EP 051 · 66 MIN GO OBVIOUSLY AI, OR GO HOME

Sora 2, Veo 3 & The AI Video Gold Rush

October 6, 202566 min Rory Flynn & Drew Brucker

The short version

Episode 51 marks Fast Hours' return after a month off with a deep look at Sora 2's remix, cameo and 'any IP' features, which Rory argues make it the easiest AI video tool yet for non-experts. Rory and Drew also compare Sora 2, Veo 3.1 and Luma's Ray3 for speed and consistency, walk through a real keyframe-troubleshooting workflow for a stunt shot, and cover Visual Electric's acquisition by Perplexity, Higgsfield's pivot to an aggregator, and Midjourney's refined Style Explorer search.

Key takeaways

  1. 01

    Sora 2's Remix button lets anyone rework someone else's post, like Midjourney's Explore page for video, and Rory says its comedic timing is unmatched by any other video model.

  2. 02

    Cameo permissions can be public (Sam Altman's is): anyone can drop a public figure's likeness into a Sora 2 scene once that person opts in.

  3. 03

    For brand use, Rory argues for going 'obviously AI' and embracing the format's flaws rather than chasing realism, which risks looking deceptive and inviting backlash.

  4. 04

    AI video still lacks context between shots: a car (or any object) generated fresh from a new start frame will show a different paint job or details than the last shot unless you control it with keyframes.

  5. 05

    Luma's Ray3 currently handles fast action and fight choreography better than Veo 3.1, which Rory says has mysteriously slowed down and lost its 'fast motion' look.

  6. 06

    Luma abruptly cut most paid creators from its creator program over follower counts with no warning, then reinstated a few; Drew and Rory call it a preventable communication failure.

  7. 07

    Visual Electric was acquired by Perplexity; Higgsfield pivoted from a lagging video model to becoming an aggregator with aggressive guerrilla marketing, including a paid WorldStar post.

Receipts · said on the record

“if you want to talk about like AI adoption, I have a feeling this will be like the best tool for AI adoption since probably Midjourney.”
Rory Flynn Why Sora 2 could drive mainstream AI adoption Play from 4:16
“Personally if I was if I was the head of a big giant billion dollar brand if I was gonna use AI I would do it in the most ridiculous fashion. So everyone knows that it's AI, but it's also good.”
Rory Flynn How brands should approach visible AI use Play from 16:42
“They just like kick people out because they were just like, you don't have a big enough following, which look. I get it from the brand perspective, they went about it the wrong way.”
Rory Flynn Luma abruptly cutting its creator program Play from 20:49
“There's no context of what the side of that car looks like. So if you're just generating as a start frame, that car lands and you want it to turn, it's going to be a totally different side of the car.”
Rory Flynn Why AI video shots lose consistency without keyframes Play from 26:00
“Why do we have two names for the same thing? And then you got Flow, and then what is Flow and Whisk? We got VO3, all these things, image gen … but we got Nano Banana. What are we doing?”
Drew Brucker Frustration with Google's fragmented product naming Play from 52:09
“unpopular opinion, but I kind of want it to be like Sora right now. I would want it to be like Sora because for Midjourney, a lot of things I just want to do it for fun.”
Rory Flynn What he wants from Midjourney's future video model Play from 54:23

