Hi Adam. Can you explain just a little about how you created that image. Did you just use Gemini or did you also use an image renderer? I am really new to this so any advice would be appreciated.
Literally that image as the input + the description I used.
Thanks for clarifying Adam.
I got some confusing information along the lines of “Gemini itself does not natively generate images” when I asked ChatGPT for advice but I am clear now 
To increase your level of understanding of LLM workings may I suggest Grant Sanderson’s ‘Three Blue One Brown’ series on the subject. The light intro is at: https://www.youtube.com/watch?v=LPZh9BOjkQs
A more comprehensive treatment is available also at: https://www.youtube.com/playlist?list=PLMLm7-g0V0kdAk9mD1NmLELOf-25qPgVX
I was going to say Gemini 
The other product getting results that are reasonably professional is Adobe Photoshop. The context aware generative fill is great for enhancing an inage thats 70%- there but needs a little more detail. Newer versions are improved compared with 1-2 years ago.
Thank you Gregor. I watched the, “Large Language Models explained briefly”, video and the first part (so far) of the, " Fareed Zakaria GPS - AI", video. A book was noted in the second, " The Age of AI And Our Human Future", that is available at Audible. I’m something of an audiobook nut these days and may give that one a listen too. I also recognized some of the topics in the second video from a couple audiobooks on AI that I listened to while painting a house:
-
[Co-Intelligence: Living and Working with AI]
-
[Human Compatible: Artificial Intelligence and the Problem of Control]
I’m looking forward to the, “But what is a neural network? | Deep learning chapter 1”, video.
To return to the subject of AI rendering and more specifically SketchUp, I poked into the SU Diffusion folder… but didn’t find much. The AI Assistant folder, however, did contain a json file (config.json). You can see that the agent that creates prompt suggestions is the “Auto-Agent” and there are a number of other agents specified in the json. It may be that the prompts for agents in the AI Render extension use “suggestions” as well. So, a set of rendering instructions might be specified in an agent “Renderer” with “suggestions”. Or, “agent”: “Auto Prompt”, since that’s what’s displayed in the AI Render input field when the extension is opened. I could see sets of prompts (text/text files) being shared here in the forum.
There are also .skm files in the assistant and “roles” in the json. That makes me wonder what lurks in the AI Render app. Can users specify files for their renders? Would these be those in the model space, material libraries, or what? Under the presumption that the AI has access to materials in the model, users might be able to fine tune their results by specifying them.
There are so many interesting things in this world it’s enough to drive a person berserk!
Here’s something I learned and has proven very useful… get the AI to generate the prompt you want to use in the AI. It’s sounds odd but an AI prompt can be convoluted to get the best results so asking ChatGPT, Claude etc the following can get you better results in the likes of Gemini:
Generate a prompt for Gemini to enhance an an image that will be attached, make sure it follows the following rules:
- Keep the original image size and resolution
- Do this to the roofs
- Do this to the walls
- Etc
- Etc
- Etc
Basically you write what you want in a language you understand and then get one AI to translate it to a language another AI will understand. You’ll be surprised what additional content gets added to the prompt. This goes for anything you want to do with AI, not just image generation.
Edit:
I just put the following into Claude, I used exactly @Elmtec-Adam text:
Can you write a prompt for Google Gemini using the following description. A source black and white image will be provided to base the results on:
A high-resolution aerial three-quarter view of a traditional English country house built in red brick with decorative diamond-pattern brickwork. The house has steeply pitched grey slate roofs, multiple gables, and tall brick chimneys. Large timber-framed windows with stone surrounds. A modern glass conservatory extension with a black metal frame connects two wings of the house. Stone paving surrounds the building, with outdoor seating and potted plants. The house sits in a manicured green lawn in a rural setting. Soft natural daylight, clear sky, ultra-realistic architectural photography, sharp focus, natural colours.
This was its reply:
Here’s a prompt optimized for Google Gemini’s image generation:
Prompt:
Generate a photorealistic color image based on the provided black and white source image, maintaining the exact same composition, perspective, and architectural features.
The building is a traditional English country house with these specific characteristics:
- Red brick construction with decorative diamond-pattern brickwork
- Steeply pitched grey slate roofs with multiple gables
- Tall traditional brick chimneys
- Large timber-framed windows with stone surrounds and mullions
- Modern glass conservatory extension with black powder-coated metal frame connecting two wings
- Stone paving (buff/natural color) surrounding the building
- Outdoor furniture and potted plants on the terrace
- Manicured green lawn extending into rural countryside
Photography style: High-resolution aerial three-quarter view, soft natural daylight from partly cloudy sky, ultra-realistic architectural photography with sharp focus throughout, natural color palette, professional property photography aesthetic.
Match the exact camera angle, building proportions, and spatial relationships shown in the source image.
Note: Gemini works best when you’re explicit about wanting it to reference your source image’s composition. You may need to adjust the prompt based on results - if certain details aren’t coming through clearly, try emphasizing them or breaking the prompt into a shorter version.
Here’s a timely video by @eric-s showing show very interactive AI rendering: Easy Render Edits with Generative AI in Photoshop - YouTube
It looks like Photoshop keeps prompts with the layer it was applied to… handy for sure.
I feel that horizon line should not be visible with house in this perspective.
AI has made the planet somewhat smaller…
That’s not an unreasonable observation, but I personally live in a hilly area, and as such, a scenario as shown in the image is not an improbable outcome. Perhaps the Prompt should reference any horizon requirements?
Yes - I was thinking about this as well. In my opinion you would see the horizon from that perspective.
The top surfaces of the chimneys are visible, and they are above the horizon “line”. The horizon should be somewhere above the chimneys.
Good point!
I live at one of the highest points in Bristol, so I can see the tops of lots of chimneys — but as you say, none of them are above the horizon
Any thoughts on how well the image generators work with a SU model that already has the desired textures in place? I always model everything with the final render in mind so I add textures, materials, and lighting as I go. Would it be better to start with that or a monochrome image with detailed prompts? This is typical of what I show my clients before rendering:
Maybe try it out and see if it’s to your liking?
I’ve been experimenting like crazy with AI image generators this week, and my conclusion is this: while they can produce amazing images incredibly easily, I still can’t get them to deliver the exact rendering I need. So for the foreseeable future, I expect to keep subcontracting my rendering work to a fellow ■■■■ sapiens.
Did the forum automatically censor a genus 
It probably censored a part of a Latin scientific name for a large ape species. It has an odd selection of forbidden words. It seems to allow the f-word, but, for instance, d a m n that some Victorians thought mildly annoying gets censored.
Ultimately the textures themselves are abstract and only meaningful to a human to a certain extent, the same applies to an AI.
The textures you are showing are excluding various aspects of color and light that make them look realistic in the first place, so ultimately they will change.
The image alone isn’t enough as it doesn’t really describe things enough - It’s actually an interesting topic for design communication, because it actually suggests that many images like that leave too much to the imagination - there is a fine line when presenting visuals to clients between things looking as you would like them to look and clients halluncinating a reality that will never come from an image that is too conceptual.