EP 069 · 71 MIN OMNI CAN'T SWAP A CHARACTER. SEEDANCE CAN

Google Dropped Too Many AI Tools. Which Ones Matter?

May 24, 202671 min Rory Flynn & Drew Brucker

The short version

Episode 69 is Rory and Drew's reaction to Google's single-week flood of AI launches: Gemini 3.5 Flash, Google Flow, Omni, Pics, and refreshed Veo and Nano Banana. Rory tests Omni and Runway Aleph 2.0's character-swap tools against a complex Seedance clip and finds both fail, tries Flow's new agent mode and in-app tool builder, and the two debate what 'agentic' should actually mean.

Key takeaways

  1. 01

    Google's Omni and Runway Aleph 2.0 both failed to swap characters in a complex Seedance shot — camera motion tracked, but bodies morphed and warped.

  2. 02

    Seedance and Kling still lead on physics and camera motion; Rory calls Veo 3.1 forgettable and says he'd mostly stopped using it after Veo 3.

  3. 03

    Higgsfield's new agentic clipper produced the same cut twice even after a detailed system prompt — auto-clipping still can't out-edit a human.

  4. 04

    Flow's agent mode generated ten art-directed variations of one Midjourney image in a single request, batching directions instead of one at a time.

  5. 05

    GPT Image 2 still can't handle relative scale — an 8-inch toy sized next to a dog — even with cropped reference images.

  6. 06

    Rory calls Claude the most genuinely agentic tool he uses, because you can watch it test and correct its own approach mid-task.

  7. 07

    SynthID-style watermarks can be stripped just by opening an AI image in Photoshop, which is a real problem for client work that must disclose AI use.

Receipts · said on the record

“The first thing I did was put a complex video in there from Seedance that I made and was like, let's try to replace the characters. No chance. Not even close.”
Rory Flynn Testing Omni/Runway character-swap against a complex Seedance video Play from 22:13
“I mean, these are just numbers to me. I don't know what goes into them. I don't know how they're tested.”
Drew Brucker Skeptical of Gemini 3.5 Flash's published benchmark scores Play from 13:05
“It's also mapping how I think. That's what I mean. It already knows.”
Rory Flynn On Gemini answering a question using context it was never given Play from 18:57
“When I think agentic, I'm thinking automated, I'm thinking automatic, I'm thinking like, I've got somebody … that's doing something that's gonna give me high quality, predictable, consistent results.”
Drew Brucker Defining what 'agentic' should actually mean in AI tools Play from 55:06
“The agentic stuff on Claude, like, that makes sense to me. Cause it's like, I'll ask you to do a task and you can see all the work.”
Rory Flynn Why Claude's visible reasoning feels more agentic than workflow tools Play from 55:56
“We have sh*t days, sh*t weeks … I'm done today. And I don't know. You can't say the word stump. I've generated 10 times and probably wasted like $15 or more on this.”
Rory Flynn On a frustrating week of tools failing simple requests

