But this doesn’t look like a finished scene that’s been “enhanced,” but rather like a very unfinished Blender scene that was then completed by AI.
For me, as a hobbyist Blender user, that would be pointless, because the fun of creating is precisely what makes it enjoyable. Otherwise, I might as well just let the AI generate the scenes/images.
It’s nothing new though. There are many add-ons like that. Probably hundreds at this point. 
There was one(at least one, I don’t know maybe even a few of them) to integrate ComfyUI nodes into Blender even.
I’ve tried that. Came to a conclusion it’s useless for me for archviz. It adds random stuff, incorrect stuff, changes things in wrong ways, lowers resolution and also requires more time. More time to reach worse results. I hate it. It’s not consistent either. You can never know if it’s going to even give you usable results. That’s kind of important when planning your work…
Yes. Some clients do. Also if I don’t care about that, why bother?.. If the details aren’t important, they aren’t important to me. 
OK, if we are talking about communication, what does it communicate? Well, AI look communicates “low effort”, “cheap”. I don’t want that perception or those emotions about my work even if it’s not really art that I am making. By the way, AI is terrible at composition. It doesn’t really make it right. It makes it worse. It seems to work when you don’t know what you want, or what you are doing, if you do, then it’s another story…
I have successfully used AI in my work once as far. I needed a wooden sculpture. I had examples of the sculptures and needed to show a different one that the artist will be hired to actually make(Oh, I hope the visualizations with AI did not offend them too much… 
). Anyway… No add-on was needed. I generated it in a part of the render, upscaled, then photoshoped it into the full resolution render so it doesn’t ruin the detail. No add-on needed
Aside from using blender “as a hobby”.. i just now see: it’s not only changing the lighting but also adding some plants and chair and towel..
If one would be satiesfied with the look but then adds some detail to the house.. maybe some different window or another one or a bigger garage or whatever.. so how does the next “enhanced” image looks like.. and maybe one does not like the different lighting, plants, chairs ??
So one has to find out if you are in control or or the machine.. this means time saved for “finalizing” the first image but then tweaking some prompt to get similar result (also how does this work in the next generation of ai/llm/addon ?).
When re-searching the parameters of the renderer (even if not using cycles or eevee) to achieve ones goal might be more “time saving” ?
Of course one can spent time to discover this.. but then doesn’t this contradicts the “time saving” argument ??
I think with this one it didn’t help at all with composition. It wasn’t good to begin with, AI did not help. I would not consider it usable. I would look for better angle.
That is a fair point, and this is exactly the balance we are trying to explore.
As someone who has worked with Blender and architectural visualization for years, I can say that the enjoyable part of 3D work changes with experience. Creating details, materials, vegetation and background assets can be satisfying at first, but when you need to produce several complete scenes within one week for a client presentation, a large part of the process becomes repetitive production work rather than creative exploration.
The goal is not to replace the design process or let AI invent the entire image. It is to reduce the time spent on repetitive modelling, texturing, asset placement and rendering while preserving the geometry, composition and architectural intent created in Blender.
The most important challenge is finding the right threshold: the minimum amount of 3D input required to maintain control, combined with the lowest possible risk of AI hallucination.
Too little 3D input, and the AI starts redesigning the project. Too much 3D work, and the time-saving benefit disappears.
Finding that sweet spot is essentially the core idea behind the tool.
The point was not to make finished archviz, it was to test the addon )) …
hey, thanks for a comment. Check my reply to “mali_zweipunktnull” bellow please. it is basically talking about this ))
That is a completely valid criticism, and honestly, most current AI workflows deserve it.
If the tool adds random objects, changes the architecture, destroys resolution and makes the result less predictable, then it is not saving time. It is simply replacing a controlled workflow with an unreliable one. For professional archviz, that is obviously a bad trade.
I also agree that details can be important. The “grain of rice” analogy was not meant to suggest that accuracy does not matter. The real question is which details carry design intent and which details are repetitive production work.
A window position, facade proportion or custom furniture piece may be critical. The exact arrangement of every grass blade, background leaf or small surface imperfection usually is not. A useful AI workflow must understand that distinction, or at least give the artist enough control to define it.
And yes, the recognizable “AI look” is a serious problem. If the final image feels generic, inconsistent or cheap, then the workflow has failed, regardless of how quickly it was produced.
What you described with the wooden sculpture is actually a good example of AI being used correctly: locally, for a specific purpose, with the result manually integrated into a controlled render. The fact that you did not need an add-on is fair.
The question for us is whether that same targeted workflow can be made faster and more convenient directly inside Blender, especially through masks, material IDs, Cryptomatte, object selection and high-resolution local editing.
So I do not think the goal should be “apply AI to every render.” It should be: use AI only where it gives a measurable production advantage, while keeping the artist in control of geometry, composition, resolution and final quality.
