CHAPTERS
- 0:00 – 3:41
Why consistent brand imagery needs a tight Midjourney process
Jamey frames consistency as less about magical prompting and more about a repeatable, manicured workflow. Claire sets the bar: generating a single great image is easy, but building a coherent brand library is the real challenge.
- •Consistency requires a system, not just ad-hoc prompting
- •Goal is a scalable ‘brand portfolio’ of images, not one-offs
- •Promise of a workflow that reduces time spent ‘prompting all day’
- •Context: using Midjourney plus tools like Nano Banana and Flora
- 3:41 – 4:42
Step 1: Build a mood board to establish visual language
Jamey starts every project by collecting a mood board in Pinterest or Cosmos to define the vibe. The board acts as a non-verbal creative brief that communicates aesthetic intent faster than text.
- •Mood boards define the ‘general vibe’ and visual codes upfront
- •Juxtaposition and ‘internet-coded’ details help create a distinctive aesthetic
- •Mood boards can be pasted directly into Midjourney or converted into references
- •Claire highlights mood boards as a substitute for missing design vocabulary
- 4:42 – 8:48
Why mood boards sometimes fail (and how to diagnose mismatch)
Early generations look cool but don’t match the mood board, revealing that Midjourney may average out diverse references. Claire suggests using AI (e.g., ChatGPT/Claude) to articulate why outputs diverge, helping non-designers build visual language.
- •Midjourney mood boards can ‘average’ styles when references are broad
- •You must evaluate outputs against the intended vibe with brutal honesty
- •Specific visual gaps: saturation, contrast, washed-out treatment
- •Use critique prompts to learn descriptive style language
- 8:48 – 11:14
Step 2: Use SREFs (style references) instead of (or alongside) mood boards
Jamey shifts to SREFs for stronger stylistic control and faster convergence. By removing one overpowering reference (green eye makeup), she steers the palette back toward the desired neutral-pink direction.
- •SREFs transfer overall style: color, camera treatment, vibe
- •SREFs often outperform mood boards for generalized ‘vibe’ goals
- •Remove dominant references that skew generations (e.g., too green)
- •Use Midjourney’s library to reuse pasted image references anytime
- 11:14 – 12:40
Test prompts for checking style consistency across subjects
To validate whether a style holds up, Jamey uses repeatable ‘test prompts’ like ethereal female model, cats (texture-rich and abundant training data), runners, and astronauts. The aim is broad-enough subjects that reveal whether the aesthetic transfers reliably.
- •Keep a small set of reusable test prompts to benchmark style
- •Cats/running/astronauts reveal lighting, texture, and composition quickly
- •Use prompts broad enough to let the style show through
- •Consistency checks happen early before investing in refinement
- 12:40 – 16:48
Step 3: Iteration with personalization codes to add your signature style
Jamey introduces Midjourney personalization codes—profiles trained by ranking image pairs—to bake in a preferred ‘late 2025’ look. She warns about style bleeding (unintended drift toward painterly/vintage looks) and emphasizes good naming because profiles aren’t easily editable.
- •Personalization codes are built via ‘this-or-that’ image rankings
- •Profiles can be layered on top of SREFs for depth and crisp modern results
- •Skipping strategically helps avoid training on mediocre examples
- •Name profiles well; you can’t easily edit them later
- 16:48 – 18:19
Prompting shortcuts: magazines, ‘editorial,’ and cultural keywords as compression
Jamey demonstrates how referencing known publications (e.g., Dazed, Vogue) transmits complex photographic direction with minimal text. These shorthand terms encode contrast, fashion sensibility, and production value without long, brittle prompts.
- •Magazine names can act as high-information style tokens
- •‘Editorial’ is a reliable shortcut for elevated photography cues
- •Use simple, human language (e.g., ‘up close macro photo’)
- •Avoid overly technical or JSON-style prompts when possible
- 18:19 – 21:31
Step 4: Image references + cropping to control composition (and remove overpowering elements)
Image references help lock composition, but they can also drag unwanted elements into the generation (like bubblegum). Jamey solves this by cropping/zooming in Midjourney to exclude the distracting detail, then re-running the prompt—faster than fighting with negative prompts.
