TL;DR
You can make a watchable 8-second AI clip with synced audio in about 15 minutes. The big change you need to know first: OpenAI shut down the Sora consumer app (sora.com and the in-ChatGPT tool) on April 26, 2026, so a beginner’s easiest paid path today is Google’s Veo, which you reach inside the Gemini app. Sora 2 still exists but only as a developer API that sunsets September 24, 2026. Below is the access reality, a four-part prompt formula, the clip-length limits, and how to choose between a realistic look and a stylized one.
Key tools and concepts:
- Veo — Google’s text-to-video model. Veo 3.1 is the current generation; in the Gemini app it now runs under the Gemini Omni video experience. This is the beginner-friendly path.
- Sora 2 — OpenAI’s text-to-video model. Consumer access ended April 26, 2026; only the paid API remains, and it sunsets September 24, 2026.
How access actually works in June 2026
Read this part before you sign up for anything — the landscape moved in early 2026 and most older tutorials are now wrong.
Veo (the beginner path)
Veo 3.1 is Google’s text-to-video model. The simplest way in is the video tool inside the Gemini app, which as of May 2026 is branded Gemini Omni (Gemini Omni Flash replaced the older “Veo 3.1” label in the consumer app — same family, new front end). You need a paid Google plan:
- Google AI Pro — $19.99/month. Includes 1,000 Google Flow credits monthly, roughly 50 Veo Fast clips or about 10 top-quality clips. This is the tier most beginners want.
- Google AI Ultra — $99.99/month. A much larger credit pool for heavy use.
Two other entry points exist for developers: Google AI Studio (free preview quota for testing) and the Gemini API / Vertex AI (pay per second). On the API, Veo 3.1 runs from about $0.03/second (Lite, no audio) up to $0.40/second (full quality with audio), as of June 2026. Beginners do not need the API — the Gemini app is enough.
What Veo does best:
- Realistic physics and natural-light scenes: street, documentary, indoor conversation, sun / wind / water.
- Grounded talking-head shots: a person speaking with lip-sync plus room ambience that reads as documentary-real.
- Photoreal humans and animals: skin, fur, and micro-expressions hold up better than Sora’s.
Sample prompt:
An elderly woman at a Paris cafe table
smiles at the camera and says "Bonjour",
natural light, ambient street audio, 35mm film grain,
eye-level fixed camera.Where Veo struggles: it is conservative on heavy stylization and surreal warping; each generation caps at roughly 8 seconds (the Gemini Omni front end advertises up to ~10s); celebrity and copyrighted-character filters are strict.
Sora 2 (now API-only)
Sora is OpenAI’s text-to-video model. As of June 2026 there is no consumer Sora app — sora.com and the Sora tool inside ChatGPT were discontinued on April 26, 2026, and ChatGPT Plus/Pro no longer include video generation. What remains is the Sora 2 API, which third-party tools can call until it sunsets September 24, 2026. Pricing is per second of output:
| Sora 2 API tier | Resolution | Price/sec | Clip lengths |
|---|---|---|---|
| Sora 2 (standard) | 720p | $0.10 | 4, 8, 12s |
| Sora 2 Pro | 720p | $0.30 | 10, 15, 25s |
| Sora 2 Pro | 1080p | $0.70 | 10, 15, 25s |
Batch mode is roughly half price with a 24-hour turnaround. For a true beginner with no coding setup, Sora is not the place to start — use a third-party app that wraps the API, or stick with Veo.
What Sora 2 still does best (via the API or a wrapper):
- Complex camera moves: dolly, tracking, aerial, low-flying drone, one-take long shots.
- Stylized looks: golden hour, 35mm grain, neon cyberpunk, low-saturation cinematic color.
- Abstract / surreal subjects: shattering glass, fluids, smoke, morphing forms, slow-motion physics.
Sample prompt:
A woman in a red trench coat slowly turns toward camera
on a neon-lit Tokyo street in the rain,
slow motion, 35mm film grain,
camera dollies in slowly from waist height,
neon reflections on wet pavement.Where Sora struggles: face identity drifts across multiple shots; hands, fingers, and on-screen text often distort; it skews stylized, so grounded photoreal scenes can look less convincing than Veo’s.
One-line picker
- New, want the simplest paid path → Veo in the Gemini app (Google AI Pro, $19.99/mo).
- Want grounded, photoreal realism → Veo.
- Want cinematic / complex camera / stylized or surreal, and you can use an API-backed tool → Sora 2 (until Sept 24, 2026).
- Both generate native synced audio (dialogue, sound effects, music) in one pass, so decide on look, clip length, and price — not on which one “has sound.”
Who this is for
You have never made an AI video and want a first result today without picking the wrong tool or burning a subscription on the wrong thing.
Make your first clip: step by step
- Pick scope. Aim for a single 4-8 second beat on your first try. Veo generations are ~8 seconds; one shot, one idea.
- Write the four-part prompt: subject + action verb + camera move + lighting. The action verb is what turns a still image into a video. You can add a
dialogue:orambient:line and the model will generate synced audio with the picture. - Generate, then change one variable at a time. Swap the subject, or the camera move, or the lighting, or the style — never two at once, or you cannot tell which change moved the result.
- Stitch in an editor (CapCut, Premiere, or DaVinci Resolve). Veo and Sora clips ship with their own audio track, so do not mute it by accident when you cut.
A realistic first session: 3-4 generations to learn the prompt grammar, then 2-3 keepers you stitch into a 20-30 second sequence.
Common mistakes
- Going long on the first try. Waiting on a 25-second render that turns out wrong wastes far more credits than three short 4-second tests.
- No verb in the prompt. “A sunset by the sea” is a still frame. Add “camera dollies slowly along the coastline” and it becomes video.
- Choosing by audio. Both models generate synced dialogue, effects, and music in one pass now. Choose by look and budget.
- Asking Veo for surreal warping. “A glass person melting into neon” will look stiff in Veo. That is a Sora job.
- Comparing tools by changing the prompt and the model at once. Fix the prompt, then swap the model, or you will not know which one moved the result.
- Following a 2025 tutorial that says “open Sora in ChatGPT.” That path no longer exists — the Sora consumer app closed April 26, 2026.
FAQ
Q: As a beginner with no coding skills, which tool should I actually use? A: Veo, through the Gemini app on a Google AI Pro plan ($19.99/month as of June 2026). It is the simplest paid consumer path, it is strong at realistic scenes, and it generates synced audio. Sora 2 is API-only now, so you would need a third-party app that wraps it.
Q: Can I still use Sora inside ChatGPT? A: No. OpenAI discontinued the Sora consumer app — sora.com and the in-ChatGPT tool — on April 26, 2026, and ChatGPT Plus/Pro no longer include video generation. The Sora 2 API still works for developers and third-party apps until it sunsets on September 24, 2026.
Q: How long can a single clip be? A: Veo generations are about 8 seconds (the Gemini Omni front end lists up to ~10s). The Sora 2 API supports 4/8/12s on standard and 10/15/25s on Pro. Either way, plan your story in 8-second beats and stitch them in an editor.
Q: Why does my AI video look stiff or jittery? A: Usually one of two causes: an over-choreographed prompt (“walks left, then turns, then sits down”) the model cannot stage in 8 seconds, or a clip stretched past its natural length. Simplify the action and keep it short.
Q: How do I get a consistent character across multiple shots? A: Both models drift on face identity across separate generations. Use image-to-video with a reference image to lock the face, generate on-screen text in your editor rather than asking the model to render it, and keep each character description identical word-for-word between prompts.