He Spots Every AI Fake — But Filmmakers Already Have the Edge, says Jeremy Carrasco – Focus Check ep105

by Felix TappeinerFebruary 26^th, 2026

Subscribe to Podcast

In this special episode of Focus Check, host Nino Leitner sits down with Jeremy Carrasco, the AI video educator and media literacy creator nearly a million people trust to tell them what’s real on the internet.

AI video has gone from novelty to noise — and Jeremy Carrasco has been watching every step of it. In episode 105 of the CineD Focus Check podcast, the live streaming veteran turned AI video educator walks us through what the latest generation of diffusion models can and can’t do, why every model still has a detectable fingerprint, and why the bigger threat isn’t fake videos fooling people — it’s real videos being dismissed as fake. From the “AI accent” in synthetic speech to the slot-machine frustration of using these tools professionally, Carrasco pulls no punches.

Apple Podcasts Spotify Amazon Music PlayerFM iHeart Radio YouTube

With models like Seedance 2.0 and Kling 3 going viral for near-convincing results, the conversation also gets philosophical: is undetectable AI video even desirable, and what happens to filmmaking when spectacle becomes free? Carrasco’s answer might surprise you — he sees a coming backlash against AI-saturated feeds and a real opportunity for human-centred storytelling to regain its edge. The advice for filmmakers? Stop trying to out-prompt AI, and double down on the thing machines can’t replace: relationships.

Chapters and articles in this episode

00:00 Introduction & Jeremy’s Background

05:30 How Jeremy Got Started Spotting AI Video

08:19 Veo 3 — The Inflection Point for AI Video

15:32 Every AI Model Has a Fingerprint

21:47 The Real Problem: Doubting Real Footage

27:30 Computational Photography & Smartphone Images

33:00 AI Video vs. Professional Camera Workflows

38:30 The Creator Economy Under Pressure

44:00 C2PA & Content Provenance — Does It Actually Help?

51:00 VFX, Extras & Where Jobs Will Really Go

57:00 The “AI Accent” & Why Audio Is Still a Giveaway

01:03:00 Is There a Legitimate Creative Space for AI?

01:09:00 AI, Social Media & The Coming Backlash

01:16:00 How Filmmakers Can Stay Irreplaceable

01:20:24 Looking 3-4 Years Ahead

01:26:18 Where to Find Jeremy & Wrap-Up

How Jeremy Carrasco became the “AI video detective”

Carrasco’s path to becoming the internet’s go-to AI video detective wasn’t planned. A live streaming technical producer and encoding engineer by trade, he left his production company last year just weeks before Google’s VO3 model dropped — a moment he calls a major inflection point for public awareness of AI video. He bought multiple subscriptions, burned through $250 in credits in two weeks, and came out the other side knowing every flaw these models produce. His background in broadcast production and encoding gave him a unique edge: he already knew everything that could go wrong with real video, so spotting the synthetic stuff became a process of elimination.

Every video has a fingerprint – that can’t be fully erased

The conversation goes deep on what Carrasco calls the “fingerprint” theory — the idea that every media-making process, whether it’s an ARRI Alexa, an iPhone, or a diffusion model, leaves traces that can never be fully erased. Just as filmmakers can distinguish between camera systems by their rendering characteristics, AI video models each carry their own telltale signatures in how they handle lip sync, camera support, noise, and dynamic range. Carrasco argues this will always be the case, even as the technology improves, because the mathematical processes behind diffusion models are no different in principle from the physical processes behind a lens and sensor.

People doubt genuine footage more than ever

But the most surprising takeaway may be about trust, not technology. Carrasco reveals that the videos his followers send him asking “is this AI?” are now 75-80% real. People are doubting genuine footage more than ever — and he considers this erosion of trust a bigger problem than the fakes themselves. As he puts it, the average person learns AI can generate fake videos and suddenly questions everything, without the tools to distinguish between the two.

Prompt engineering: a worthless skill

The episode also covers practical ground for working filmmakers. On prompt engineering as a skill investment, Carrasco is blunt: he considers it overblown and expects the prompting process to become so simplified that it offers no competitive advantage. On AI’s threat to professional production, he draws a sharp line between transactional work like stock footage — which he considers effectively dead — and creative, relationship-driven filmmaking, which he believes becomes more valuable as AI-generated content floods every platform.

C2PA, state of AI audio and how not using AI might be beneficial for Hollywood

Carrasco and Nino also dig into the C2PA metadata standard, the state of AI audio (still far behind video, even according to OpenAI’s own Sora team lead), why Hollywood may actually benefit from NOT using AI once everyone else has access to the same tools, and how social media — not cinema — is the sector most immediately disrupted by generative video.

The full conversation runs over 90 minutes and is one of the most philosophical episodes of Focus Check to date — an essential listen for anyone trying to understand where AI video is headed and what it means for the craft of filmmaking.

Watch or listen to the full episode on YouTube, Apple Podcasts, or Spotify. Follow Jeremy Carrasco at @ShowToolsAI on Instagram and TikTok, and check out his publication at riddance.ai. You have feedback, comments, or suggestions? Write us at [email protected]