Matts AI for Architecture

05/14/2025

Google Earth to Photo 2: "Aerial Europe Daytime"

I’ve recorded the second tutorial in this series. Editing tomorrow, online Wednesday Night or Thursday. This one focuses on Europe and Scandinavia from above: classic architecture, winding streets, and terracotta rooftops instead of skyscrapers. Typically the shiny modern museum would be your 3D rendering so you won't have to convert it like I tried to do here.

My Europe AI model is designed to capture an old-world feel, focusing on unique historical buildings and detailed landscapes.

This process isn’t automated. It still needs a human touch, like adding details in post. If you want me to handle it, I offer a conversion service. Send me your rendering or context image, and I’ll get back to you with a quote and turnaround time. Prices generally range from $250 to $600 per image.

10/19/2024

I’ve developed a new Upscaler extension for Forge and Automatic1111 called Regional Prompt Upscaler. It works similarly to Ultimate Upscaler but takes things further by breaking your image into tiles and using AI to generate unique prompts for each region. It also uses an overall image prompt to make the tile descriptions even more accurate.

Key Features:

- Automatically resizes tiles for perfect image coverage
- Uses CLIP and BLIP to generate prompts based on content (Pro version includes 3 additional models for even more accuracy)
- Lets you filter out unwanted words or categories
- Exports and loads prompt grid for easy customization
- Overlap and feathering options for seamless blending

If you're looking for a fresh way to upscale your images, give it a try and let me know what you think.

https://www.hallett-ai.com/upscaler

10/02/2024

My tutorial is ready to turn Google Earth into Night. I can keep working on this, but is it ready for marketing? Comments please!

10/01/2024

I've always wanted to learn to code, and now with AI I was able to make a simple program that's actual useful. Its made with ChatGPT, and I'm hosting it on a Github like a real developer.

Its a GUI that uses CLIP to sort your photos with tag words of your choice. "Nature, City" etc.

I made sure it was easy to install and use, especially for typical Window users.

Check it out!
https://github.com/HallettVisual/Photo_Sorter

09/25/2024

I've made improvement to my Day to Light Stable Diffusion Model and Workflow, and I think its ready for sale.

Its even possible to adjust the time of day from sunet to night and works on any large city with skyscrapers.

In this example I converted a Google earth image of London's Gherkin into a realistic night image.

I will post more examples this week! Please let me know what you think, and feel free to message me if you would like access to advanced copy.

09/17/2024

I've been working a set of renderings for a few days straight, and just finished a set, but I have one issue thats bothering me - The image in this post.

Normally I use Stable Diffusion just to enhance my high-res rendering finals, but this patio sunset rendering is upscaled from 1500 pixels to 7000 and all the textures have been replaced.

It was out of necessity, the final over night high-res version didn't save properly this morning, but I had a low-res draft. Since I was going to enhance it with Stable Diffusion or Flux anyways, I decided not to re-render and apply all the upscale and enhancing tricks I know.

Surprisingly, I feel like I did something wrong, or was dishonest. Its irrational because I'm pretty sure its a better product than I could have done traditionally given the amount of time.

On a side note: Flux is amazing at people and straight lines, but s**t at fine grain details. Trees look like cartoon versions. A mixture of both here allowed me to pull this off in 3 hours of messing around. I had zero feedback other than the one draft, and that I should reduce the lake a bit.

09/13/2024

When was the last time you visited Huggingface? There are so many cool "spaces" as they call it were you can test all sorts AI applications for free. Flux, Image 2 Mesh, LLMs, Background removal, you name it.

List of spaces using ZERO-GPU

09/13/2024

Here's my latest attempt to turn Google Earth day into night.

09/11/2024

Does this look accurate for a night image? I had a comment that my previous attempts looked a little to much like an HDR blend.

09/07/2024

Version 2 of my "Aerial City at Night" SDXL model is ready, and I'm still developing a universal ComfyUI workflow to easy convert Google Earth images into photo realistic night.

This is only a 2x upscale, and single pass conversion. Eventually we'll get this to 4x with sharp details and more lights on the buildings.

I have made a similar model for Flux but it doesn't work at all for image to image, only text to image.

I still don't know how to market this LoRA model. My current thought is to make the workflow available for free so you can test it, and realize the only way it works is by purchasing the model. It dopes all the heavy lifting.

09/06/2024

The Stable Diffusion by Abdullah plugin for Photoshop is dying and no one is updating it. Updating Forge and Automatic1111 often equals connection issues, so we need an alternative.

Check out Krita. I just got FLUX and SDXL working using a plugin that has a local ComfyUI backend.

Krita is open source photo editor.

https://krita.org/en/

https://github.com/Acly/krita-ai-diffusion/releases

08/25/2024

I made an AI WWII movie trailer this weekend. Various techniques, see details in description. This version 2, I made another version last year. This one uses mostly Runway.

A faux WWII movie trailer made with Images Generated with Stable Diffusion, Flux, Trained Flux LoRA on 100 images from WWII movies for Mood and Era Specific ...