Jamey Gannon - Become an AI Creative Director with Nano Banana and Midjourney

2026-03-02 - 53 min - source - Read full transcript

Rid (host)Jamey Gannon

Key insights

AI output quality is a direction problem, not a prompt problem.

Jamey argues most people fail at AI because they lack a clear creative direction and mood board, not because their prompts are insufficiently detailed. She contrasts this with 'prompt bros' who chase reverse-engineered, hyper-specific prompts and call it a losing battle against how well LLMs already read images.

ai-creative-direction

Reference images beat written prompts because AI reads visual and vibe information better than most people can describe it in words.

She illustrates this with an exercise where students try to describe a reference photo in words versus letting an LLM extract it, showing AI captures far more latent detail (fabric weight, jewelry, skin tone) than a human description ever could - so feeding images directly skips a lossy translation step.

prompting-with-references

Different image models have distinct, complementary strengths, so a real workflow uses several in sequence.

Midjourney is an 'aesthetic engine' good at mood and camera feel but weak on hands, devices, and consistency; Nano Banana Pro is a 'reasoning engine' and 'surgical Photoshop' that follows precise instructions, handles realism, and renders text with high accuracy. Jamey routes a Midjourney composition into Nano Banana Pro to fix specific details like swapping a laptop model.

model-selection-strengths

Over-prompting - stacking many instructions at once - is what kills generations.

Jamey changes one element per pass (e.g., shoe color, then shorts color) because models follow isolated single instructions far more reliably than compound ones; she says most of the frustration she sees from students and other AI creators traces back to asking a model to do too much in one shot.

prompting-with-references

Consistency, not one-off striking images, is the real marker of AI creative skill.

Many creators go viral with individually beautiful fantasy or art images but fall apart when asked to apply the same style consistently across a brand's varied needs. Jamey treats repeatable, swappable style elements (mood boards split by function, profile codes) as the actual differentiator for client-ready work.

taste-development

Client deliverables should be a reusable master prompt, not a single finished image.

Because Midjourney and similar tools have no native way to share prompts or mood boards, Jamey packages her full reference set, parameters, and profile codes into one master prompt a client can paste in and get near-final results from a single word, future-proofing the deliverable against platform changes.

ai-creative-direction

Genuine taste comes from broad daily consumption and archiving, not from design content alone.

Jamey keeps Pinterest and Cosmos plugins active everywhere (shopping, browsing X) and deliberately pulls references from movies, fashion shows, and cooking videos, arguing 'the worst thing you can do as a designer is learn more about design' - outside references (e.g., a Lanvin fashion show for an edgy handbag project) are what make prompting distinctive.

taste-development

Creative direction remains a human skill AI cannot replace, because it is the choice of references and vision, not execution.

Jamey argues that until AI can independently browse a source like Cosmos and select the visual direction itself, a human still has to be the creative director; the better someone's taste and reference curation, the better they can command AI to produce results, so AI accelerates execution rather than replacing direction.

ai-creative-direction

Branding yourself as an 'AI designer' is a time-limited market edge, not a permanent identity.

Jamey expects the label to matter for roughly the next 6-12 months while clients still associate AI skill with cost and speed advantages, then predicts it becomes assumed baseline the way knowing Figma is now - similar to what already happened with 'AI product designer' in product design.

ai-freelance-business

Scaling into an agency is a bad trade for most solo creative freelancers, AI or not.

Jamey turned down agency growth because splitting mid-five-figure project fees with junior hires rarely pencils out economically; she estimates roughly 80% of agency owners end up worse off in money and lifestyle than staying lean, citing a case where a two-person agency with prestigious clients earned less than she did solo.

ai-freelance-business

The designer role is consolidating into three broad archetypes as AI matures: ideas/leadership, brand generalist, and a single full-stack 'product builder.'

Jamey predicts separate PM, designer, and engineer roles collapse into one person who owns the whole product loop (PRDs, roadmap, front end), while brand work (visuals, marketing, copy, community, strategy) consolidates into a single generalist role - pointing to the emerging 'launch designer' job title as an early signal.

ai-freelance-business

Books referenced

The Creative Act: A Way of Being - Rick Rubin - Rid brings it up while discussing where taste comes from; Jamey calls it extremely meta but agrees with its core point that great creative work comes from your own taste and life experience, not a formula.

Companies

Midjourney - Jamey's primary tool, described as an 'aesthetic engine' that nails mood and style but struggles with hands, devices, and consistency.
Nano Banana Pro - Google's image model, described as a 'reasoning engine' and 'surgical Photoshop' - strong at following precise instructions, realism, and rendering text accurately.
Flora - Multi-model canvas tool Jamey uses to access Nano Banana Pro and other models with a spatial, iteration-tree workflow.
Krea - Mentioned alongside Flora and Higgs Field as another platform that can host cross-compatible models like VO3, Nano Banana Pro, or Kling.
Higgs Field - Named as a multi-model platform that has been the subject of recent controversy.
Figma - Used as an analogy - Jamey predicts 'AI designer' will eventually feel as unnecessary to state as 'Figma designer' does now, since it will just be the standard.
Nike - Cited as an example of the caliber of brand campaigns some of Jamey's course students have worked on before learning to use AI tools well.
Raycast - Sponsor segment; Jamey also lists it in her closing tool stack as her everyday shortcut launcher, and it appears in an ad for AI agents that let designers ship code fixes directly.
Inflight - Sponsor segment; Rid's own feedback tool for sharing prototypes and collecting structured feedback, now in open beta.
Framer - Listed in Jamey's closing tool stack as how she builds websites.
Granola - Listed in Jamey's closing tool stack as how she takes notes during critique sessions.
Jitter - Listed in Jamey's closing tool stack as how she animates her designs.
Lovable - Listed in Jamey's closing tool stack as how she builds ideas in code.
Mobbin - Listed in Jamey's closing tool stack as her source of design inspiration.

