Nov 15, 2024 · 1h 3m · startup-ideas
The ULTIMATE guide to Midjourney (AI Design Tutorial)
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Midjourney expert Nick St. Pierre and host Greg Isenberg present a masterclass on advancing from basic text prompts to high-end visual craftsmanship using style references, numeric Sref codes, model personalization, and canvas inpainting.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Greg holds 12.1% of the talking time here. How this is scored →
Greg as informed peer 1.6 Guest teaching 4.8 Guest disagreement 0.1 Greg pushing back 0.0
speaking balance: gold is Greg, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 6:08 Telling users to get out of Discord
Nick emphatically orders users to abandon Discord in favor of the new web application, calling Discord a hellish environment.
Hardest push from Greg ▶ 55:40 Challenging the premise that AI art is inherently crappyGreg admits his longstanding skepticism that Midjourney generated unconvincing, non-artistic assets before bringing on Shahed Khan to test Nick's output.
Biggest teaching moment ▶ 30:50 Demonstrating model personalization vs default stylizeNick proves that default high-stylize generations scream generic AI, while user personalization radically transforms the aesthetic into high-end analog photography.
Greg holds their own ▶ 45:30 Synthesizing the slot-machine reframe as stealing like an artistGreg demonstrates sharp conceptual understanding by articulating how deterministic style mixing transforms AI prompting from random gambling into genuine artistic synthesis.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Greg as informed peer | Guest teaching | Guest disagreement | Greg pushing back | Why |
|---|---|---|---|---|---|---|
| Shifting from Text Prompting to Visual Driving | 1 | 5 | 1 | 0 | Greg admits he has been relying on basic Twitter prompt lists and text prompting. Nick gently educates him on moving beyond semantics into visual-driven style spaces. | |
| Selecting Visual Inspiration from Vintage Posters | 2 | 2 | 0 | 0 | Greg provides creative inspiration by sharing vintage Japanese exhibition posters from his Twitter feed, which Nick adopts immediately as baseline reference material. | |
| Navigating the Midjourney Web App and Explore Feed | 1 | 4 | 0 | 0 | Nick guides Greg through the Midjourney web UI and explore feed, emphasizing stealth mode for client work, while Greg asks clarifying questions about public generation visibility. | |
| Understanding the Creative Value and Power of Midjourney | 3 | 2 | 0 | 0 | Greg prompts Nick on the commercial and creative utility of Midjourney, offering his own perspective on how high-level assets can live across landing pages and social media. | |
| Testing Initial Style References and Prompt Parameters | 1 | 6 | 0 | 0 | Nick demonstrates live prompting with the giraffe concept, teaching Greg how style weights (--sw), aspect ratios, and minimal text prompts calibrate visual output. | |
| Distinguishing Style References from Image Prompts | 1 | 6 | 0 | 0 | Nick explicitly breaks down the mechanical difference between standard image prompts (which pull subject matter and pixels) and style references (which decouple aesthetics from subject). | |
| Sponsor Message: BoringMarketing.com | 1 | 4 | 0 | 0 | Following an ad read for BoringMarketing.com, Nick demonstrates subtle versus strong variations and using the editor to paint out unwanted text elements. | |
| Calibrating Model Personalization and Stylize Settings | 2 | 7 | 0 | 0 | Nick explains how rating images trains personalized models (--p) and how stylize parameters alter output. Greg admits his biggest gripe was generic Midjourney aesthetics and had no idea this lever existed. | |
| Exploring Multi-Dimensional Sref Style Codes | 2 | 7 | 0 | 0 | Nick introduces style reference codes (sref) as multi-dimensional coordinates and showcases his database tool. Greg suggests Midjourney build a creator marketplace around personalizations. | |
| Fine-Tuning Sref Weights and Elevating AI Artistry | 3 | 5 | 0 | 0 | Nick illustrates multi-style weight syntax (::) to eliminate the feeling of playing a slot machine. Greg connects this method to Austin Kleon's 'steal like an artist' philosophy. | |
| Executing Permutation Prompts for Optimal Style Weights | 1 | 6 | 0 | 0 | Nick demonstrates permutation bracket prompting to batch-test multiple style weights simultaneously. Greg is stunned by the high quality of the resulting giraffe art. | |
| Contextual Mockups Using the Inpainting Canvas Editor | 1 | 5 | 0 | 0 | Nick uses the canvas editor to place the generated artwork into a museum gallery setting, removing style codes to allow natural photographic background rendering. | |
| Shahed Khan Joins for a Full Workflow Review | 2 | 4 | 0 | 0 | Loom co-founder Shahed Khan joins to review the workflow and confirms the output looks museum-grade. Nick recaps the entire pipeline and shares his Pioneer Works physical lab plans. |
Statements from this episode (13)
St. Pierre: Visual references unlock aesthetics beyond semantic text prompts
“We can really do a lot more when we're using visuals to drive visuals as opposed to semantics to drive visuals. We can kind of like explode out aesthetics and style space into almost like infinite directions. We don't have to rely so much on the base models in…”
St. Pierre recommends defining a subject and medium for AI design
“I kind of like to, I kind of like to start usually with like a subject and like maybe a medium if I'm after a particular medium, like a painting or an illustration or a photograph. Or something three D. It's nice to kind of start there because we can bend and …”
Midjourney generations are public by default without Pro stealth mode
“So if your account is on the standard plan or below, everything is kind of public by default. When you're on the pro plan, you can go into stealth mode and you can prompt privately.”
Midjourney style weight parameter defaults to 100 and maxes at 1000
“And one of those parameters is style weight, dash dash sw. What this will let you do is determine how much influence that this style reference has over the generation. So by default, it's at a hundred, but it goes all the way up to a thousand. So if we run thi…”
Midjourney style references decouple aesthetic style from an image's subject
“So Midjourney is essentially looking at this image and it's trying to understand what makes it what it is. And it tries to do that by decoupling aesthetic from subject matter.”
St. Pierre: Dragging images into Midjourney defaults to image prompts
“So this is an important distinction because a lot of times people will just, you know, they'll just drag an image in and by default it's an image prompt, but there's a huge difference When you click that little paperclip.”
St. Pierre explains the difference between Midjourney's Subtle and Strong variations
“Now, Variation Subtle, what they're going to do is keep most of the details of this image. They're going to keep most of the details intact, but it's going to change, like, the subtle things. Variation Strong are going to have more dramatic effects on the comp…”
St. Pierre predicts Midjourney v7 will enable model personalization by default
“Personalization is something everybody should do. And it's probably going to be default with the seven.”
St. Pierre: Midjourney is a new medium blending complex layered aesthetics
“I think of me mid journey is like a new medium entirely, right? Like it's a different type of canvas. It's a different way of problem solving. You have not just paint, but you have like entire aesthetics that have all of these layers and this complexity and yo…”
St. Pierre: Midjourney's 'style raw' parameter produces more photorealistic results
“I'll even drop like style raw, which leans a little bit more photo realistic.”
Khan: MoMA visitors would never guess high-end Midjourney art was AI-generated
“Like, I, honest to god, if you printed a million of these and just had a exhibit in MoMA, I don't think anyone would ever guess that it was made using AI.”
St. Pierre: Midjourney has over 4 billion unique aesthetic style codes
“There's like two to the 32nd power of these style references, these S ref codes and each kind of is like coordinates to its own little aesthetic universe.”
St. Pierre is building a physical creative AI lab in Red Hook
“So we're building a lab downstairs here in Red Hook in partnership with Pioneer Works. And we're essentially trying to create a space for people to come and learn how to use these tools at a deep level, not just mid journey, but all of the creative AI tools. B…”