma-labs-dream-machine: Creative agents that make you prolific: AI video and image generation for creators
• minimax-h3: MiniMax H3 (Hailuo 3.0) is a 33B open-weight omni-modal model, released July 31 2026, generating 2K/24fps video with native stereo audio in one forward pass.
• pika: AI creative platform by Pika Labs. Generates video, image and audio from prompts or photos using its own model plus outside AI engines. Free to start; paid plans scale up from there.
• pixverse: AI video generator valued at over $2 billion, with native audio, camera controls and a free plan; paid tiers available.
• seedance-2.5: Seedance 2.5 is ByteDance's 2026 video generation model, producing one continuous multi-reference clip per pass with synchronized audio, launched inside Jimeng AI and Doubao Pro.
• synthesia: AI video platform transforming text into professional, multilingual videos with realistic avatars
• veed: Browser-based video editor with AI subtitles, avatars, voice cloning, and text-to-video: no download needed.
• veo-3.1: Google DeepMind's Veo 3.1 generates 4, 6 or 8-second video clips with native synchronized dialogue and audio, rendering up to 4K.
• vidu: AI video generator with text-to-video, image-to-video, and reference-to-video modes, native audio-video co-generation up to 16 seconds, and multi-subject consistency: starting free with paid plans.
• wan-3.0: Tongyi Lab's public-beta video generator (since August 6, 2026) makes native single-shot clips up to half a minute long from text, image, audio, video or documents.

Sources