Chapters

  1. 0:00:00 Intro + return from creative hibernation
  2. 0:00:42 Sora 2 drops: “the floodgates just opened”
  3. 0:01:27 Sora 2’s new tricks: editing, sound, dialogue, and IP
  4. 0:02:17 Veo 3 vs Sora 2 vs Ray 3
  5. 0:03:20 Rory’s South Park AI episode + the “bar-only” friend
  6. 0:05:00 Why dedicated AI-only spaces actually matter
  7. 0:05:48 Remix culture, Mr. Rogers AI, and the rise of comedic timing
  8. 0:07:00 The “Remix” button, mass adoption, and kids using Sora
  9. 0:08:43 The IP gray area: Minecraft meets Mr. Rogers
  10. 0:09:43 Cameos, access codes, and mobile vs desktop creation
  11. 0:12:25 Sequential storytelling: AI understands chronology now
  12. 0:13:29 Toyota ads, guerrilla launches, and OpenAI’s flood strategy
  13. 0:15:00 IP risk vs reward — how far can brands push it
  14. 0:18:00 AI performance comparisons
  15. 0:23:00 Fight scenes, motion control, and why keyframes matter
  16. 0:27:50 Workflow troubleshooting and micro-decision fatigue
  17. 0:34:50 Too many tools? Runway, Aleph, and the Weavy advantage
  18. 0:35:45 Shout-out to Weavy + tool consolidation predictions
  19. 0:36:00 Higgsfield pivots, Pika memes, and the marketing gap
  20. 0:37:00 Visual Electric’s acquisition and the coming consolidation wave
  21. 0:38:20 Midjourney updates: Style Explorer, Smart Search, and new unlocks
  22. 0:40:30 Playing with EXP mode + hidden color/style refinements
  23. 0:42:30 Style Finder, Style Creator, and mood-board personalization
  24. 0:43:57 Style ranking feature + --r 40 nostalgia meltdown
  25. 0:46:00 Midjourney V8 speculation & dataset rumors
  26. 0:50:30 Google’s product chaos: Gemini, Nano Banana, Flow, and Veo 3
  27. 0:53:00 Why Google can fail (and still win)
  28. 0:55:10 ChatGPT’s image text features & the next AI video wave
  29. 0:59:30 The Weavy renaissance and workflow automation discussion
  30. 1:02:00 New creative problems worth solving
  31. 1:06:00 Why “easy” AI creation still stings for creatives

Questions this episode answers

What makes Sora 2 easy to adopt, according to Fast Hours?

Rory says Sora 2 needs almost no prompting skill or background to get a good result, calling it potentially the biggest AI-adoption driver since Midjourney, thanks to features like Remix and cameo videos.

Should brands use AI video that looks obviously fake?

Rory argues yes: he'd rather a brand 'do it in the most ridiculous fashion so everyone knows it's AI, but it's also good,' since chasing photorealism risks looking deceptive and drawing backlash.

What happened with Luma's creator program?

Luma abruptly removed most paid creators from its program over follower-count thresholds with little warning, then reinstated a few; Drew and Rory call the rollout poorly communicated even if the underlying business logic made sense.

Is Veo 3.1 or Luma's Ray3 better for fast action?

Rory says Ray3 currently handles fast motion and fight choreography better, while Veo 3.1 has mysteriously slowed down and struggles to render fast-moving cars or people without looking like slow motion.

Does Fast Hours want Midjourney's future video model to look like Sora?

Rory says, as an 'unpopular opinion,' he'd rather Midjourney's video model be fun and meme-able like Sora 2 rather than purely cinematic, arguing fun is what actually drives adoption.

Show notes

Fast Hours Podcast — Episode 51: Sora 2, Veo 3 & The AI Video Gold Rush

Drew and Rory crawl out of their creative caves after a month-long “break” (read: total burnout) and immediately get smacked in the face by Sora 2, Veo 3, and Midjourney Style Explorer updates that make everything they said last episode obsolete.

They break down what’s actually working in Sora 2 (full edits, audio, dialogue, even IP?), why Veo 3 can’t seem to go fast anymore, and how keyframes are quietly becoming the holy grail of AI video.

Expect rants about IP chaos, Mr. Rogers cameos, and why OpenAI’s guerilla launch strategy might be the smartest marketing move of 2025. It’s equal parts therapy session, tech panic, and creative caffeine rush — exactly how they like it.

Keywords: Sora 2 update, Google Veo 3 review, Midjourney V8, AI video generation tools, Luma AI, keyframes, AI video editing, AI marketing trends 2025, generative video, Style Explorer Midjourney

Every word Read the full transcript

If you liked this one