If it cannot do that reliably, then I agree with you: it is useless for professional archviz.
My point usually is. And that’s the reason this doesn’t work
For some reason whenever it’s about AI it is always about testing and proof of concept with some small details left to be ironed out. how much testing do we need? It’s been years of testing at this point… And it always turns out those small details cannot be solved.
Not only masks. Depth, normal passes also whatever it is that they call canny
There seem to be way more control options than you are aware of. ComfyUI has them all and it’s possible to use its nodes in Blender through those add-ons… Still no joy. Try them. I don’t think you are the first one to think of this. A lot of stuff exists.
And the answer is no.
It may work for some things… In some very limited circumstances. This is not new at this point…
A couple of people have made this point already but I wanna re-iterate :
Gen AI comes out with a single layer ‘Final’ image. Everything this AI decided to add into or enhance the image with is baked in. This is nothing like a composite, like photoshop or something, where edits are part of layers. Neither is it adding objects into the scene. I would be left with no control over the edits made by AI, I have to re-prompt for the smallest adjustments.
‘No control’ is an exaggeration, I know, but the amount of agency you loose is very problematic to many workflows. Having no .blend file or .psd file to quickly make adjustments to is the loss of control I am referring to
Gen AI also tends to struggle with consistency. I’m actually curious how this add-on will handle rendering the same yard from multiple angles. Will it keep everything looking similar? Will it be the same flower bed..or the same paintings? Maybe because the different angles were requested together, it might hold the memory to keep things consistent, but what if a 5th angle was prompted 1 week apart and the result you get is completely different from the first 4…how do you fix?
Okay, Architecturally, these details don’t matter, I am nitpicking.
But the ceiling has changed…That’s a major change! It did not follow the geometric intent you gave it, and because the inner workings of AI is a blackbox, how do you fix and diagnose without re-prompting? I am not picking out the error to say “Haha, AI made a mistake, hence its useless”
I am more worried about how you would fix it, without going into a cycle of edit prompt > regenerate > repeat.
Exactly. I don’t know how other people work, but I can give an example of how I make interior visualizations. I create the models of the space I am tasked to visualize first, all the exact products, finishes, I light it, find nice camera views and render as many images as it makes sense to actually show everything the designer worked hard to create. The idea is to show the design decisions and the hard part is modelling and texturing so there is no limit to how many images my clients get - as many as it makes sense to show everything nicely, right? So let’s say I visualize a big fancy living room joined with the kitchen and a dinning area. There can be 25 - 30 images there if its really fancy! Now what if the designer decides “hey I want darker floor”? OK, no problem, I change one node in my compositing file, hit F12 - done. 5 minutes to do including writing the email to tell the client that changes are done. Now what if I was using AI?.. Would anybody in their right mind choose that in my place? Even if I need to actually re-render, I will just make the changes, hit render and all the 30 image will magically appear where they need to be the next day. Even if we ignore all the numerous other huge deal breakers, AI doesn’t scale. With AI you need to “work” with every single image individually again and again and again if there are changes. And there are changes. That’s the purpose of visualization - to decide before building something. Changes are the whole point of it. @gugusheep, do you think it would be a good idea to use AI in the situation I describe?
I kind of want to try a fried banana with rice now… maybe some syrup or jam as well? And ice cream. There would need to be some ice cream with that.
Yes, yes, and yes.
I keep reading “AI generates all the stuff you don’t want to deal with” - how the hell does it KNOW what I considered to be something of value? Does it see some smudge marks that I intentionally painted on a wall, and decide the wall should be more even and clean? Or a shadow that I’ve intentionally clipped for artistic purposes, what happens to that? And as Martin said - suddenly some area has just changed in color, outside of the design palette, because AI thought the lighting looked wrong.
All of this is a craft. We make intentional decisions. Yes, it can mean a lot of manual labor that we might wish to avoid… but AI doesn’t just remove the manual labor, and replace it with “what we would have liked”. It just vomits out some result, and the Comfy chatbot hopes you just go with it.
This doesn’t save time, because any time you “saved” by handing it over to Comfy is now going to be time spent studying the AI result, finding what it did wrong, then unfucking it.
AI may help in the hand of experts to discover “new ways” because they can judge the “results”.. but at least in the software and “image producing industry” i think, it’s a weird thing that those who “praise” AI the most are the “AI service provider” and the ones who “produce content” all the time (which is something different than “knowledge” and often anything else than “important” .. if you know what i mean).. ..and.. (short sighted ?) “financial managers”..
“Old knowledge” can be lost..
The engineers did not made the former decision to fire people and “embrace” AI.. they might have been able to use it for some things.. but “AI results/hallucination” seems to be higher valued than expert judgement.
