AI arch viz expert needed – trying to build a faster, cheaper alternative workflow for creating CGIs

I run an architectural visualisation studio producing high-end photorealistic CGI, primarily for workplace designers pitching to win projects. I’m exploring whether we can use AI to offer clients a faster, lower-cost alternative to traditional CGI — ideally achieving around 80% of the result for roughly 50% of the cost. Importantly, I’m not looking for someone to build a complicated automated workflow or ComfyUI system at this stage. For now, I want to work manually with someone who is very good at current AI image tools and test what’s actually possible. What I need help with I’d like to take a few real projects and experiment with different approaches. For example, we might start with a basic SketchUp model or clay render to establish the correct geometry and camera, then use AI manually to create the photorealistic result and make subsequent changes. I want to understand things like: * What’s the best starting point — CAD, SketchUp, clay renders, mood boards, reference images, etc.? * How well can we preserve the correct architecture and layout? * Can we introduce specific furniture, materials and finishes? * Can we make targeted changes without AI altering unrelated parts of the image? * What types of client feedback can we realistically accommodate? * What can’t we do reliably? * How consistent can we be across revisions or multiple images? * How much manual work is actually required? * Is it genuinely faster and cheaper than traditional CGI? The aim is to establish a simple, practical manual process that we could start offering to clients relatively quickly, along with clear rules about its limitations. Once we’ve proven that process works, we may look at scaling or automating parts of it using ComfyUI or similar tools. That’s a later stage and isn’t what this job is about. It’s also completely possible that our testing shows AI isn’t currently suitable or cost-effective enough for what we need. That’s a useful outcome too. Examples I’ve recorded a Loom explaining what we currently produce and what I’m trying to achieve: Loom: https://www.loom.com/share/3d7cccb42ce84f41ac6420ea2f8916d9 Here’s also an example Miro board from a real CGI project showing the kind of detailed client feedback we normally receive: Miro: https://miro.com/app/board/uXjVLaKDXtc=/?share_link_id=997036365817 I’d particularly like to understand which types of comments on that board could and couldn’t realistically be handled using an AI-based approach. Who I’m looking for I’m looking for someone who is hands-on and very capable with current AI image-generation/editing tools, particularly where maintaining composition, architecture and visual consistency matters. You don’t need to build us a system. I want you to work with us manually on a few test projects, try different approaches, explain the caveats, and help us establish what does and doesn’t work. If you’re interested, please watch the Loom and take a quick look at the Miro board, then tell me how you would approach our first test project and which AI tools you would use.