- \n
- What does \u201cremix\u201d mean here (not collage, not filter)?<\/li>\n
- What are the constraints that can disqualify you?<\/li>\n
- What does a jury usually reward in this kind of competition?<\/li>\n<\/ul>\n
Then I asked Gemini to take a step further: assume the jury mindset<\/strong>. Even without knowing names, you can infer the logic: museum culture values craft, coherence, restraint, and concept. A jury may appreciate bold ideas \u2014 but they punish sloppy execution fast.<\/p>\n
That early analysis gave me a north star:<\/p>\n
If the physics doesn\u2019t work, the concept doesn\u2019t matter.<\/strong><\/p>\n
\nStep 2: Define a \u201cwinning vibe\u201d \u2014 not a subject<\/h2>\n
This was the first major decision: I didn\u2019t want to \u201cdecorate Vermeer.\u201d I wanted to collide worlds<\/strong>.<\/p>\n
We defined the vibe as:<\/p>\n
Dutch Golden Age stillness<\/strong> meets digital-era anxiety<\/strong> \u2014 without losing realism.<\/p><\/blockquote>\n
It had to feel like a real moment, not a staged fantasy. A believable scene with a crack in reality.<\/p>\n
\nStep 3: The modern lens \u2014 street photography, not \u201cAI aesthetics\u201d<\/h2>\n
I already work with street photography in the AI space (including my ongoing project \u201cStreets of Chicago\u201d), so I leaned into what I know: the feeling of a captured moment.<\/p>\n
Gemini suggested three street photographers whose visual language could bridge classical calm with contemporary strangeness. I analyzed all three and chose Julie Hrudov\u00e1<\/strong>.<\/p>\n
Why?<\/p>\n
Because her work isn\u2019t just \u201csurreal.\u201d It\u2019s often surreal in a street-photography way<\/em>:<\/p>\n
- \n
- directness<\/li>\n
- flash energy<\/li>\n
- a sense of \u201ccaught in the act\u201d<\/li>\n
- the humor and discomfort of reality behaving slightly wrong<\/li>\n<\/ul>\n
That mattered \u2014 because the weakest AI images tend to look too polite. Too balanced. Too \u201cmade.\u201d<\/p>\n
Hrudov\u00e1\u2019s influence gave me permission to break the perfection.<\/p>\n
\nStep 4: Choose the right Old Master \u2014 not the most famous one<\/h2>\n
Gemini proposed three Rijksmuseum-connected anchors:<\/p>\n
- \n
- Vermeer \u2014 The Little Street<\/em><\/li>\n
- Rembrandt \u2014 The Night Watch<\/em><\/li>\n
- Breitner \u2014 The Bridge<\/em><\/li>\n<\/ul>\n
We chose Vermeer\u2019s View of Houses in Delft (The Little Street)<\/em> (1658)<\/strong> for a very practical reason:<\/p>\n
It\u2019s quiet enough to accept disruption.<\/p>\n
Rembrandt fights back \u2014 it\u2019s already dramatic. Vermeer doesn\u2019t. Vermeer invites intrusion, which makes the intrusion louder.<\/p>\n
Then we did the part people skip: we analyzed Vermeer\u2019s DNA \u2014 not the story, but the mechanics:<\/p>\n
- \n
- composition geometry<\/li>\n
- texture contrast (brick vs plaster vs wood)<\/li>\n
- light logic<\/li>\n
- the calm of everyday domestic architecture<\/li>\n<\/ul>\n
That \u201cDNA\u201d became my checklist for realism later.<\/p>\n
\nStep 5: Find a real modern street that carries the same soul<\/h2>\n
I asked Gemini a strange question:<\/p>\n
If Vermeer painted today in Amsterdam, where might he stand?<\/strong><\/p>\n
Gemini suggested three streets with similar \u201cold Dutch passage\u201d energy. We explored them and chose Begijnensteeg<\/strong>, because it offered:<\/p>\n
- \n
- a portal-like depth<\/li>\n
- heavy brick texture<\/li>\n
- a natural stage for light vs shadow<\/li>\n
- enough visual silence to make one surreal detail feel enormous<\/li>\n<\/ul>\n
Next, we planned the shot like a real photographer would:<\/p>\n
- \n
- framing<\/li>\n
- focal length feeling<\/li>\n
- light direction<\/li>\n
- the moment of contrast<\/li>\n<\/ul>\n
This mattered because if the \u201cbase photo\u201d is weak, every later step becomes damage control.<\/p>\n
\nStep 6: Generate the base \u2014 then accept that the real work starts after<\/h2>\n
Using Gemini \/ Nano Banana Pro prompts, I created a strong base image in the spirit<\/em> of Hrudov\u00e1: a street-photography frame that already felt plausible.<\/p>\n
Then I pushed it toward a remix.<\/p>\n
At this stage, my goal wasn\u2019t \u201cmore weird.\u201d My goal was:<\/p>\n
Make the original feel true \u2014 then introduce one impossible thing.<\/strong><\/p>\n
So I started adding elements:<\/p>\n
- \n
- a surveillance camera<\/li>\n
- a heron perched on it (a natural watcher)<\/li>\n
- voxel\/pixel \u201cdecay\u201d in the wall \u2014 as if the street is turning into a digital simulation<\/li>\n