But.. we might get off-topic now.. 
It doesn’t know other things as well. It cannot deal with textures. It simply doesn’t know. It’s not in its training data. Training data does not exist for that. Images don’t exist online, so it cannot possibly know, how the finishes look. I often get a sample like this:
And I have to somehow find some pictures of it and then find some similar textures of wood an photoshop all the other properties I need in it and make something like this:
AI cannot do this… The finishes it makes are terrible. It knows 5 kinds of wood at best, while for example at this very moment I am working on a project where I have 2 different finishes:
They are different. I need to show them as different. AI doesn’t know the difference and it cannot know how they look when the texture needs to be big. Actually I don’t know either. …but I guess better. …usually …I think…You could probably convince it to make something similar in a few hours. It’s not going to be even close to what I do in 15 minutes and it’s going to be quarter of the resolution, but… Well, I don’t have a few hours for every single image out of the 30 for every material anyway and it cannot keep them consistent across those images anyway. It’s like asking a 4 year old to do my job for me. It’s ridiculous, it fails at everything at every step. Why am I talking about this at all?.. We wouldn’t be discussing the idea to let a 4 year old do this job. Maybe I am somehow hoping to stop @gugusheep from wasting time… It’s better to learn to do it properly.
It seems like it could. It turns out it’s pretty useless once you try it. AI doesn’t understand different materials. It can make a few well and when you really want to experiment, its knowledge of materials ends. Everything looks the same AI slop kind of way. Even the biggest image generating models just aren’t trained on enough data. Because such data doesn’t exist. As artists we can use our experience of the actual reality and can identify and recreate very specific properties of materials in very specific ways and AI image generators don’t come even close to those abilities and speed. It seems as if it could be extremely useful if looked at from a beginner perspective who doesn’t have the skills to make even simple materials, then since AI manages to do that it seems as if it is some sort of a solution.
If you are interested in trying it for yourself, you can. Check out ComfyUI. The add-on the OP is talking about is nothing compared to that in terms of functionality. And it even comes as an installable app nowadays and you don’t even need to deal with complex setup. Rendering whatever passes you need and using them in ComfyUI is not a problem really. You can also use whatever image generating models and APIs of huge paid models in it. You will see what the actual problems are. It’s not like there is a lack of advanced tools for using AI… it seems gugusheep is just a little late to the game. You can control stuff. It just sucks. At best you will get AI look that screams “low effort” and “cheap” anyway. Even though using ComfyUI might not be so low effort in reality… It might be quite a lot more effort actually.
Adding to what @MartinZ just posted while i was typing..
What if you found something interesting on that “exploration” ?? It’s all “halucinated” so you do have not the added objects nor marterialss or lighting setup..
..so you have not only to replicate a look from a real image or the reality itslef, which might be hard enough, but from something which might even add some “unnatural” impossible lighting.. well sometimes this is also done for presentational purposes, but nayway:
One needs to know how to do this anyway.
So i do not see the “improvement”, speed up, time save or whatever one want to call it.
Also: any profesional might “see” that this is “halucinated”.. maybe also asking for " a simliar render with slighly changed context.. and then ??
So maybe “only” for exploration “in house” but i think never for:
.. but that’s only my one and half cents
I wouldn’t overlook the cheap and and low effort feel about AI as well. I think it’s quite strong in people’s minds at this point. Even if I worked hard on the whole project and then added just a few details with AI, few accessories, few things that don’t really matter… I think they actually do matter in a big way, I think it doesn’t improve much, it just damages reputation at this point and leaves the viewer guessing if everything they see is just cheap mindless AI. In my mind, that’s a quick way to ruin an image and the whole visual communication. If I am looking at an image myself and see AI elements in it, it just instantly becomes all about judging the creator and their laziness and guessing what else in the image may or may not be AI. I am too afraid of that at this point to use more AI in my own work. It’s instant “slopify and cheapen filter” in my view. I’ve used AI before to see how it could work and just never showed the results to my clients, because it just makes my work less, it doesn’t add much, it takes things away that you cannot easily get back later… and I am not proud of the times I did already either… Actually, I am bit ashamed. 
It’s weird, because, usually I don’t get ashamed easily… I’ll cheat any way I can to get results without much afterthought. There must be something about this emotional intuition…
So it’s really a pity, because I could clearly improve my renders by for example adding extremely attractive people to them this way making the images without a doubt a whole lot more aesthetically pleasing:
…but for me this whole idea remains just this - a joke…
Where can I get that addon?
That’s a sharp dressed man.. all fitting colors or is this somekind of camouflage ?
Now how is the living room and.. do you change when moving their ? 
Yeah, that’s the thing with generative AI: you can tell immediately by seeing such perfect bodies and too much hansome faces
