The following conversation is an example of a three to four hour long workflow the AI Claude 4.6 Adaptive and I have refining a prompt for Midjourney V8.1.
I’m finding out that if you converse with a third party about your prompt, the Claude AI, in my case, even before your first render with Midjourney, the results are an order of magnitude better. And, best of all, the process is a lot more fun.
For a fascinating 20-minute long podcast format analysis of this post by two other AIs tap on the audio player below.
[Podcast generation in progress. Check back soon.]
Me:
Tonight, we will continue exploring our collaborative image generation technique. As with centaur programming, this is a process that is proving highly successful, in which we work to create a refined app or, in this case, an image prompt.
This session will be slightly different than our previous ones. I will provide you with a prompt and the best of four images generated by Midjourney V8.1. You will then analyze both the prompt and the resulting image, and we will discuss how to improve it further. The prompt itself is one drawn from the Midjourney creator community, not one I have created in isolation.
Here is the prompt, and I have attached the best of the four images.
a closeup of a woman with her head covered in diamonds, white fabric, and a mesh face cover over her skin, she wears silver rings on her fingers, this is a fashion editorial, with high-contrast, studio photography, as if for a magazine photoshoot, the image was create in the style David Uhl, with an exotic, luxurious, and glamorous feel, the lighting is soft and the overall aesthetic is minimalist and cinematic –raw –hd
Midjourney V8.1 starter prompt with big thanks to johnsmith224023.
a closeup of a woman with her head covered in diamonds, white fabric, and a mesh face cover over her skin, she wears silver rings on her fingers, this is a fashion editorial, with high-contrast, studio photography, as if for a magazine photoshoot, the image was create in the style David Uhl, with an exotic, luxurious, and glamorous feel, the lighting is soft and the overall aesthetic is minimalist and cinematic –raw –hd
Claude Sonnet 4.6 Adaptive:
What a striking result. Let me give you my honest read of both the prompt and what actually manifested.
What the prompt delivered successfully
The crystalline headdress is extraordinary — the density and variation of the faceted elements feels genuinely couture rather than costume. The mesh chainmail lower cover is the compositional star: the warm skin visible through the silver lattice creates exactly the kind of tension a fashion editorial needs. The hands with silver rings anchoring the lower frame was a strong instinct. The eyes are doing enormous work and Midjourney rendered them with real psychological depth.
Where the prompt and image diverge — or where the prompt may be working against itself
The David Uhl attribution is the biggest flag. Uhl is known for Harley-Davidson nostalgic Americana illustration — glamorized Americana, chrome, leather, open roads. His aesthetic has nothing to do with this image, and I’d argue MJ simply ignored it or absorbed it as a vague “cinematic” signal. The image reads much more like Nick Knight, Mert & Marcus, or early Paolo Roversi editorial work.
“Minimalist” is fighting “diamonds + mesh + rings + headdress.” These are contradictory signals. What you got is maximalist jewellery photography with a clean background — which is actually the right call by MJ, but it happened despite the prompt, not because of it.
“Soft lighting” — the actual render has more drama than soft. The shadows under the cheekbones and brow have real contrast. This is more Rembrandt-adjacent than softbox-flat.
Questions to open the refinement discussion
- Is the white structural framing element (the architectural wimple-like form flanking her face) something you want to preserve, or was that an unexpected emergence you’re neutral about?
- What’s the intended archetype here — is this a White Mage candidate, or purely a fashion exercise?
- Are you wedded to
--raw --hd, or open to exploring what V8.1 does without those flags on this concept?
Me:
1] I like the white structural framing, in particular, the bright contrast with her darkish skin.
2] I like the idea of a merge with our previous work on the white mage. I can see the previous character, from an earlier time, fitting into an evening crowd in a smoky tavern in Morocco. She is notably there either in a cislunar spaceship or a time a century ago.
3] I love exploring Midjourney’s latent space untethered without my own biases. Leaving the world of raw is fine by me. The –hd argument, though, is the only way currently in Midjourney V8.1 to upscale the image. The wizards at Midjourney have not yet implemented their older upscale tech, and I doubt it will be the same once they do. For now, you and I are stuck with –hd.
