EP 067 · 66 MIN A $150K TYPEFACE, BUILT IN 30 MINUTES
GPT Image 2 Is Good. But Is It Nano Good?
The short version
Episode 67 marks the show's rename from Midjourney Fast Hours to Fast Hours and centers on OpenAI's new GPT Image 2. Rory turns a logo brief into a complete, working typeface, alphabet, numerals, ligatures and an OTF file, in about 30 minutes using GPT Image 2 and Claude, work Drew notes can otherwise cost $10K to $500K. Both hosts also find Veo 3.1 currently unusable next to Seedance and Kling.
Key takeaways
- 01
Rory turned a logo brief into a full OTF font file in about 30 minutes, using GPT Image 2 and Claude.
- 02
Custom typefaces normally cost real money: Drew cites $10K-35K for a single display face and $150K-500K for a full enterprise type system with global language support.
- 03
Instead of asking creators for their prompt, feed the reference image to an LLM and ask it to reverse-engineer a reusable prompt template.
- 04
GPT Image 2 beats Nano Banana on sports-graphic micro-detail (stitching, stubble, bat grain) but still shares Nano's slightly noisy/grainy skin rendering.
- 05
Veo 3.1 and 3.1 Lite are currently unusable next to Seedance, Kling and LTX — Rory saw people walking through walls.
- 06
Kling's new native 4K generation is a real step forward for video text and skin detail, because more pixels finally give small text room to render.
- 07
Claude Design locked Drew out of his $200/month Max plan for a week after he ran through his tokens testing it — a rough first impression.
Receipts · said on the record
“That shouldn't be possible. I shouldn't be able to get a new font file in 30 minutes. I should not. It makes me feel weird about some things. But it's also like, you need people to do weird stuff.”
“I'm just like, are you kidding me? I pay $200 a month. You guys should not have rolled this out if I gotta wait a week to re-access this sh*t. That's unbelievable.”
“There is a middle ground that all these brands have to decide on. And I don't think it's either spectrum, all or nothing ever.”
“I know it was hand done and it's got somebody's personality and time embedded into it. I'm still gonna f*cking pay the money for that.”
“For the micro details that they're getting, like, look at the beard, right? I'm seeing the stitching on the SL logo on his hat.”
“Unusable in comparison to what else is out there. Like, in comparison to Seedance and Kling 3 and LTX and like, I don't know. It just doesn't hold up anymore.”
Chapters
- 0:00:00 Fast Hours is (re)born
- 0:03:36 Going tool-agnostic
- 0:04:34 GPT-Image-2 quietly drops
- 0:05:31 Text becomes the unlock
- 0:07:31 The AI backlash returns
- 0:10:57 Hype, fear, and the middle
- 0:12:10 Typography gets weird
- 0:14:50 What custom fonts cost
- 0:15:43 GPT-Image-2 vs Nano Banana
- 0:17:39 Rory’s font experiment
- 0:18:47 Fiddleheads become a typeface
- 0:19:39 Building the type deck
- 0:20:36 The nine-slide image unlock
- 0:21:14 Geometry, spacing, and logic
- 0:22:11 Turning images into font files
- 0:23:02 Micro-text gets better
- 0:24:19 Claude builds the font package
- 0:26:41 The revision loop changes
- 0:27:50 Context is the silver bullet
- 0:32:11 Drew makes a handwriting font
- 0:35:35 Why designers obsess over type
- 0:37:52 Reverse-engineering prompts
- 0:39:51 Richer color and sports graphics
- 0:41:27 Fixing artifacts and details
- 0:42:37 Nano Banana vs GPT-Image-2 tests
- 0:44:26 Sports realism gets scary good
- 0:45:27 Why teams need this now
- 0:46:43 Freepik settings and ratios
- 0:48:36 Testing, tokens, and limits
- 0:49:44 Brand kits and rebrand concepts
- 0:53:19 Google I/O and the next model
- 0:53:48 Veo 3.1 falls behind
- 0:55:04 Kling adds native 4K
- 0:56:40 Character sheets and macros
- 0:58:07 Rebrands as visual prototypes
- 1:00:53 Building a reference library
- 1:01:36 Three weeks in a row
- 1:02:58 Claude Design tease
- 1:03:37 Tell your local [fill in the blank] spam finale
Questions this episode answers
How did Fast Hours turn an image into a working font file?
Rory generated a full typography system (alphabet, numerals, ligatures, brand applications) in GPT Image 2 from a logo and a fiddlehead reference image, then had Claude read the geometry and output a complete OTF/TTF font package in about 30 minutes.
Is GPT Image 2 better than Nano Banana?
Drew and Rory call it roughly even to slightly ahead — GPT Image 2 renders sharper micro-text and sports-graphic detail, but it shares Nano Banana's slightly noisy, grainy skin texture, and OpenAI has gotten more conservative about real-person likeness.
What did Fast Hours say about Veo 3.1?
Rory calls Veo 3.1 and 3.1 Lite currently unusable next to Seedance, Kling and LTX, citing broken physics like people walking through walls and car doors that won't shut.
How much does a custom brand typeface normally cost?
Per the figures Drew reads on the episode: a single custom display face runs $10K-35K, a small brand family $25K-75K, a full brand type system $75K-150K, and an enterprise-wide proprietary system $150K-500K.
Why did Fast Hours drop 'Midjourney' from the podcast's name?
Rory and Drew explain the show outgrew being only about Midjourney — they now cover the whole generative AI ecosystem of image, video and LLM tools — so they renamed it Fast Hours, while saying Midjourney is still special to them and isn't going away from the coverage.
Show notes
Fast Hours has entered the witness protection program. Same Drew. Same Rory. But fewer syllables and more chaos.
In this episode, Drew Brucker and Rory Flynn officially drop “Midjourney” from the podcast name and relaunch as Fast Hours, a broader home for the creative AI ecosystem: image models, video models, LLMs, vibe coding, Claude, ChatGPT, Midjourney, and whatever tool drops five minutes after they hit publish. Naturally, the rebrand lasts about four minutes before they’re elbows-deep in GPT-Image-2, OpenAI’s new ChatGPT image model that quietly showed up and immediately started making designers question their calendar, career choices, and relationship with kerning.
The big topic: GPT-Image-2 is shockingly good with text, typography, brand systems, visual decks, product mockups, and multi-image outputs. Rory walks through how he used ChatGPT and Claude to create a custom typeface from visual references, generate a premium typography presentation, extract geometry, and turn the whole thing into usable font files. Drew then shows how he turned his own handwriting into a working typeface, because apparently “personal brand” now includes making your lowercase g file a tax asset.
They also dig into the uncomfortable middle ground of AI creative work: when it saves time, when it still needs human judgment, why anti-AI panic and AI hype both miss the point, and why the real advantage is context. Not prompts. Not magic buttons. Context.
The episode also covers GPT-Image-2 vs Nano Banana Pro, richer color rendering, micro-text improvements, AI-generated sports graphics, brand kit concepts, Freepik settings, Claude Design, 4K video generation, Kling, Veo 3.1, Seedance, and the strange reality that a custom brand typeface can now go from “that’ll be $150K” to “Rory did it before lunch.”
Basically, it’s an episode about the exact moment creative production stops feeling like a tool demo and starts feeling li ke a factory someone accidentally left unlocked.