- •Image references primarily guide composition but often influence style
- •Crop/zoom to remove dominant elements (bubblegum) instead of ‘no’ prompts
- •Dragging the cropped image back in can preserve quality and structure
- •Use editing steps in Midjourney UI to steer results quickly
- 21:31 – 29:01
Step 5: Iterative prompting for subject + setting + style (the deer-in-NYC example)
Jamey shows a practical iteration loop: keep the style stack constant and adjust only what’s missing in the scene. Small additions—‘NYC skyline in the window,’ ‘matte black leather couch,’ ‘at night,’ plus a camera callout—incrementally nail the concept.
- •Add missing scene info with plain language rather than complex prompts
- •Use a ‘luxury’ keyword to imply high-rise, upscale interior details
- •Camera names serve as fast styling modifiers for realism/era
- •Iteration targets specific failure modes (too fantastical, wrong setting)
- 29:01 – 30:02
Scaling the aesthetic: generate a broader brand set, fix artifacts, and borrow from Explore
With a stable SREF stack, Jamey scales across themes (tech, culture, motifs) while maintaining cohesion. She uses Midjourney’s subtle/strong variations for fixes (like hands) and mines the Explore page for prompt patterns when building larger sets.
- •Reuse the same SREF ‘recipe’ to expand across many subjects
- •Use ‘very subtle/very strong’ to iterate away common artifacts
- •Explore page is a prompt inspiration library for stock-style breadth
- •Aim for a cohesive set that still has enough variation to avoid looking flat
- 30:02 – 35:56
Packaging for clients: build a final mood board and deliver a reproducible recipe
Once the target look is achieved, Jamey curates successful outputs into a new mood board and tests it on fresh prompts (e.g., turtle). She documents the final setup—profiles, SREFs, references—and shares it in Figma so clients can self-serve future assets without re-hiring an agency for every shoot.
- •Create a second, ‘validated’ mood board from your best generations
- •Combine mood boards (e.g., ‘Real Skin’) to influence realism and texture
- •Accept that some subjects (animals) may need new references to work well
- •Deliverables: prompt recipe + references + example outputs in Figma
- 35:56 – 43:19
Targeted refinements with Nano Banana (Photoshop-by-prompt) and Flora workflows
Jamey uses Nano Banana as a precision editor: upscale, swap objects (like replacing a laptop), and preserve composition. She also demonstrates building realistic AI self-portraits by combining reference selfies with style/pose references in Flora for thumbnails and social content.
- •Nano Banana excels at targeted edits (object swaps, upscales)
- •Use constraints: ‘don’t change anything else,’ ‘keep position/size same’
- •Reasoning models can infer common items (e.g., MacBook Pro) without refs
- •Self-portrait workflow: reference selfies + pose/style references in Flora
- 43:19 – 47:04
Building a visual reference library: X lists, Cosmos, Pinterest systems, and taste practice
For ongoing inspiration, Jamey maintains curated feeds—especially an X (Twitter) list of aesthetic accounts—and uses Cosmos and Pinterest as structured archives. She emphasizes consistent saving/archiving as a ‘daily taste practice’ and keeps mood boards simple to avoid decision fatigue.
- •Create an X list of ‘Tumblr-like’ aesthetic accounts for steady input
- •Cosmos is strong for art direction and brand inspiration; Pinterest remains a hub
- •Use browser plugins to save inspiration from anywhere into boards
- •Keep boards broad and reusable to reduce decision fatigue
- 47:04 – 49:49
Troubleshooting when AI won’t cooperate: step away, diagnose, simplify, and recompose
Jamey’s core troubleshooting tactic is taking a break to regain clarity, then returning to identify the real problem (too busy, too many SREFs, a dominant color). The fix is often subtractive: simplify references, rebuild the mood board, and align with how the model ‘wants’ to interpret inputs.
- •Take a break; fresh eyes reveal the true failure mode
- •Diagnose precisely: ‘too busy,’ ‘too Midjourney,’ ‘style bleeding’
- •Remove dominant colors/elements even if you like the reference
- •Stop forcing the model—adapt the workflow to how it responds
