Asset 20 8 2
Does AI recommend your business? Run the free check →

Join 15,000 business owners, marketers and entrepreneurs. The Sunday newsletter you'll be annoyed only arrives once a week.

Article

Sora Prompts for Realistic People That Hold Up

If you are skim reading
The short version: Realistic people in Sora come from specific, boring-sounding prompts (exact age, specific lighting, named camera type, imperfections called out) rather than words like "photorealistic" or "4K", and even then you'll still spot the AI tell in

The short version: Realistic people in Sora come from specific, boring-sounding prompts (exact age, specific lighting, named camera type, imperfections called out) rather than words like "photorealistic" or "4K", and even then you'll still spot the AI tell in the hands, the teeth, or the way fabric moves. Write for a documentary photographer, not a magazine cover, and you'll get closer to real.

Why most "realistic person" prompts fail before you even hit generate

I've watched a lot of people type "photorealistic portrait of a woman, 8K, hyper detailed" into Sora and then wonder why she looks like a waxwork with good lighting. The word "photorealistic" is doing nothing. It's a leftover from Midjourney prompting culture and it doesn't map onto how Sora's model reasons about a scene.

Sora isn't rendering a photo. It's simulating a short piece of footage, which means it's thinking about light sources, motion, skin under movement, and camera behaviour all at once. The prompts that work are the ones that give it a real-world context to simulate, not a style adjective to slap on top.

So the first shift: stop describing the finish you want and start describing the shoot. Who's the photographer. What lens. What time of day. What's slightly wrong with the person, because nobody is symmetrical and nobody has perfect skin at 2pm on a Tuesday.

The anatomy of a prompt that produces a believable face

A strong Sora prompt for a realistic person has five layers, and I'd argue most people only ever write two of them:

  • Specific demographic detail (not "a woman" but "a woman in her late 40s with fine lines around her eyes and slightly uneven eyebrows")
  • Named camera and lens behaviour (phone front camera, 35mm film, handheld DSLR with slight motion blur)
  • Light source and time (overcast window light at 4pm, not "cinematic lighting")
  • Context and action (she's mid-sentence, she's reaching for a mug, she's not looking at camera)
  • Deliberate imperfection (flyaway hair, a slightly crooked collar, a half-finished blink)

Here's a worked example. Say you run a small bakery account and you want a short clip of a "customer" reacting to a cake for a promo, without hiring anyone. A weak prompt: "photorealistic woman tasting cake, happy expression, 4K." A far stronger one:

"A woman in her early 30s with slightly tired eyes and a loose bun, sitting at a wooden café table near a window on an overcast afternoon. She takes a bite of chocolate cake, pauses, and breaks into a genuine, slightly lopsided smile, eyes crinkling. Shot on a handheld phone camera, natural light only, slight grain, background softly out of focus with a blurred coffee cup visible."

That second version gives Sora a believable physical world to simulate rather than a vague mood to guess at. The difference in output quality, in my experience testing prompts like this, is not subtle.

The uncomfortable bit nobody wants to say out loud

Here's what most guides won't tell you: even with a perfect prompt, Sora still struggles with hands holding objects, teeth during speech, and the physics of hair in wind. You can prompt around some of it, but you cannot fully fix it with wording alone, and anyone claiming otherwise is selling you something. If your use case needs a close-up of someone signing a document or playing piano, expect to regenerate multiple times and still cut around the worst frames in editing.

That's not a reason to avoid Sora for people content. It's a reason to plan your shot list around its weaknesses the same way a real director plans around a nervous actor. Wide shots, side profiles, and movement that distracts the eye from hands are your friends. Static close-ups of a face talking straight to camera are where the illusion breaks fastest, which is exactly why AI generated videos still look a bit weird even when everything else about the scene is convincing.

Prompt structures that work for different "realistic person" scenarios

Everyday candid shot

Structure: [age + specific physical detail] + [activity, mid-motion, not posed] + [location detail with one imperfection] + [camera type] + [light source and time of day]

Example: "A man in his late 50s with greying stubble and reading glasses pushed up on his forehead, scrolling his phone while waiting at a bus stop. Slightly cracked pavement, overcast light, shot on a 50mm lens with shallow depth of field, background blurred, no direct eye contact with camera."

Talking-to-camera style (for social clips)

Structure: [age + role, not generic descriptor] + [what they're saying, described not written] + [setting detail] + [camera wobble or imperfection] + [lighting]

Example: "A woman in her mid-30s, a small business owner in a slightly cluttered home office, explaining something enthusiastically to camera with natural hand gestures, occasional glance away as she thinks. Webcam-style footage, slightly soft focus, warm desk lamp light mixed with daylight from a side window."

Work with me

Want AI doing the heavy lifting in your marketing?

I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.

