CHAPTERS
Why consistent brand imagery needs a repeatable process (not endless prompting)
Jamey frames the core challenge: one-off great images are easy, but building a cohesive brand library requires a tight, repeatable workflow. Claire sets the goal—creating a consistent portfolio of “pink internet-coded” brand visuals without spending all day prompting.
- •Consistency across a set matters more than a single standout image
- •A “manicured process” reduces time wasted on trial-and-error prompting
- •Goal: create brand assets that feel uniquely on-brand and scalable
- •Tools mentioned as part of the broader workflow: Midjourney, Nano Banana, Flora
Step 1: Build a mood board to define the visual language
The workflow starts in Pinterest (or Cosmos) by collecting a mood board that captures the intended vibe—color, contrast, subject weirdness/juxtaposition, and overall aesthetic. The mood board becomes a shared visual spec for both the human and Midjourney.
- •Use Pinterest/Cosmos to collect a cohesive “general vibe” mood board
- •Juxtaposition (objects in unexpected contexts) helps create a distinctive brand feel
- •Mood boards provide visual language when you lack design/photography vocabulary
- •A more consistent board (e.g., same subject/style) tends to yield clearer results
Step 2: Turn mood board images into Style References (SREFs)
Jamey shows two ways to use a mood board in Midjourney: paste images directly as a mood board, or use them as SREFs. SREFs often work better for capturing consistent style traits like coloring, contrast, and camera treatment.
- •SREFs = style references that transfer vibe, palette, and treatment
- •You can drag/paste images directly from your Pinterest board into Midjourney
- •Initial goal is speed: generate fast to learn what inputs are doing
- •Don’t be precious in early tests—use simple prompts to probe the style
Diagnose mismatch: Compare generations vs. mood board and adjust inputs
They review early outputs that look cool but don’t match the target aesthetic. Claire highlights a technique: ask a text model (ChatGPT/Claude) to describe why outputs differ, helping non-designers build the vocabulary needed to correct course.
- •Be brutally honest: “cool” images aren’t enough if they’re off-brand
- •Mismatch cues: saturation/contrast, washed-out grading, inconsistent vibe
- •Use an LLM to explain differences and suggest style descriptors
- •Early phase is about identifying what the model is over/under-emphasizing
Fix the palette and bias: Remove overpowering references and refine SREF set
Jamey demonstrates iterative pruning: a single strong color element (e.g., green eyeshadow) can dominate generations. By removing that SREF, results shift closer to the desired neutral/pink balance and composition.
- •Midjourney may “average” a broad mood board—SREFs can be more controllable
- •One dominant element can steer the entire output distribution
- •Remove problematic references instead of fighting them with long prompts
- •Saved SREF images persist in a personal library for reuse later
Step 3: Add Personalization Codes to lock in your signature style
To deepen consistency and make outputs feel more like ‘your’ look, Jamey layers in personalization codes (trained via Midjourney’s preference voting). She explains how to build and manage these profiles and warns about unintended style bleeding.
- •Personalization codes are trained by choosing between image pairs (or skipping)
- •Create multiple profiles (e.g., “late 2025 aesthetic”) for different use cases
- •Skipping strategy: select only what you’d actually want to generate
- •Watch for “style bleeding” when liking too many images with a distinct art style
Step 4: Use publication/style keywords as high-information shortcuts
Jamey begins light prompting with references like “Dazed editorial,” “Vogue,” or “high fashion,” which compress complex style directions into a few words. This approach avoids long prompt blocks while still steering lighting, contrast, and editorial tone.
- •Magazine/publication names can encode lighting, contrast, and fashion cues
- •“Editorial” is a powerful shorthand for composition and polish
- •Short, human language prompts can outperform overly technical prompt walls
- •Goal: raise quality while keeping prompts “lazy” and repeatable
Image References and cropping: control composition without fighting the model
To match a specific pose/composition, Jamey uses image references—then crops out the dominating detail (bubblegum) to prevent it from hijacking results. The key idea: edit the reference, not the prompt, to remove unwanted bias quickly.
