How to get AI to
help me choose an AI video generator.

What you’ll be able to do

By the end you will have one AI video generator picked for the video you are actually making, a clear idea of what free gets you, and a rule for when paying is worth it. You will also have run the same test prompt on two tools, so you choose from what you have seen rather than from a list on a website.

“Best AI video generator” lists put everything on one level, but these tools do different jobs. Some make a ten-second scene from a sentence. Some make a person talking to camera from a script. Some are editors with generators tucked inside. Pick the job first, then the tool.

One warning up front: Sora is gone. OpenAI announced on 24 March 2026 that it was discontinuing Sora in the consumer app and API (Engadget quotes the announcement), so any guide that still tells you to use it is out of date.

Quick picks: what are you making?

• Want a quick clip or a photo that moves? Gemini Omni and Google Flow: ten-second clips with sound, inside Google’s AI plans.
• Want a realistic or cinematic shot? Runway: the clearest free trial of the specialist generators, and it hosts Kling too.
• Want sound and a scene built step by step? Kling: Kling 4.0 with its Canvas workspace.
• Happy to pay from day one? Luma: paid plans only, commercial use on Plus.
• Want to test for free without a watermark? Pika: free plan, but not a commercial licence.
• Need to join, caption and polish clips? CapCut: an editor first, with AI models inside.
• Need a person talking to camera? Synthesia or HeyGen: script in, presenter out.
• Need a launch video for your website, free? Screen-record the site, cut it in CapCut, and let an assistant write the script.

Before you start

These tools sort into four tiers. Work out which tier matches your video, then pick inside it.

The video generator scale

• Tier 1: Inside your assistant: video generation that comes with an AI plan you may already have (Gemini Omni, Google Flow).
• Tier 2: Specialist app: you spend credits on each clip or order (Runway, Kling, Luma, Pika).
• Tier 3: Editor with AI: an editor first, with generators and AI credits inside (CapCut).
• Tier 4: Presenter avatar: not scenes, but a person speaking your script (Synthesia, HeyGen).

Comparison chart

Generators make clips of seconds, not films, and you should expect several tries for every one you keep. To show your real product, use a screen recording: a generator will invent something that looks a bit like it. Vendor details here were checked on their own pages on 4 November 2026, and prices change, so each block points to the pricing page.

Get set up

Pick one tool from the tier that fits your video.

For a launch video, the free route is to screen-record your site or product and cut it together in CapCut, with an assistant writing the script. “How to get AI to make a launch video” walks through both routes.

Gemini Omni and Google Flow

Tier 1: Inside your assistant

Best forquick clipsphotos into videoediting by chat

Gemini Omni has replaced Veo in the Gemini app. Gemini Omni 1.1 Flash makes 10-second videos with native audio, can turn up to five photos into a video, and you edit by chatting to it. It needs a Google AI subscription and features vary by plan and country. Videos carry Google’s SynthID watermark. On the plans page the free plan lists limited Google Flow access, Google AI Plus lists 200 Flow credits, Pro 1,000, and Ultra 10,000 or 25,000. Google Flow (flow.google.com) is Google’s AI creative studio for video scenes and stories, run on the same credits.

Cost: limited on free, then a paid Google AI plan; check gemini.google/subscriptions. Skip it if: you cannot have a watermark on the finished video.

Runway

Tier 2: Specialist app

Best fortext to videoimage to videotrying several models

Runway’s free plan is 125 one-time credits, and once they are gone there is no monthly top-up. Paid plans start in the low tens of dollars a month and list “No watermarks” and 4K upscaling. Gen-4.5 is Runway’s own model, and it also hosts Kling 4.0 and other third-party models, so you can try a couple of engines in one place.

Cost: free to try, then a monthly plan; check runway.com/pricing. Skip it if: you do not want to watch a credit count.

Kling

Tier 2: Specialist app

Best fortext to video with audioimage to videoscene building

The current model is Kling 4.0, and the site also offers an agent-driven workspace called Canvas for planning scripts, images and videos in one place. Its pricing page did not load for us, so the free allowance and whether free clips are watermarked are unconfirmed here. Look at the page before you plan around it.