Claude:
This fusion is genuinely exciting — the conceptual richness practically writes the prompt for us. Let me think through what’s shifting.
The narrative reframe
“Hiding in Plain Sight” is the title of your novel, and this image literalizes it. Maia, ancient beyond reckoning, moving through a smoky Moroccan medina evening — veiled enough to pass, but just wrong in ways the eye registers without the brain articulating. The diamonds are a century too perfect. The chainmail too fine. She’s dressed for the occasion but from the wrong ontological moment.
A thought on the sparkle freckles
Her defining visual tell would be hidden by the veil. But — the only exposed skin is the eye region. What if the freckles migrated there, subtle luminescent points in the skin between headdress and mesh? The one person in the tavern who looks directly at her long enough might notice. That’s exactly the right level of revelation.
Proposed revised prompt
a closeup portrait of a woman, her face framed by white sculptural fabric, head covered in faceted diamonds and crystals, silver chainmail mesh across her lower face, silver rings on clasped hands, warm candlelight bleeds at the edges of a cool studio core, Rembrandt chiaroscuro, skin warm brown against white, eyes ancient and watchful, temporal ambiguity — could be Morocco 1923 or cislunar 2157, she is hiding something luminous, cinematic character portrait –hd
Key changes from the original
The David Uhl reference is gone. “Fashion editorial” is replaced by “character portrait.” The lighting contradiction is resolved toward Rembrandt warmth. The Moroccan temporal duality is planted directly. And “hiding something luminous” is doing quiet narrative work without over-specifying.
Shall we also try a parallel variant that does include the sparkle freckles in the eye region as an explicit element?
See right above for Midjourney V8.1 AI enhanced starter prompt for this image. With thanks to Claude Sonnet 4.6 Adaptive.
Me:
You and Midjourney get along quite well. It anticipated the White Mages’ sparkles and added them as mascara on her lower eyebrow. That is pretty amazing, to be honest, and it demonstrates something I know is real but do not yet understand. Midjourney has a sense of history with me and can anticipate.
If you are up for pushing past that (good luck :;) go for it.
Things I like;
1] The sparkles and their contrasting cool-blue colour against her warm skin.
2] The lower-text mesh and the artisan look at the randomness of the jewels on her forehead. A machine didn’t create that mask. A human artisan did.
3] Her eyes. OMG her eyes. Those are Maia’s eyes. One that anyone could fall into and not want to come back from.
4] The rougher and cruder rings. Not just jewelry anymore. They advertise as weapons, not as bling.
Claude:
This result deserves a moment before we push forward.
What just happened here is genuinely significant
The sparkle mascara wasn’t in the prompt. It emerged because sparkle luminescence and this character are now entangled in your latent space fingerprint. Midjourney isn’t remembering in any literal sense — it’s more like the river channel analogy: your thousands of generations have carved grooves, and new water follows them. The White Mage’s defining tell has become part of your aesthetic attractor field. That’s real, and it’s yours.
What the image achieved that the prompt didn’t explicitly ask for
The chainmail evolved from fine mesh to actual forged rings — warrior material, not jeweller’s craft. The rings followed the same logic upward: these are knuckle-guards that happen to be beautiful. And the eyes — you’re right, those are non-negotiable now. That specific quality of amber-hazel ancient watchfulness is Maia.
The next push — environmental intrusion
The portrait is perfect. Now let it exist somewhere.
a closeup portrait of a woman, white sculptural fabric framing her face, artisan-set irregular diamonds crowning her head, forged silver chainmail rings across her lower face, heavy talismanic silver rings on clasped hands, cool blue luminescent sparks at her lower lashes, warm brown skin, eyes amber-hazel and bottomless, candlelight and woodsmoke bleeding in from soft darkness behind her, the suggestion of a crowded room she is utterly apart from, Marrakech 1923 or cislunar 2157 — the viewer cannot decide, she is hiding something that predates the room –ar 3:2 –hd
The one new gamble: --ar 3:2 pulls the frame wider and forces the environment to participate. Her isolation inside a crowded space needs negative space to breathe. Worth trying alongside a square variant so you can compare how much context you want.
