EP 075 · 99 MIN JUST DRAW IT. STOP PROMPTING EVERYTHING
How to Make AI Images That Don’t Scream “I Used AI”
The short version
Finn McKenty, a former Abercrombie & Fitch designer and fractional CMO, tells Drew Brucker and Rory Flynn why he skips prompting for most of his process: he sketches by hand, hand-recolors Midjourney line art, and references physical, out-of-distribution objects (like a scanned 1968 flea-market jersey font) that no model has ever seen. He argues subject-matter expertise, reading model documentation directly, and material-specific prompting beat speed, and calls fake AI UGC 'cancer' for brand trust.
Key takeaways
- 01
Finn's rule: prompting is maybe 5% of the work. Sketch your idea by hand first, then use that drawing as a structural reference for the model.
- 02
He references genuinely 'out of distribution' objects, like a scanned 1968 flea-market jersey font, since no model has ever trained on something that rare.
- 03
Name the actual physical process in a prompt — 'chenille with a felt layer,' not 'embroidered' — because those specific material words carry real semantic weight.
- 04
Finn reads model documentation himself instead of asking an LLM to summarize it, and keeps a months-long chat thread just for learning how image models actually work.
- 05
He calls fake AI UGC 'cancer' for real brands: the entire point of UGC is that it's real, and audiences don't forgive feeling deceived for decades.
- 06
Higgsfield's terms reportedly grant it a perpetual, irrevocable, sub-licensable license to everything you upload or generate — Rory says no corporate attorney would allow that.
- 07
Finn thinks Midjourney's narrow, aesthetics-only bet was right all along: it's bootstrapped and profitable, so David Holz never has to chase venture-scale returns.
Receipts · said on the record
“There's literally a trillion steering wheels in this thing, and the more you understand what each of those does, the better you can get where you want to go. Slop is just people pressing on the gas with no hands on the wheel.”
“The value is in doing things that are either out of distribution completely or at the edge of the distribution. All that stuff in the middle is worthless.”
“It's cancer … unless you're some dropshipper, fly-by-night brand that's not gonna exist in three months. But if you're any real brand, that should be just absolute cancer.”
“You just can't. No corporate attorney that looks at that is gonna be like, cancel this account immediately.”
“Before, it was like when we create crazy face portraits with paint and different textures, and it was fun. Then it got like, how do you build this production pipeline, and it became like, now I have to write 50,000-word system prompts.”
Chapters
- 0:00:00 Who is Finn McKenty?
- 0:03:13 Why did Finn quit YouTube?
- 0:05:30 Do creators need a large audience?
- 0:12:05 Which existing skills create an AI edge?
- 0:14:26 Does expertise accelerate AI mastery?
- 0:16:27 Are creatives too attached to process?
- 0:18:17 Why do companies fail to adopt AI?
- 0:20:17 Why does advanced AI work cause burnout?
- 0:27:33 Where should creators invest brainpower?
- 0:29:31 How do creators steer probabilistic AI?
- 0:31:27 What made Midjourney click for Finn?
- 0:34:00 How did AI produce 10x SEO traffic?
- 0:37:44 Why reject polished AI perfection?
- 0:38:41 Why does handmade design feel better?
- 0:41:06 How should AI and manual craft combine?
- 0:46:24 Why does making more have zero value?
- 0:50:57 How does material knowledge improve prompts?
- 0:55:03 Why choose Weavy over ComfyUI?
- 0:55:44 How do sketches improve AI images?
- 0:58:33 How should creators divide work with AI?
- 1:02:12 What does out-of-distribution mean?
- 1:05:56 How do original references beat AI slop?
- 1:07:03 Why does fake AI UGC destroy trust?
- 1:11:31 What went wrong with Higgsfield?
- 1:22:33 Why did Midjourney acquire Co-Star?
- 1:24:09 Has Midjourney had the right vision?
- 1:27:30 Why are creators returning to Midjourney?
- 1:29:10 Why does bootstrapping protect creativity?
- 1:34:01 Why does AI creative work need fun again?
- 1:35:04 Does direct attribution kill creativity?
Questions this episode answers
Who is Finn McKenty?
Finn McKenty is a fractional CMO and partner at URM Academy (known for Nail the Mix), with a prior career as a designer at Abercrombie & Fitch and Hollister, an early team member at CreativeLive, and a YouTuber with roughly 750,000 subscribers across two channels.
What does Finn McKenty mean by 'out of distribution' in AI images?
It's a machine-learning term for anything a model's training data never covered, so the model literally cannot produce it well, if at all. Finn's example is a font scanned from a real 1968 high-school jersey found at a flea market: because no image model has ever seen it, using it as a reference steers the output somewhere no generic prompt could reach.
Why does Finn McKenty sketch by hand instead of just prompting?
He estimates prompting is only about 5% of the real work, so he'll doodle an idea (even on his phone) and feed that drawing in as a structural or style reference. He argues you should spend your limited time and skill on the part you're genuinely good at, like hand-drawing letters, and let AI handle the part that would otherwise take disproportionately long, like rendering a chrome effect.
What did Fast Hours say about Higgsfield's terms of service?
Drew Brucker reads out language that reportedly grants Higgsfield a perpetual, irrevocable, worldwide, sub-licensable license to everything a user uploads or generates. Finn McKenty and Rory Flynn agree no corporate attorney would let an enterprise use the tool under those terms, and Finn says he personally canceled his subscription and found the cancel flow deliberately hard to complete.
Does Fast Hours think Midjourney made the right call by not chasing realism?
Finn argues yes: rather than compete with Nano Banana, GPT Image, and Flux as a fourth generalist model, Midjourney doubled down on aesthetics with its own tightly integrated model and front-end. Being bootstrapped and profitable, he says, lets founder David Holz make that call without venture-capital pressure to chase generalist parity.
Show notes
AI images stop looking generic when creators steer models with real expertise, original references, hand-built assets, and selective craft.
Finn McKenty joins Drew Brucker and Rory Flynn to explain why the fastest route to better AI work may involve drawing by hand, reading the manual, and making fewer things. The three alleged efficiency experts also wander into AI burnout, fake UGC, Higgsfield trust problems, Midjourney’s long game, and the deeply inconvenient possibility that one great image may beat the 1,000 they can generate before lunch.
Topics and tools covered: Midjourney, Weavy, ComfyUI, Claude, Gemini 2.5, GPT Image, Nano Banana, Flux, Krea, Figma, Adobe Photoshop, and Higgsfield anchor the practical discussion. Sketch references, image references, style references, out-of-distribution inputs, material-specific prompting, systems thinking, AI adoption, fake UGC, creator trust, and durable AI-assisted SEO define the core concepts.