Cost: check kling.ai. Skip it if: you need to know the cost and watermark rules before you start.

Luma

Tier 2: Specialist app

Best forpaid-for generationcommercial usemixing models

Luma’s pricing page lists paid individual plans only (Plus, Pro and Ultra), and Plus lists commercial use. Credits buy Luma’s own models and some third-party ones. No free plan is shown, so if you want to test first, start elsewhere.

Cost: paid only; check lumalabs.ai/pricing. Skip it if: you want to try it free first.

Pika

Tier 2: Specialist app

Best forfree testingno-watermark clipssocial posts

Pika has a free plan with all models available through credit packs, and free exports carry no watermark. Free is not a commercial licence, though. The paid plans (Starter, Creator, Fancy) include one. So free is fine for experimenting or posting for fun, and not for selling or advertising with.

Cost: free to start, paid plans add the licence; check pika.art/pricing. Skip it if: you are making an ad and need the licence from day one.

CapCut

Tier 3: Editor with AI

Best forjoining clipscaptionspolishing

CapCut is an editor first. It lists AI video models inside (Seedance and Veo 3.1 among them), but its free plan is templates and basic editing tools, and AI credits come with Pro and Ultra. Use it free for cutting, captions and music, and expect to pay if you want it to generate.

Cost: free for editing, paid for AI credits; check capcut.com/pricing. Skip it if: you want a generator and have no clips to edit.

Synthesia

Tier 4: Presenter avatar

Best fortraining videosexplainerstalking presenters

It does not make scenes. You write a script, pick an avatar, and it makes a person saying your words. The Basic free plan carries a Synthesia watermark, and paid plans remove it.

Cost: free with watermark, paid to remove it; check synthesia.io/pricing. Skip it if: you want a scene rather than a person talking.

HeyGen

Tier 4: Presenter avatar

Best forpresenter videosan avatar of yourselfmany languages

The same job as Synthesia. Free is 3 watermarked videos a month, up to a minute each. The Creator plan removes HeyGen branding and adds a custom avatar with your own voice and likeness.

Cost: free with limits, then Creator and up; check heygen.com/pricing. Skip it if: you want a scene, or are not comfortable uploading your own face and voice.

Try it yourself

Pick two free options and run the same prompt on both. That’s the whole test. Reading about them only gets you so far, because you don’t know what “free” feels like until the credits start dropping.

If you’re not sure which tier you need, ask an assistant first. ChatGPT, Claude, Gemini or Copilot will all do this. Paste this in and fill the brackets:

I want a video for [purpose] on [platform], about [length]. Which of these should I use: an AI video generator, an AI presenter, or an editor with templates? Ask me three questions first, then recommend one.

Answer its three questions properly. It’s doing the sorting the tiers above do, but with your actual video in mind.

Now the test clip. Keep it to one scene and one action, because more than that is where these tools fall apart:

A 10-second shot of [one scene]. Camera: [slow push-in / static / handheld]. Light: [time of day]. One action only: [what happens]. No text on screen, no logos, no real people.

Something like: a barista pulling an espresso shot, slow push-in on the portafilter, warm morning light, espresso flowing into a white cup. Run exactly the same words on both tools.

If your idea is bigger than one shot, don’t cram it into one prompt. Split it:

Turn this idea into a shot list of five 5-second shots, each with one subject and one action, so I can generate them separately: [idea]

Then generate each shot on its own and join them in an editor. Shorter shots usually look better and cost fewer credits.

Check the result

Run these on both clips. Don’t just trust your first impression, it’s easy to be wowed for about four seconds.

Faces and hands. Pause and zoom. Count fingers. Look at teeth, eyes, where a hand meets a cup. If you asked for no real people, check nobody’s wandered into the background anyway.

Motion. Does it look like it has weight? A cup should be lifted like it’s got something in it, a door should turn on its hinge. If it floats or slides, that’s a tell.

Text. If any writing appears, read it. Garbled letters are common, even when you asked for none.

Prompt match. Go back to your prompt and tick off each part: camera, light, the one action. Write down what it missed.

Credits per usable clip. This is the one people skip. If it took four goes to get a keeper, that’s four goes of credits. Divide what you spent by what you can actually use, and you’ll know what free really gets you.