- •Image references guide composition but often influence style too
- •Crop/zoom in Midjourney UI to remove distracting or overpowering elements
- •Avoid spending hours on “no bubblegum” style negation attempts
- •Image reference can also inadvertently transfer facial structure and features
Step 5: Minimal prompting for subject + setting + style (with “camera cheat codes”)
Jamey shows when she does add prompt details: defining setting cues (NYC skyline, luxury apartment, time of day) and swapping materials (matte black leather couch). She also uses camera mentions as quick styling levers to shift realism and era.
- •Use simple scene directives: subject + setting + style framework
- •Single words like “luxury” can imply high-rise, upscale interior context
- •Camera names/types act as shortcuts for realism and time-period aesthetics
- •Iterate with small prompt deltas instead of rewriting everything
Scaling the asset set: variations, Explore page inspiration, and prompt “stealing”
To expand from a few wins into a full brand library, Jamey keeps the same SREF stack and generates across many subjects (tech, culture, motifs). She uses Midjourney’s subtle/strong variations to fix issues (like hands) and mines the Explore page for reusable prompt patterns.
- •Reuse the same SREF stack to maintain a consistent brand through-line
- •Use subtle/strong variations to correct common artifacts (e.g., hands)
- •Pull inspiration and prompt structure from Midjourney Explore for breadth
- •Build a cohesive set by testing multiple subject categories in the same style
Reinforce and remix: create a new mood board from your best outputs
Once a style is working, Jamey selects ~30 strong images from her gallery and turns them into a new mood board to “lock in” the look. She also demonstrates combining multiple mood boards (e.g., ‘Real Skin’)—then notes SREFs still tend to be more reliable for consistency.
- •Turn your best generations into a new mood board to reinforce the style
- •Blend multiple mood boards to introduce realism (e.g., skin) or other traits
- •Expect domain mismatch: an editorial SREF set won’t optimize for animals
- •Sometimes you must restart references (e.g., find NatGeo-like animal refs)
Delivering to clients: share the recipe (profiles, refs, settings) in Figma
Jamey explains how she packages the work so clients can self-serve: final prompt setup, profiles, SREF images, and key settings—often pasted into Figma since Midjourney lacks robust sharing. Claire highlights how this shifts creative services from retainer dependence to upfront system-building.
- •Deliver the “system,” not just the images: refs + profiles + key settings
- •Figma becomes the handoff hub: paste final prompt and reference set
- •Most outputs can come from a single consistent setup once dialed in
- •This model empowers clients to generate new assets without re-hiring immediately
Post-production workflow: fix hands/logos and upscale with Nano Banana (via Flora/Higgsfield)
For final polish, Jamey takes Midjourney images into Flora/Higgsfield and uses Nano Banana as a “talk-to-Photoshop” tool. She demonstrates upscaling and swapping an unrealistic laptop for a specific ‘2026 midnight black MacBook Pro’ while preserving composition and style.
- •Nano Banana is used like Photoshop: targeted edits, swaps, cleanup, upscaling
- •Prompt with constraints: ‘don’t change anything else,’ keep size/angle constant
- •Reasoning models can recognize common objects without reference images
- •Useful for fixing Midjourney issues: hands, logos, weird computers, resolution
Inspiration and troubleshooting: build taste libraries, take breaks, and edit inputs not prompts
In the lightning round, Jamey shares her inspiration system: X/Twitter lists, Pinterest/Cosmos saving habits, and a daily taste practice. Her troubleshooting philosophy is to step away, identify the real failure mode (too busy, too many refs, wrong color), and simplify or rebuild references.
- •Create an inspiration pipeline: X lists, Pinterest boards, Cosmos searches
- •Keep mood boards simple to avoid decision fatigue; archive everything for reuse
- •When stuck: take a break, then return to diagnose the real issue
- •Solve problems by removing/adjusting references instead of expanding prompts