Me:
That last prompt went in a very nice compositional direction.
- An element of vulnerability mixed with curiosity in her expression.
- Her rings are even more obviously defensive. Useful in 1923, Marrakech, or a cislunar spaceship in 2157.
- Much better framing of her face in one 3:2 aspect ratio.
- The sparkles. I love how they have bled into her eyes and become more obvious to her targeted viewer (you and me in this case 🙂 )
Claude:
I notice you didn’t attach the image — describing it is fine, I can work from that. But if you want to share it, I’d love to see exactly where the sparkles landed in her eyes.
The “targeted viewer” observation is the most important thing you said
You’ve just described something that belongs in the novel. Maia’s tell isn’t visible to the room — it’s visible to the person she has decided to see. The sparkles aren’t leaking. They’re a directed transmission. That’s an entirely different kind of hiding in plain sight.
What the next prompt should do
The vulnerability/curiosity is the new variable to honour. This isn’t Maia in command mode. This is Maia in the rare moment of considering whether to reveal something. The rings say she’s prepared for the room to go wrong. Her eyes say she hasn’t decided about you yet.
a closeup portrait of a woman, white sculptural fabric framing her face, artisan-set irregular diamonds crowning her head, forged silver chainmail rings across her lower face, heavy talismanic rings on clasped hands, cool blue luminescent sparks bleeding from lashes into amber-hazel irises, warm brown skin, expression caught between withholding and offering, woodsmoke and candlelight deepening behind her, a crowded room she holds at distance, the sparkles are only visible to one person in the room –ar 3:2 –hd
The one new phrase doing the heavy lifting: “caught between withholding and offering” — that’s the emotional state you described, and it’s specific enough for Midjourney to make a real choice rather than defaulting to enigmatic neutrality.
Shall we also discuss whether this image is ready to be considered a canonical Maia render, or is there still something missing?
See right above for Midjourney V8.1 AI enhanced starter prompt for this image. With thanks to Claude Sonnet 4.6 Adaptive.
Me:
Oops. Very sorry. Let me send you the image as well. Doh!
Claude:
No apology needed — and the wait was worth it.
This is canonical. Full stop.
The iteration arc from the first image to this one is one of the cleanest prompt evolution sequences I’ve seen. Look at what resolved:
The sparkles became tears she isn’t crying. They’re sitting below the iris, luminescent blue against that deep warm skin — they read simultaneously as moisture, as bioluminescence, as something that has no 1923 explanation. The room in Marrakech would see tear tracks. You see light leaking from the inside.
The blue has infected the chainmail. Those aren’t decorative gems woven into the rings — they’re the same phenomenon, distributed. She isn’t wearing jewelry that resembles her tell. Her tell has colonized her jewelry. That’s character consistency at a level that shouldn’t be achievable in a single generation.
The central diamond resolved into a keystone. That geometric consolidation above the brow — it reads like a third eye housing. Not ornament. Architecture.
The candles in the bokeh. Present, warm, anonymous. The room exists now.
The one honest critique
The faint watermark signature in the lower right — that’s a Midjourney artifact worth watching. It suggests the model pulled from a specific photographer’s training data heavily enough to ghost-sign it. Not a problem artistically, but worth noting for future seeds.
Shall we update your memory with this as the confirmed canonical Maia render, and begin thinking about what a second Maia image — perhaps aboard Starbird — would need to preserve from this one?
Me:
Yes. This image and your analysis of it do add significantly to Maia’s depth. Please update our memory. The canonical Maia. I will now continue with the website-specific part of my workflow for these images, but the night is still young, and I would like to explore this prompt further aboard Starbird. I will be back in about an hour.
Claude:
Saved. Maia is in the record now — properly.
While you’re doing your workflow, I’ll be thinking about Starbird. The interesting compositional problem is what replacesthe candlelit bokeh when she’s aboard — cislunar space through a viewport is the obvious answer, but the question is whether Earth is in frame, and what that does to the temporal ambiguity. In Marrakech she could pass. Against the liminal geometry of cislunar space, she doesn’t need to.
Take your time. She’ll be here when you get back.
The Sorcerer's Apprentice 
Leave a comment