This structure is close to what you'd use for product explainer content too. If you're building out a library of these for a brand, it's worth reading through how I'd approach Sora prompts for product videos that convert, because the same "give it a real shoot, not a style" principle applies.

Group or crowd scenes

This is where Sora gets shakiest. Faces further from camera tend to blur or distort more than foreground ones. If you're prompting a group, put your one "important" character in sharp focus and deliberately describe the rest as soft, backlit, or partially obscured. You're not hiding a flaw, you're matching how real cameras behave in a crowd, and it buys you a lot of forgiveness from the viewer's eye.

Settings and workflow that matter as much as the words

Prompting is maybe 60 percent of the result. The rest comes from how you handle generation and selection:

  • Generate at least 3 to 4 variations per prompt. Realistic-person results are inconsistent run to run, and the best one is often the third or fourth attempt, not the first.
  • Use shorter clips (4 to 8 seconds) for anything with a face talking. Longer clips give more time for the model to drift into uncanny territory.
  • Avoid prompting multiple named emotions in sequence ("starts sad then becomes happy") in one shot. Split it into two generations and cut between them, it looks far more natural.
  • If skin looks too smooth, add a specific flaw: "a small scar above the eyebrow," "visible pores," "slightly chapped lips." Vague terms like "realistic skin texture" rarely do much on their own.

When Sora isn't the right tool for the realism you need

I'll say plainly what a lot of content on this topic avoids: if your project needs a specific recognisable person doing something precise, close up, for more than a few seconds, you may be better served by filming a real person on a phone than fighting Sora for twenty generations. Sora is brilliant for background characters, atmosphere, b-roll of "a person," and stylised realism. It is not yet a replacement for a real actor in a close, high-stakes shot. Knowing which jobs to hand to which tool is half the skill, and it's covered in the broader comparison of best AI video generators for social media if you want to see where Sora sits against the alternatives.

It's also worth being realistic about cost and time before you commit a campaign to this approach. Regenerating a realistic-person clip ten or fifteen times to get one usable take isn't unusual, and that adds up in credits and hours. If you're scoping a project, it's worth reading through how to budget for AI video creation before you promise a client or your own team a delivery date.

A quick prompt checklist before you generate

  • Have you given an exact age range, not "young" or "old"?
  • Have you named a camera or lens style instead of saying "photorealistic"?
  • Have you described light by source and time, not mood words like "cinematic"?
  • Have you included one deliberate imperfection in appearance or framing?
  • Is the person mid-action rather than posed and staring at camera?

If you can tick all five, you're already ahead of most people typing prompts into Sora. And if you're building this into a wider content operation rather than a one-off clip, it's worth stepping back and looking at where generative AI video tools fit into your business process, because the prompt is only ever the smallest part of making this work at scale.

For anyone pairing this with text or image prompting for the same campaign, there's a lot of crossover in how specific you need to be, which I've gone into in the wider ChatGPT prompts guide covering fun prompts, photo edits, and daily caps.

Frequently asked questions

Why do Sora-generated people still look slightly off even with a detailed prompt?

Hands, teeth, and fine hair movement are still the weak points of current video generation models, and no amount of wording fully fixes that. Plan your shots to avoid close-ups of these details rather than trying to prompt your way around a known limitation.

What's the single most useful change to make to a Sora people prompt?

Replace vague style words like "photorealistic" or "4K" with a specific camera type, light source, and one physical imperfection. That shift alone usually improves believability more than any other single edit.

How many times should I generate a prompt before giving up on it?

Run at least 3 to 4 variations before judging a prompt as weak. Realistic-person outputs vary a lot between attempts on identical wording, and the usable result is often not the first one you see.

Is Sora good enough to replace a real actor in marketing videos?

For background characters, atmosphere, and short b-roll it's useful. For a recognisable person speaking in close-up for more than a few seconds, filming a real person is still usually faster and more convincing.

Run a business, a brand or a blog? I feature tools, products and expert guest posts for an audience of 15,000. See how it works

Published and maintained by the Lilach Bullock team, covering marketing, AI and business growth.
Your buyers are asking AI who to use. Does it say you?

See for free whether ChatGPT, Claude, Perplexity, Gemini and Google name you, and get the plan to become the answer.

Check my AI visibility →
Sundays only

Get the Sunday newsletter.

One email a week. AI experiments, marketing tactics, and the workflows Lilach is building right now in her own business.

Subscribe free

Let’s get your marketing running on AI.

Book a free 30-minute call

We figure out what you need, where AI fits in, and what working together would look like.

Book the call →

Or take the 30-second calculator

You’ll see the hours and the money quietly leaking out of your week, and the three workflows worth building first.

Take the calculator →

Or grab the free AI resource library

Prompt packs, templates, checklists, and swipe files. The exact tools I build for paying clients. Yours, free.

Get the library →
Keep reading

More from the blog.