How to make AI videos: choosing a tool for beginners

Text-to-video vs image-to-video

With text-to-video you type something like "a puppy sitting at a rainy window," and the AI draws the whole clip from scratch. You get a lot of freedom, but the AI decides the fur color and the shape of the window, so the result often differs from what you pictured.

With image-to-video, your photo becomes the first frame and the AI adds motion. If you need it to be your cat, this is the better choice. Some tools also accept several reference images to keep a person or object consistent. For example, Google's help page says you can upload one video and up to five images in the Gemini app.

Both use credits or usage allowance every time you generate. Starting from a photo cuts down on trial and error.

Tools compared (as of November 2026)

The table lists only what I confirmed on official pages. Plans and limits change often, so check each service's pricing page before you pay.

In short: if you already pay for a Google plan, the Gemini app is the simplest start. If you want longer clips, higher resolution or several models in one place, an aggregator like Higgsfield fits better. The table is a list of verified facts, not a ranking. Quality depends heavily on your prompt and subject, so you have to try it yourself.

• Gemini app (Google)

  What I confirmed

  Video is made by Gemini Omni (Gemini Omni 1.1 Flash on Google's overview page), which replaces the earlier Veo 3.1. It can generate audio with the video, makes 10-second videos, accepts one video and up to 5 images, and lets you pick the aspect ratio before generating.

  Free?

  Needs a Google AI plan on a personal account. Not available to users under 18.

• YouTube Create app / Shorts remix

  What I confirmed

  Gemini Omni clips up to 10 seconds, 900-character prompts, up to 3 reference images from your gallery. Listed countries include the United States, Canada, Australia, India and South Korea.

  Free?

  YouTube's page does not say whether it costs anything. It calls the features experimental and available to a limited number of creators.

• Higgsfield

  What I confirmed

  A service that offers several video models in one place. Per its help center, Seedance 2.0 makes 4 to 15 second clips at 480p to 4K. Seedance 2.5 makes 4 to 30 second clips and is available on Pro and Plus plans and above, not Starter or Basic. Native audio is included.

  Free?

  Credit cost depends on model, resolution and length, and is shown on the Generate button. I could not confirm current plan prices on an official page, so check the pricing page.

• Kling (Kuaishou)

  What I confirmed

  The official site shows Kling 4.0 as the latest model, with a focus on audio-visual performance and stable motion.

  Free?

  The page I read mentions no free credits or prices. Check your account after signing up.

• Sora (OpenAI)

  What I confirmed

  On March 24, 2026, an OpenAI spokesperson told CBS News: "We've decided to discontinue Sora in the consumer app and API."

  Free?

  Not recommended as a tool to start with.

What credits are

Credits are the allowance an AI service spends each time you generate a video. Video takes far more computing than text or images, so each clip costs more. Longer and higher-resolution clips cost more still.

A good habit is to test the composition with a short, low-resolution clip, and re-render only the ones you like at a longer length or higher resolution. Some services charge credits even for failed generations, so read the refund policy and expiry dates before you buy.

On subscriptions, credits are usually refilled each month, but whether unused credits roll over differs by service. Some, like the Gemini app, manage use with daily or monthly limits instead.

A prompt structure that works

A video prompt works better in a fixed order than as a long, flowery paragraph. Write one line for each of the following and the AI is less likely to get confused.

• 1. Subject and action: who does what. Use only one action.
• 2. Place and light: where it is and what the light is like, such as "a living room with morning sun."
• 3. Camera: fixed, slow push-in, top-down. If you skip this, the AI moves the camera however it likes.
• 4. Mood and sound: cozy, quiet, a soft clock ticking. If the tool generates audio, describe it.

A gray cat slowly tilts its head at a window. A bright living room with morning sunlight. The camera stays fixed at the cat's eye level. Cozy, quiet mood, with the soft sound of a small clock ticking. Vertical 9:16, 8 seconds.

Common problems and fixes

The "falls apart after about 6 seconds" row is a trend many users report, not an official figure. It varies by tool and version, so test it yourself.

• Extra or blurry fingers

  Why it happens

  Small, complex details are still hard for AI

  What to do

  Avoid close-ups of hands. Show the face or the whole body instead.

• Garbled signs or captions

  Why it happens

  The AI draws letters as pictures

  What to do

  Leave text out and add captions in your editing app.

• Face changes from your photo

  Why it happens

  The photo is dark or small, or the motion is large

  What to do

  Use a bright, front-facing photo, small movements and a "keep the face" line.

• Shapes break down in longer clips

  Why it happens

  The AI loses track of earlier frames

  What to do

  Make clips of 4 to 8 seconds and join them. Many people see problems after about 6 seconds.

• Awkward movement

  Why it happens

  You asked for several actions at once

  What to do

  Describe one action and say "slowly."

• Comes out wide instead of vertical

  Why it happens

  Some tools default to landscape for text prompts

  What to do

  Pick 9:16 before you generate.

After you make it

Once you have a few clips, join them in an editing app, add captions and music, and post them as Shorts or Reels. YouTube, Instagram and TikTok all ask you to label realistic-looking AI video. The details are in the guide on AI video labels and monetization on this site.

Do not make scenes where animals are hurt or in danger, or videos of real people saying things they never said. These can break service policies and get your video removed. For more projects, see the make section.

Related terms

Text-to-video

Making a video from a written description. Flexible, but hard to predict.

Image-to-video

Using a photo as the first frame and adding motion.

Credits

The allowance spent each time you generate a video. Longer and higher-resolution clips cost more.

Reference image

A photo you upload so the AI keeps a person or object consistent. The number allowed differs by tool.

SynthID

An invisible watermark Google adds to videos made in the Gemini app.

FAQ

Keep going

• Make with AI: step-by-step projects