Step 1: Which kind of AI video do you need?
"AI video" covers four quite different outputs. Pick the one that matches the job before choosing a tool.
| Type | What you get | Tools and models | Good for |
|---|---|---|---|
| A presenter talking to camera | An AI avatar reads your script | HeyGen, Synthesia, Colossyan, Elai | Training, onboarding, updates, translated versions |
| A video from a script or article | Stock or generated scenes with an AI voice and captions | invideo AI, Fliki, Pictory, Lumen5, Visla | YouTube explainers, social posts, blog-to-video |
| Generated clips from prompts | Raw shots you edit together (no story, no voiceover) | Veo 3.1, Seedance 2.5, Kling 4.0, MiniMax H3, Runway, Pika, Luma | B-roll, ads, music videos, concept shots |
| Motion graphics | Animated type, UI and shapes, designed scene by scene | nod (from your product page), Hera, Jitter, MotionVid | Software launch and explainer films, product ads |
Prices and free plans for each tool are in best AI video generators by use case; model specs in AI video models compared. Explaining a product? See AI explainer video generators and AI motion graphics tools.
Step 2: How do you write the script?
Write the voiceover first, then the on-screen text and shots for each line. Spoken scripts run at about 150 words a minute, so the length sets your word budget:
| Video length | Words of voiceover (at 150 wpm) | Typical use |
|---|---|---|
| 15 seconds | about 38 | Ad, teaser, Story |
| 30 seconds | about 75 | Social ad, product teaser |
| 60 seconds | about 150 | Explainer, feature launch |
| 90 seconds | about 225 | Launch video, walkthrough |
- Generate a timed, scene-by-scene draft with the free launch video script generator (15, 30, 60 or 90 seconds).
- Or fill in a two-column video script template with word budgets per beat.
- Time a finished script with the script length calculator; the rates behind 150 wpm are in words per minute.
- For generated clips, turn each line into a shot: subject, action, camera, light, and whether it needs sound.
Step 3: How do you generate the video?
Avatar and text-to-video tools take the whole script and build the video; you pick a voice, avatar or template and regenerate scenes you don't like. Model clips are generated one shot at a time, so a 30-second video is several generations:
| Model | Max per clip | Clips for 30 s | Rate (tier) | 30 s, one take |
|---|---|---|---|---|
| Pika H3 Flex | 15 s | 2 | $0.033/s (H3 Flex 768p, Starter monthly) | $1.00 |
| Gemini Omni Flash | 15 s | 2 | $0.07/s (720p) | $2.10 |
| MiniMax H3 | 15 s | 2 | $0.08/s (MiniMax-H3, 768p) | $2.40 |
| Veo 3.1 | 8 s | 4 | $0.10/s (Veo 3.1 Fast, 720p, with audio) | $3.00 |
| Wan 3.0 | 30 s | 1 | $0.10/s (720p) | $3.00 |
| Runway Gen-4.5 | 10 s | 3 | $0.12/s (Gen-4.5) | $3.60 |
| Kling 4.0 | 30 s | 1 | $0.126/s (Runway API, 720p (to Nov 30)) | $3.77 |
| Seedance 2.5 | 30 s | 1 | $0.231/s (720p) | $6.93 |
Computed from each developer's published per-second price (Kling 4.0 via Runway's API; Pika from its Starter credits), checked November 10, 2026. Luma prices its API in credits without a dollar rate. Budget for several takes per shot.
- Keep one look across shots: reuse a reference image or the last frame of the previous clip. Seedance 2.5 takes up to 50 references, Kling 4.0 up to 15, MiniMax H3 up to 12 files.
- Draft cheap, finish high: Gemini Omni Flash drafts at 360p ($0.023/s) and Seedance 2.5 has a 480p Draft mode before the final render.
- Check the rules: Veo 3.1 takes English prompts; Gemini Omni Flash doesn't accept negative prompts; Seedance 2.5 won't take reference images with real human faces.
Step 4: How do you edit an AI video?
Put the clips on a timeline in script order, trim each to the line it covers, then add the voiceover, music and captions. Runway Gen-4.5 and Luma Ray3.2 generate no sound, so add it all here; models with native audio still need levelling against your voiceover. Make it readable with the sound off: burn in captions or on-screen text.
Editors that help: Descript (edit by transcript), Kapwing and VEED (browser, auto-subtitles), Runway's Aleph 2.0 (change what's in a shot). Compared in best AI video editing apps.
Step 5: How do you export for each platform?
Export one 16:9 master at 1920×1080 for YouTube, LinkedIn, X and your site, and a 9:16 cut at 1080×1920 for Reels, Shorts and Stories. Each platform's length, size and ratio limits are on social media video sizes; check a finished file with the video specs checker (nothing uploads), and plan file size with the video file size calculator.
- Generate in the ratio you'll publish: most models offer 16:9 and 9:16, and several add 1:1, 4:3 and 21:9 (aspect ratio guide).
- Some models deliver unusual files: Seedance 2.5's 1080p output is 10-bit H.265, so convert to an H.264 MP4 if your editor or platform won't take it.
- Generated clips carry provenance marks in some cases: Veo 3.1 supports Content Credentials and Gemini Omni Flash adds an invisible SynthID watermark.