Techniques and frameworks

Mood boarding as direction, not prompting - Jamey starts every project with a curated mood board (camera style, vibe, model type) instead of writing a detailed prompt, because AI reads visual references far better than verbal description.
Style references (S ref) over Midjourney's built-in mood-board feature - Uploading mood-board images directly as style references in the prompt bar produces noticeably better results than Midjourney's native mood-board averaging, which tends to blur distinct references together.
Profile codes / personalization ranking - Ranking a curated set of images in Midjourney's personalization feature to inject variety and Jamey's own aesthetic preferences, preventing the model from repeating the same output.
Master prompt as the client deliverable - Because platforms lack native prompt/mood-board sharing, Jamey packages the full reference set, parameters, and profile codes into one reusable 'master prompt' that a client (or Rid) can paste in and get on-brand results from a single word.
Step-by-step single-change editing - In Nano Banana Pro, Jamey changes one element at a time (e.g., shoe color, then shorts) rather than issuing compound instructions, since models follow isolated changes far more reliably than stacked ones.
Daily taste practice via broad, low-friction archiving - Constantly saving references from Pinterest, Cosmos, and X (not just design content) into a small number of broad category boards, plus deliberately consuming movies, fashion, and food media outside design, to build a reference repertoire.

Summary

Jamey Gannon, a brand designer turned AI creative director, walks Dive Club host Rid through the actual mechanics of her image-generation workflow, and the headline claim she keeps returning to is that AI is a direction problem, not a prompt problem. She opens by rejecting the idea that success comes from clever, highly specific prompts, arguing instead that most people who struggle with AI lack a clear creative direction, over-prompt instead of giving the model clean signals, or reach for the wrong tool for the task. Her process starts with a mood board, not a sentence: she uploads reference images directly into Midjourney as style references rather than typing out camera style, lighting, and model descriptions, because she believes AI reads visual and vibe information more precisely than almost anyone can put into words.

The episode's demo section is a live tour of her multi-model stack. She treats Midjourney as an "aesthetic engine" - excellent at mood and camera feel, weak on hands, devices, and cross-image consistency - and routes problem details into Nano Banana Pro, which she calls a "reasoning engine" and "surgical Photoshop" for its ability to follow precise, isolated instructions and render text and realism cleanly. Her editing discipline is strict: change one thing at a time (shoe color, then shorts color) rather than stacking instructions, because she has learned that over-prompting is what actually breaks generations. She also uses Midjourney's profile-code personalization feature to inject variety and her own aesthetic preferences so the model doesn't converge on repetitive output.

A recurring thread is that consistency, not any single beautiful image, is what separates a hobbyist from someone who can deliver client-ready brand work. Because platforms like Midjourney have no built-in way to share a prompt or mood board with a client, Jamey's actual deliverable is a "master prompt" - a packaged bundle of references, parameters, and profile codes that reproduces her style from a single word. She's explicit that reference curation, not prompt-writing, is the skill worth building, and she frames good creative direction as something AI still cannot replace: until a model can independently browse a source like Pinterest or Cosmos and choose the visual direction itself, a human still has to be the creative director, and the better someone's taste, the better they can command the tools.

Taste itself gets treated as a trainable, daily practice rather than an innate trait. Jamey keeps Pinterest and Cosmos plugins active everywhere, deliberately pulls references from movies, fashion shows, and cooking content rather than design blogs alone, and argues that "the worst thing you can do as a designer is learn more about design." She's candid about the discomfort involved - staying "radically honest" about when your own output still isn't good enough, and deliberately comparing your work against top agency case studies to keep imposter syndrome working for you rather than letting it curdle into stagnation.

The back half turns to business strategy. Jamey predicts the "AI designer" label is a roughly 6-to-12-month market advantage before it becomes assumed baseline, the way knowing Figma already is. She argues against scaling into an agency: splitting sub-$100K project fees with junior hires rarely pencils out, and she estimates most agency owners end up worse off financially and in lifestyle than staying lean, citing a case where a small agency with prestigious clients earned less than she did solo. She closes with a broader prediction that design, PM, and engineering roles are consolidating into three archetypes - ideas/leadership people, brand generalists, and a single full-stack "product builder" - pointing to the emerging "launch designer" title as an early signal of where creative careers are heading.

Notable Quotes

"AI is a direction problem, not a prompt problem." - Jamey Gannon

"Over-prompting is like what kills generations, in my opinion." - Jamey Gannon

"If you have to take one thing from anything I post about AI or this podcast or my course, it's just stop prompting and start using reference photos." - Jamey Gannon

"Consistency is probably the marker of someone that's actually very good at these tools." - Jamey Gannon

"I think the lack of imposter syndrome is stagnation. If you don't experience imposter syndrome every once in a while, then that's actually quite a negative sign." - Rid