- subtle modern hints (like the scooter), placed so they feel discovered, not inserted<\/li>\n<\/ul>\n
This is where the process turned into an endurance sport.<\/p>\n
\nStep 7: The brutal part \u2014 physics, anatomy, and the tyranny of shadows<\/h2>\n
The hardest phase wasn\u2019t creativity. It was coherence.<\/p>\n
Every time I added something, Gemini became my technical critic:<\/p>\n
- \n
- Is the shadow direction consistent?<\/li>\n
- Does the bird actually grip<\/em> the camera, or is it floating?<\/li>\n
- Do the pixels have weight and depth, or are they stickers?<\/li>\n
- Do the modern materials look like metal and glass, or like painted clay?<\/li>\n
- Does the scene still feel like one photograph<\/em>?<\/li>\n<\/ul>\n
This created loops. Many loops.<\/p>\n
That\u2019s the hidden truth:<\/p>\n
High-quality AI art is iterative art direction.<\/strong>
\nIt\u2019s \u201cgenerate \u2192 critique \u2192 fix \u2192 critique \u2192 fix.\u201d<\/p>\nNot once. Dozens of times.<\/p>\n
\nStep 8: The \u201cmasterstroke\u201d detail \u2014 the reflection<\/h2>\n
At some point, we hit a conceptual gap: the image was beautiful, but still felt like a tableau.<\/p>\n
Gemini suggested a single addition that changed everything:<\/p>\n
Put the photographer\u2019s reflection in the security camera dome.<\/strong><\/p>\n
This did three things at once:<\/p>\n
- \n
- A sophisticated nod to Dutch painting traditions (hidden reflections and optical play).<\/li>\n
- A street-photography trick: it becomes a captured event, not a staged scene.<\/li>\n
- A narrative loop: the watcher is being watched.<\/li>\n<\/ol>\n
The reflection had to be subtle, warped by the dome curvature, not a selfie. Done right, it becomes a reward for the viewer who leans in close.<\/p>\n
Done wrong, it becomes gimmick.<\/p>\n
We iterated until it felt physically believable and narratively necessary.<\/p>\n
\n![\"\"](\"https:\/\/darkopavic.art\/wp-content\/uploads\/2026\/01\/end-3-Gemini_Generated_Image_om72v6om72v6om72-201x300.png\")
Finished artwork<\/figcaption><\/figure>\n <\/p>\n
Step 9: Packaging for jurors \u2014 title and statement that don\u2019t overtalk<\/h2>\n
I asked Gemini to help craft the title and artist statement \u2014 not to sound \u201cAI poetic,\u201d but to land fast with jurors who read dozens of entries.<\/p>\n
My title: The Glitch in Delft<\/strong><\/p>\n
And the statement was built to do one job:<\/p>\n
- \n
- explain the concept in plain language<\/li>\n
- name the key visual loop (surveillance \u2192 heron \u2192 reflection \u2192 viewer)<\/li>\n
- frame the voxel decay as meaning, not decoration<\/li>\n<\/ul>\n
In competitions, clarity wins.<\/p>\n
\nStep 10: Finish like a printmaker, not a prompter<\/h2>\n
The final step was production: upscaling, noise handling, sharpening, and print-readiness.<\/p>\n
If you\u2019ve ever printed large, you know the rule:
\nA1 size will reveal every weakness.<\/strong><\/p>\nSo I treated the file like a serious print:<\/p>\n
- \n
- controlled noise reduction (don\u2019t smear texture)<\/li>\n
- restrained sharpening (don\u2019t create halos)<\/li>\n
- checks at 100% on critical areas: edges, sky transitions, the heron, the reflection<\/li>\n<\/ul>\n
The goal wasn\u2019t \u201ccrispy.\u201d The goal was \u201cbelievable.\u201d<\/p>\n
\nWhat I learned<\/h2>\n
If I had to compress the whole experience into a few sentences:<\/p>\n
- \n
- AI doesn\u2019t replace craftsmanship \u2014 it relocates it.<\/li>\n
- \u201cMuseum-grade\u201d is 80% about physics and restraint<\/strong>.<\/li>\n
- The difference between \u201cnice\u201d and \u201cserious\u201d is coherence: light, weight, texture, intent.<\/li>\n
- The best AI workflow is less about prompting and more about creative direction and ruthless editing<\/strong>.<\/li>\n
- A strong concept is necessary \u2014 but realism is the gatekeeper.<\/li>\n<\/ul>\n
Or, put simply:<\/p>\n
If your shadows lie, your story collapses.<\/strong><\/p>\n
\nIf you want, I can also produce a shorter LinkedIn version of this (tight hook + 3 lessons + 1 image + CTA), or a Forbes-style \u201c5 principles\u201d format that\u2019s more skimmable.<\/p>\n","protected":false},"excerpt":{"rendered":"
Most people think AI art is about the prompt. It\u2019s not. If you\u2019re trying to create something that could survive a serious jury \u2014 especially one connected to a museum tradition \u2014 AI becomes less like a magic wand and more like a brutally honest creative partner: it helps you see the gaps you didn\u2019t… <\/p>\n
- Rembrandt \u2014 The Night Watch<\/em><\/li>\n
- Vermeer \u2014 The Little Street<\/em><\/li>\n