Chapters

  1. 0:00:00 Cold open
  2. 0:00:32 Google’s AI naming avalanche
  3. 0:01:39 AI hype vs actual workflow value
  4. 0:02:34 Why AI launches feel like iPhone upgrades
  5. 0:06:12 Google’s “throw everything” strategy
  6. 0:07:08 Omni vs Veo 4 expectations
  7. 0:07:43 Video physics and speed problems
  8. 0:09:03 Google Pics, Flow, Omni, and Flash
  9. 0:10:04 How Rory actually uses Gemini
  10. 0:11:51 Gemini 3.5 Flash breakdown
  11. 0:12:38 AI benchmarks feel like marketing
  12. 0:13:42 Gemini as a better search layer
  13. 0:15:18 Creepy Gemini context awareness
  14. 0:17:35 Why AI data connections feel too early
  15. 0:19:15 The privacy tradeoff gets darker
  16. 0:21:19 Google Omni vs Runway Aleph 2.0
  17. 0:22:12 Google Omni testing starts rough
  18. 0:23:39 Google Veo 3.1 feels forgettable
  19. 0:25:21 Why Omni feels early
  20. 0:26:19 Higgsfield clipper test fails
  21. 0:27:59 Why auto-clipping still misses
  22. 0:31:30 Rory tests Flow and Omni live
  23. 0:32:41 Omni character swap struggles
  24. 0:33:33 Runway Aleph panda test
  25. 0:34:07 Flow’s new interface and tools
  26. 0:35:02 Building custom tools inside Flow
  27. 0:36:10 The joy of making tools from nothing
  28. 0:37:39 Agent mode for still-image workflows
  29. 0:39:05 Batch creative directions in Flow
  30. 0:40:03 Omni turns six images into video
  31. 0:40:47 Driving physics still feel off
  32. 0:41:55 Why consistency matters for adoption
  33. 0:43:03 Kling, Seedance, and the update race
  34. 0:43:59 Seedance handles complex camera motion
  35. 0:45:42 GPT Image setup for golf video
  36. 0:46:53 Testing the same prompt in Flow
  37. 0:49:25 Why agentic platforms can feel thin
  38. 0:51:10 The need for visual design systems
  39. 0:52:21 Flow’s golf swing result
  40. 0:53:56 Everyone is racing toward agentic
  41. 0:54:18 What “agentic” actually means
  42. 0:56:03 Claude feels more genuinely agentic
  43. 0:57:04 Josh Hart quote analysis detour
  44. 0:58:44 Reverse-engineering creative patterns
  45. 0:59:53 Pizza, calzones, and prompt structure
  46. 1:00:26 SynthID and GPT Image 2 watermarking
  47. 1:01:47 Metadata problems for client work
  48. 1:02:51 Google Pics enters the chat
  49. 1:04:03 Too many image models to track
  50. 1:04:52 Midjourney color still hits different
  51. 1:06:01 GPT Image 2 quality frustration
  52. 1:06:59 Image models still struggle with scale
  53. 1:08:26 Bad AI weeks happen too
  54. 1:09:20 Midjourney 8.2 speculation
  55. 1:10:01 Tell your florist

Questions this episode answers

What did Fast Hours think of Google's Omni and Runway Aleph 2.0 for character swapping?

Rory tested both on a complex Seedance video and found neither could cleanly replace characters — Omni only tracked camera motion, and Runway's panda swap caused the figures to morph mid-shot.

Is Seedance still better than Google's Veo for AI video?

Yes, according to Rory — Seedance and Kling handle complex camera motion and physics better than Veo 3.1, which he calls forgettable and says he'd largely stopped using in favor of Veo 3.

What does Fast Hours mean when they call something 'agentic'?

Drew argues most so-called agentic tools are really just automated workflows; Rory says Claude is the one tool that feels genuinely agentic because you can watch it test and correct its own approach to a task.

Why is GPT Image 2 adding a SynthID watermark?

Drew explains OpenAI is adding a SynthID watermark to GPT Image 2 outputs, matching what Nano Banana already does, though he notes the watermark can be stripped if the image is later edited in a tool like Photoshop.

Does Google's Flow work well for AI video?

Mixed, per Rory — its 'ingredients' feature turning six images into a clip was decent and agent mode generated ten art-directed image variations from one request, but driving physics still looked wrong and two of his custom in-app tool experiments failed outright.

Show notes

Drew and Rory are back for episode 69, which is legally required to begin with at least one immature joke before immediately collapsing under the weight of Google’s latest AI product avalanche.

This week, they dig into Google Omni, Gemini 3.5 Flash, Google Flow, Google Pics, Nano Banana, Veo, and whatever else Google launched before anyone had time to make coffee. The big question: are these actually meaningful creative upgrades, or did Google just throw 19 AI names into a blender and call it innovation?

They break down early Omni and Flow tests, why video physics still feel weird, where Seedance and Kling may still be ahead, and why Runway Aleph 2.0 feels promising but imperfect. Rory shares hands-on examples with character swaps, driving videos, golf swings, agent mode, and Flow’s new tool-building features. Drew tries to keep the conversation coherent while quietly wondering if every AI product now needs a map, glossary, and mild sedative.

The episode also gets into Gemini as a search replacement, creepy context awareness, privacy tradeoffs, AI tools connecting to personal data, the fuzzy definition of “agentic,” the limits of auto-clipping tools, GPT Image 2’s SynthID watermarking, metadata headaches for client work, and the universal pain of wasting $15 trying to make an image model spell “stump.”

If you’re trying to understand what Google’s AI updates actually mean for creators, marketers, AI video workflows, image generation, creative direction, and the future of agentic media tools, this episode is half useful breakdown, half group therapy for people with too many tabs open.

Every word Read the full transcript

If you liked this one