EP 068 · 87 MIN THEY ROASTED A REAL MONET AS AI SLOP
The Prompt Is Dead, Long Live the Reference + The Monet Trap
The short version
Episode 68 argues that AI prompting is shifting from typed text to visual references: Rory builds reusable 'style card' sheets in Codex and GPT Image 2, and Seedance's Omni-reference carries voice and environment between shots. Freepik rebrands to Magnific, Midjourney V8.1 remains the best model for faces despite a returning blue/pink color cast, and a viral prank reveals people rate a real Monet as obviously fake AI art.
Key takeaways
- 01
Freepik rebranded to Magnific mid-episode; its real differentiator is enterprise indemnification across roughly 40 image and 35 video models in one account.
- 02
Pinterest built its own internal AI image model (Pinterest Canvas) trained solely on its own photos — not public, but a sign more platforms will do this.
- 03
Claude now has MCPs into a Meta Ads Manager and into Higgsfield, letting an agent pull data and run bulk ad creative without a dashboard.
- 04
Rory builds a reusable 'style card' in Codex and GPT Image 2 — color rules, exposure curve, prompt nucleus — then skips re-writing the prompt each time.
- 05
Midjourney V8.1's old blue-and-pink color cast is creeping back, especially at higher --exp values; re-ranking images without that hue helps reset personalization.
- 06
The viral 'Monet' AI-art prank proves the point: people rated a cropped real Monet painting as obviously bad fake AI art once they thought it was AI-generated.
- 07
Seedance's Omni-reference can take a voice clip as an ingredient to carry spatial-depth-aware dialogue into a generated shot — still rough, but the best hack going.
Receipts · said on the record
“It's like, everything's going to be agentic. If it can just interact with the infrastructure, I can just tap into Claude and Claude goes in there and pulls the data that I need and brings it back and runs reports.”
“Claude has an MCP for your meta ads manager. Now Claude has an MCP for Higgs Field to build like workflows and you know, like generate and batch.”
“This guy's collecting all these responses. It takes off and reveals that this is actually a Monet painting that he cropped into.”
“We used to this with clients. We used to give them two different images. One was real, one was AI. And say the real one was AI and say the other one was real. And they always choose the real one no matter what.”
“I'm happy people wanna make stuff. Just keep making stuff, keep being weird. Like that's it. Like that's what this is all about.”
“There's nothing better in the game for shots of people. You can't convince me otherwise. I'll go head to head with you on creating real imagery of faces. And this is just unbeatable at this point in time.”
Chapters
- 0:00:00 Cold open
- 0:01:12 AI news fatigue is real
- 0:01:44 Claude Code runs the day now
- 0:03:11 Remote work and coffee shop crimes
- 0:08:21 3 Ninjas nostalgia break
- 0:10:09 Freepik becomes Magnific
- 0:11:47 Why Magnific works for enterprise
- 0:13:00 Model aggregators vs internal tools
- 0:18:01 Pinterest builds its own AI image model
- 0:22:11 Salesforce goes "headless"
- 0:24:08 Higgsfield, MCPs, and Meta ads
- 0:29:51 Codex for GPT-Image-2 workflows
- 0:32:45 Pulling style from video frames
- 0:34:05 Building visual style reference sheets
- 0:37:36 Codex and textured illustration systems
- 0:39:34 The evolution beyond text prompts
- 0:41:50 Seedance storyboards and visual prompts
- 0:43:01 Reference images as reusable seeds
- 0:44:51 AI video still has an audio problem
- 0:46:13 Audio reference hacks in Dreamina
- 0:50:52 Omni-reference for video control
- 0:52:37 Midjourney V8.1 updated take
- 0:53:28 The blue and pink problem returns
- 0:55:37 Midjourney still owns realistic faces
- 0:58:56 Reworking old prompts with Describe
- 1:01:10 Luma Uni vs Midjourney color
- 1:02:11 Midjourney 8.2, 8.3, and V9
- 1:03:46 Midjourney edit model questions
- 1:07:13 Midjourney plus Seedance films
- 1:09:03 Midjourney’s strange lane
- 1:12:28 The Monet AI social experiment
- 1:15:13 Why people over-detect AI
- 1:20:09 AI backlash and disclosure debates
- 1:22:50 AI as a career unlock
- 1:24:30 Keep making weird stuff
- 1:25:45 Wrap-up and seamstress CTA
Questions this episode answers
What did Fast Hours say about Freepik becoming Magnific?
Rory and Drew note Freepik quietly rebranded to Magnific, the name of the upscaling tool it had acquired years earlier; they credit the platform's strong enterprise indemnification and roughly 40 image and 35 video models in one stack for its traction with brand teams.
Is Midjourney still the best model for realistic faces, according to Fast Hours?
Yes — Drew calls Midjourney V8.1 unbeatable for photorealistic faces and skin texture, though both hosts note the old blue-and-pink color cast has crept back in, especially at higher --exp settings.
What is the Monet AI experiment Fast Hours discussed?
A creator posted an AI-generated 'Monet-style' image and asked people to critique it; commenters tore it apart as obviously fake AI art, then he revealed it was actually a real Monet painting he'd cropped — proof, Rory and Drew say, that people judge images by their label, not their content.
How are Rory and Drew using Claude and MCPs for agency work?
Rory describes Claude connecting via MCP directly into tools like a Meta Ads Manager and Higgsfield, arguing this 'headless,' agentic approach — also seen in Salesforce's shift away from a heavy UI — lets an agent pull data and run workflows without a human clicking through a dashboard.
What's Rory's style-reference-sheet workflow with Codex and GPT Image 2?
Rory extracts frames from a reference video, has Codex and GPT Image 2 build a standalone 'style card' (color grade rules, exposure curve, a prompt nucleus), then reuses that single reference image to generate consistent sets of on-brand images without re-writing the prompt each time.
Show notes
In this episode 68 of the Fast Hours Podcast, Drew and Rory return from a two-week hiatus to prove that yes, the AI news cycle did continue without their permission. Rude.
They dig into Freepik changing its name to Magnific, why enterprise AI image tools are starting to feel more like creative operating systems, and how brands may be better off using approved model aggregators instead of building weird internal Franken-tools that immediately become outdated.
Then things get nerdier. Obviously.
Rory breaks down how he’s using Codex, GPT-Image-2, Claude Code, MCPs, Higgsfield, Seedance, and visual style reference sheets to create repeatable image systems, character references, and bulk creative workflows without living inside a giant text prompt forever. Drew pushes into where Midjourney V8.1 still dominates, especially photorealistic faces, color, texture, and images that do not look like corporate stock photography that lost the will to live.
They also talk about Midjourney’s upcoming 8.2, 8.3, V9 roadmap, edit model ambiguity, personalization drift, Luma Uni comparisons, Pinterest’s internal AI image model, Salesforce going headless, and why AI video audio still sounds like it was recorded inside a cursed podcast booth.
And because no episode is complete without accidentally getting philosophical, they close with the viral Claude Monet AI social experiment, the weird bias people bring to AI-generated images, and why “how it was made” keeps hijacking whether people can actually see what’s in front of them.
Basically, it’s an episode about the future of AI creative tools, with two guys trying to sound calm while the ground turns into soup beneath them.