Watermark. Is it there, and could you live with it where you’re posting?

Then two things before you publish. Labels: YouTube requires you to disclose realistic AI-generated or altered content through the “AI use” setting in YouTube Studio. That means a real person made to say or do something they didn’t, altered real footage, or a realistic scene that didn’t happen. Instagram and TikTok each have their own AI label, so check each app. The guide “How to get AI to make short videos” covers it. Commercial terms: check them before you sell or advertise with the output. Free plans often differ from paid, and Pika’s free plan is a good example.

If it doesn’t work

Garbled text. Don’t fight it. Generate the clip without text and add the words in an editor like CapCut.

Wrong motion. Shorten the shot and ask for one action only. If it still floats, try the other tool, because the same prompt can behave quite differently on a different model.

Credits gone. Switch to another tool, or pay for one month only and finish the job in that month. You don’t have to stay subscribed.

You actually need a presenter. Then you’ve picked the wrong tier. Use Synthesia or HeyGen, not a scene generator.

Privacy and what not to hand over. Never put a real person’s face, voice or likeness into a generator without their consent. That includes photos of friends, colleagues and customers. Never make fake news footage, or a fake endorsement where someone appears to back something they haven’t. Don’t paste in private details, client information, passwords or anything under an NDA, because you don’t know how a given tool stores or uses what you upload, so read its terms. And don’t upload your own face and voice to a custom avatar unless you’re comfortable with how that tool handles them.

Questions people ask

Which AI video generators are free to start with?

Pika has a free plan with no watermark on exports, Runway gives 125 one-time credits that do not refill, Synthesia's Basic free plan is watermarked, and HeyGen's free tier is 3 watermarked videos a month of up to a minute each. CapCut is free for editing, but AI credits come with Pro and Ultra.

Can I use free AI video clips for adverts or selling?

Check the commercial terms before you sell or advertise with the output, because free plans often differ from paid. Pika's free plan is a good example: free exports carry no watermark but it is not a commercial licence, while the paid Starter, Creator and Fancy plans include one.

Should I still use Sora for AI video?

No. OpenAI announced on 24 March 2026 that it was discontinuing Sora in the consumer app and API, so any guide that still tells you to use it is out of date.

How do I test two AI video generators properly?

Pick two free options and run exactly the same one-scene, one-action prompt on both, specifying camera, light and the single action. If your idea is bigger than one shot, split it into five 5-second shots, generate each separately and join them in an editor.

What should I check before publishing an AI-generated video?

Pause and zoom on faces and hands, check motion has weight, read any text that appears, tick off each part of your prompt, count credits per usable clip and decide if the watermark is liveable. YouTube requires you to disclose realistic AI content via the "AI use" setting in YouTube Studio, and Instagram and TikTok have their own labels.

Keep exploring

Related guides: “How to get AI to make a video”, “How to get AI to make short videos” and “How to get AI to animate a photo”. To pick an assistant first, see “How to get AI to help me choose an assistant”. For the stills that often start a video, “How to get AI to help me choose an AI image generator” sorts the image tools the same way.

Our own sites: isthereathing.com gives an honest yes or no on whether a tool exists for a job, with what it costs and the catch, and canyouuseai.com gives a verdict on whether to use AI for something, with the UK rules.

Sources and review notes

https://gemini.google/gb/overview/video-generation/ https://gemini.google/subscriptions/ https://flow.google.com/ https://runway.com/pricing https://kling.ai/ https://lumalabs.ai/pricing https://pika.art/pricing https://www.capcut.com/pricing https://www.synthesia.io/pricing https://www.heygen.com/pricing https://launchvideomaker.com/ https://engadget.com/ai/openai-is-shutting-down-its-sora-video-generation-app-211023358.html https://support.google.com/youtube/answer/14328491

Vendor details were checked on their live pages on 4 November 2026. Kling’s pricing page would not load, so nothing is claimed about its free allowance. Instagram and TikTok label rules were not re-checked for this guide.

Review date: 4 November 2026

Disclosure: Launch Video Maker is one of our products — this site is made by the same studio, Adapt Progress Evolve. We have tried to compare fairly, and the alternatives named are real and worth a look.