Can AI really make ASMR video that works? I put it to the test

There's a breakout ASMR AI video trend on TikTok and Instagram for ASMR AI videos, showing a knife cutting through glass fruit, which is engaging. But how easy is it to do? You hear AI is simple – write a prompt and click 'generate' – but is it that easy? And more importantly, is it fun?

With the hype surrounding AI tools, and my own experience of visiting the AI Upscale conference, I decided to test out the latest AI tools and try and make my own polished videos. I decided on using Freepick as this platform has access to every AI model, including Veo 3, an ideal gen AI model for realistic video.

I went in optimistically thinking it's as simple as writing a prompt and letting the AI take over, but there's a little more to it, and the more I attempted and failed to get the video I aimed for, the more frustrated I became. I hear it's a common reaction to using AI – you simply hand the control and keys to Veo 3 and let go.

My first ASMR video was meant to be a relaxing and tactile scene of a person melting small holes into a glass fruit, my take on the ASMR AI glass fruit trend. And who doesn't love holes? The output from Veo 3 looked good, but there were weird little discrepancies – the person is burning holes from an imaginary blowtorch. My prompt is below:

'A close-up shot: A skilled artisan carefully burns small holes into a vibrant glass fruit under soft backlighting. The camera gently zooms in to highlight the intricate details of the glass as the flames flicker and dance around the surface.'

Prompting isn't fun

The issue was me. I just didn't think through the prompt and how Veo 3 understands its instructions. You need to get under the belly of what each AI model needs to know to get a video that comes close to what's in your head.

I do a little research and decide to give Veo 3 as much info as I can muster, including style hashtags and camera descriptions. It's a test of my descriptive writing skills; you need to imagine the video in your head and describe every angle, colour, emotion, and other details. Is it fun? Not sure, and for many artists who like to put their imaginations onto paper or screen directly, I can see it being a turn-off.

My new prompt? It's below…

'A cozy nighttime study café during a gentle rainstorm, shot in cinematic 4K. The scene is warm and softly lit with ambient golden light. Outside, raindrops streak down large café windows, reflecting neon signs from across the street. Inside, a steaming cup of coffee sits beside an open notebook with handwritten notes and a slowly writing pen. Soft lo-fi music plays in the background, mixed with relaxing ambient ASMR sounds: the patter of rain, pages turning, faint barista sounds, the hiss of a milk frother, and distant chatter. Candlelight flickers gently, casting moving shadows. The camera slowly pans across the room, creating a calm, meditative atmosphere. No characters, no narration, just immersive ambient storytelling.'

I add in these hashtags: #ASMR #LoFi #RainyNight #CinematicCafe #StudyAmbience

The consistency problem

It's okay, the video is getting better. But ASMR needs to feel intimate and personal; this new video lacks real emotion. My video needs more context. And here's the issue with AI, it's often impressive in isolation, but to get any sense of emotional connection, you need to string together scenes, at which point consistency becomes AI video's biggest Achilles heel.

Characters generated in one scene don't quite match their appearance in the next. Backgrounds varied in lighting, style, and resolution. You can’t yet ask most AI tools to give you five shots of the same character in the same location and expect them to match. It’s like working with five different illustrators who never speak to each other.

I test this by first creating a face for my ASMR video, a woman sitting at a table in the same context as before – a rainy night with moody, cosy lighting. I reuse the same prompt as before, but add to it, trying to develop the scene. The previous street cafe vibe is now a home, which is nicer, but are they connected? Not really.

I love the opening shot, the realism of the character, the lighting, and Veo 3 has chosen to add in its dialogue and whispered audio, but the next shot disappoints, the fingers are dangling into an empty cup, tapping something that doesn't exist.

My prompt:

#ASMR #CinematicCafe #RainyNight #LoFi #WhisperedASMR #FirstPerson #FirstPerson #ComfortCore A soothing first-person ASMR experience inside a softly lit, cozy room at night. You are seated at a wooden desk with a warm desk lamp glowing softly. Across from you, a calm and friendly person (gender-neutral appearance, soft features, warm expression) makes gentle eye contact and begins a relaxing personal attention session. They whisper softly and slowly, asking how your day was and if you're ready to relax. They carefully brush your face with a soft makeup brush, adjust a cozy blanket around your shoulders, and tap and scratch on various objects — a ceramic mug, wooden box, and a leather journal. Ambient background sounds include a ticking clock, soft rain against a window, and distant thunder. The person’s movements are slow, intentional, and caring, with frequent moments of soft-spoken affirmations and close-up attention. The overall mood is calming, safe, and comforting

Making more scenes

I feel like I'm getting somewhere, but really, this experiment is showing one thing – getting what you want from AI models takes time and patience; it's trial and error. When I was at Upscale conference, I asked Jason Zada, half jokingly, what his team did all day. I now know – prompting, trying things, editing, deleting, rinse and repeat.

To finish my little experiment, I decided that what's missing is a sequence of scenes. This can be done using AI in a number of ways. You can generate multiple clips with a prompt like, 'Continue from the last scene: move from tapping a mug to scraping a tin bowl,' or you can generate short clips and ask Veo 3 to embed time codes to edit in the best video editing software like Premiere Pro or DaVinci Resolve. You can even ask Veo 3 to create a loopable 60-second ASMR sequence.

I choose to ask Veo 3 to create three scenes and give each a chapter and description, and retain the same prompt text from the previous two video attempts, to ideally get some consistency.

AI ASMR, was it worth the time?

My first foray into AI video revealed several things: I admire artists who are using this technology to create new, imaginative, and consistent results – artists like GMUNK and Henry Dubraez – because it's hard and takes skill; the idea AI is a time-saver is a misnomer; you can create something in seconds, but that something isn't always good, usable or original; and my dabble in AI showed the current limitations of what it can do.

This last point is important, as while gen AI can create images, video and audio you still need to know how to use 'traditional' software to edit the raw assets; indeed the best uses of AI are projects that begin with original art, photos and video and end with it, using AI to enhance, create accidents and develop ideas.

You still have to have ideas and make those judgment calls: timing, pacing, audio syncing, transitions, and more. That’s not something the current AI tools handle well. If anything, the more assets I generated, the more work I had to do wrangling them into a narrative or something close to what I had in my head.

Finally, there's a paradox inherent in AI – having so much choice of models and an endless supply of AI-created art and video sounds, freeing, but in practice, it can be overwhelming. Freepik is a great platform as it offers everything under one subscription, but unless you've done some research, you can be left asking, 'Why is one better than the other?'

I spent far more time tweaking prompts, regenerating scenes, and second-guessing styles than I ever expected. Creative control was still in my hands, but the abundance of possibilities and the trial and error of getting what's in my head into the model slowed me down.

Don’t get me wrong, AI can be incredibly helpful. It can also be fun as something to experiment with, and there's a gacha-like excitement as results are revealed. My biggest take? Right now, the myth is that you can click a button and produce a TikTok-ready video from scratch. That’s just not true, at least, not yet.

If you don't want to go near AI, read our guide to the best digital software and try Procreate, an AI-free app for the best iPads for drawing.