EP 032 · 89 MIN MIDJOURNEY HAS NO IDEA WHAT A WALKMAN IS
Midjourney Questions Answered
The short version
Episode 32 is Fast Hours' first live audience Q&A, streamed through Goldcast, recapping Midjourney office-hours news on V7's timeline and consistent-character jump before Rory and Drew take listener questions live. Topics include de-modeling faces, stacking personalization on moodboards, juggling Midjourney against Runway, Kling, MiniMax, Veo and Sora, and Midjourney's blind spots on specific cultures and old tech like a Sony Walkman.
Key takeaways
- 01
To de-model a face, stack tokens like "everyday" and "natural," negative-prompt "no makeup" or "no airbrushing," and train a personalization profile on ordinary iPhone photos of real people.
- 02
Midjourney's David reportedly rated the then-current consistent-character feature 4/10 and promised V7 would hit 9/10 - the hosts plan to judge that skeptically.
- 03
Stack your global personalization profile on top of any moodboard: tested live, it visibly shifts composition and skin texture closer to the creator's own taste.
- 04
Keep more than one personalization profile, for example a gritty default, a softer look, and a purely photo-realistic one, and switch between them depending on the shot.
- 05
For flat 2D or vector illustration, drop stylize to roughly 20-50; Midjourney's default of 100 tends to add unwanted depth and a 3D-rendered look.
- 06
Midjourney defaults "Brazil" to favela imagery and doesn't recognize a Walkman, showing how prompts for specific cultures or old tech can hit blank training data.
- 07
For period-accurate photos, lower stylize to about 70-75 and quality to 0.5, and prompt "documentary" or "amateur photography" rather than "editorial" or "cinematic."
Receipts · said on the record
“It can't look fake, it just cannot have that AI feel to it. It needs to look like it was taken at that time. I even tried putting a Walkman in, and it's just like, what is a Walkman? I don't know what that is.”
“Midjourney sometimes feels like, Brazil, everybody's poor. So you just write down Brazil and then it just makes favelas or Rio. It just forgets that it's a huge country and there's so much culture here.”
“David said something that kind of shocked me. Someone asked what he'd rank the current consistent-character feature, and he said it's like a four out of ten, but the next one's going to be a nine out of ten.”
“I call it the prompt elevator: I type in the word car, see what it generates, then add in a village, then with a fire in the background, building step by step so you can see what each token does.”
“It's like getting the SREF. Once the SREF is cracked, the rest of it falls into place. It's finding that SREF and translating it into the prompt that's the hard part.”
“We've hit this critical mass of how much further you can push photo-realism or style. It's like now it has to go into storytelling.”
Questions this episode answers
Was Episode 32 of Fast Hours a normal recording?
No. It's Fast Hours' first live audience Q&A, streamed through Goldcast with listeners submitting and upvoting questions and a few brought on stage. Rory and Drew recap Midjourney office-hours news before taking questions live.
What did Fast Hours say about Midjourney V7's consistent characters?
The hosts relayed that Midjourney's David rated the then-current consistent-character feature 4 out of 10 and promised the V7 version would hit 9 out of 10, while saying they'd judge that claim skeptically once it ships.
How do you stop Midjourney faces from looking like a stock-photo model?
Rory suggests tokens like "everyday" or "natural," negative prompts such as "no makeup" or "no airbrushing," and a personalization profile trained on ordinary iPhone photos of real people to pull faces away from Midjourney's default polished look.
Why does Midjourney default "Brazil" to favelas?
An audience member said prompting "Brazil" consistently produced favela or Rio imagery that erased the country's size and variety. The hosts attributed it to biased training data and suggested prompting a specific neighborhood or era instead of the country name.
What stylize setting works best for flat 2D or vector illustration in Midjourney?
Rory recommends dropping stylize to roughly 20-50, since the platform's default of 100 tends to add unwanted depth and a 3D-rendered look to flat illustration or vector-style prompts.
Show notes
Fast Hours Podcast Ep. 32—Midjourney Questions Answered (with Drew Brucker & Rory Flynn)
Drew and Rory decided to take this episode live, which means things got real fast. Questions flew in from all over the world, Rory’s toddler made her obligatory guest appearance (in spirit), and somehow, we managed to sound like we knew what we were talking about.
We covered the latest Midjourney updates (V7? Maybe. Maybe not.), tackled impossible prompting challenges (Brazil is not just favelas, thank you), and gave our best shot at solving the age-old problem of getting Midjourney to chill out with its flawless model faces. Also, Walkmans? Apparently, AI has no idea what those are.
If you’ve ever wondered how to create consistent characters, why stylize settings actually matter, or what happens when you bring audience members on stage, this one’s for you. Just don’t ask us to prep next time—clearly, we do better without it.
🎧 Listen now on YouTube, Spotify, or Apple Podcasts.
