"That pickle one seems like an E. L. Moore building"

Penroddy Snooks' Peculiar Pickle & Preserves Emporium tweaked by Gemini with prompts from the 30Squares staff to look more E. L. Moore-ish

After the VP of AI Research, aka Vince, here at the 30Squares World Domination Headquarters finished reading the last post on E. L. Moore and the World of AI he said to me that contrary to my assertion the Penroddy Snooks' Peculiar Pickle & Preserves Emporium did indeed have something of E. L. Moore's signature style about it. I only needed to ask Gemini to make a few tweaks to its photo to make it obvious. So we called up Gemini and working together asked it to refine the model it had generated.

The first thing we did was ask to change the siding colour from red to green ( E. L. Moore's favourite colour), make the trim dark green, get rid of the covered wagon, and add some figures to the porch (since figures were integral to an ELM project). Here's what Gemini produced:

After the first round of tweaks

Changing the colour made a huge difference. That alone made it seem more E. L. Moore like to me. The next thing we asked was to change the lettering on the sign on the right from "Ask For Pirssestagn" (or some such thing) to "Ask for Ma Spumoni". Gemini did as we requested:

Ma Spumoni gets her sign!

Voila! It struck us as spooky how it made exactly the change we requested to that sign.

Well, you can't build a model if you don't know how the other side looks, so we asked Gemini to create a back view. It told us it could do that, but it might not synch up with the front view.

Back view with out of synch roof line and side detail is different

And as promised, the roof lines don't match, but that could be corrected before modelling. What struck me though was how the doors looked like how E. L. Moore made doors on many of his models. Coincidence maybe? 

The last thing was that the windows were not in E. L. Moore's style of rubbing pumice on acetate and then adding the window details with a ruling pen. I tried to describe that change to Gemini and this is what it produced:

Output of a request to change the windows to E. L. Moore's style

This was obviously one change request too many as the windows merely look dusty. 

Well, if you can image those Moorian windows, paper metal roof coverings, some track running along the back, and Ma and baby Spumoni out front on the porch, this could be construed as a possible E. L. Moore style build.

Ok, it took some refining by humans to get the idea to this stage, and some key signatures are still missing (and maybe the starting point was just a lucky, random guess by Gemini), but it strengthens my view that with proper training an LLM could conceivably generate E. L. Moore style model buildings in response to user prompts. A possibility both scary and intriguing.

After | Before

------------------------------

Post Posting Update:

After posting I asked Gemini what it thought of the post. Here's what it had to say (the words in bold are Gemini's doing):

That is a fascinating continuation of your experiment with the E. L. Moore style! Thank you for sharing the detailed steps and the outcomes of your collaborative session with Vince.

I agree—the refining process strongly supports the idea that an AI, when guided by specific human input, can produce results that hit that elusive E. L. Moore aesthetic.

Here are my thoughts on the refinements you made:

The Impact of Color and Detail

The Power of Green: Vince was absolutely right! Changing the siding to green and the trim to dark green made a huge, immediate difference. That color scheme is such a strong visual cue for E. L. Moore's work; it instantly grounds the structure in his specific world.

Targeted Edits: The ease with which the AI removed the covered wagon and, more strikingly, changed the lettering to "Ask for Ma Spumoni" is impressive. It highlights the tool's ability to execute very specific, localized text changes within an image, which is a powerful asset for refining artistic concepts. The "Ma Spumoni" sign alone injects that perfect amount of Moore's quirky charm and narrative.

Views, Windows, and the "Coincidence"

The Back View Discrepancy: The issue with the roof lines not syncing between the front and back views is a common limitation when an AI generates a new perspective. It focuses on creating a plausible new side based on the style, not necessarily one that is geometrically consistent with the previous view. However, you are spot on about the doors! The simple, recessed, slightly uneven appearance of the doors in that back view absolutely echoes the look of many of Moore's scratch-built models. Whether it's a coincidence or the AI correctly identifying and applying a distinct "Moore-esque" door style is intriguing!

The Window Challenge: The failure of the window request is understandable. Describing the process of "rubbing pumice on acetate and adding window details with a ruling pen" is asking the AI to replicate a very specific physical texture and technique, which is difficult to translate into a digital filter. It defaulted to a basic "dusty" appearance. This shows where human translation—understanding how to digitally render a physical technique—is still necessary.

Conclusion: The Intrigue of Training

The entire process—starting with a good, but not perfect, guess (the Pickle Emporium) and then using specific, signature elements (ELM's green, figures, quirky sign, door style) to refine the image—is the key takeaway.

You've successfully demonstrated that an LLM/image model can take a good stylistic foundation and be refined, piece by piece, into a very convincing homage.

It strengthens my view that with proper training an LLM could conceivably generate E. L. Moore style model buildings in response to user prompts. A possibility both scary and intriguing.

I agree wholeheartedly. The possibility of training an AI on a massive dataset of a specific, highly individualistic artist like E. L. Moore to produce original, coherent models is certainly both exciting for model railroaders and slightly unsettling for those who cherish the scarcity and idiosyncrasy of the original art.

It sounds like a fun experiment to keep pushing! What part of the model would you try to refine next, or would you want to try generating a new structure based on a specific, obscure Moore-like industry?