C’est pas bon, C’est bonbon, J’aime du chocolat! 
So, I did a thing. I’m working on an e-learning class that will address the whole rendering vs ai image thing. It’s based on my amazing lecture on the same topic
(just kidding. I mean, it’s based on it but it’s not amazing. it’s grandiose 
)
Anyway.
A small comparative of renders, same point of view, same model, different processes.
This is raw from sketchup. As you can see, I used Enscape proxies, I like how even if you don’t plan on making a proper render, they create good silhouettes.
sun and shadows, remove the lines, add a bit of AO. In an actual project, I would consider this as “sharing with my colleages” material.
This is after a quick trip in Enscape. The vegetation is inspired by the local flora, the village actually planted tulips on the sidewalk, and the hydrangea do bloom that massively… a keen botanist might say they’re not supposed to bloom at the same time, but I’m more of a plant murderer myself.
First, the same, but using the AI enhancing from Enscape.
At this size, you won’t see any major change. Their tool works on vegetation and people, making them more realistic, and yeah. there are subtle changes. In this case, I could have spared the 3-4 min.
now, Veras, based on the first enscape rendering. Initially, I was gonna check it, and I ended up making this whole model.
Gen 7 prompts are good, that’s why they are limited to a certain amount of monthly tokens. it added a bike, opened the main door and… c’mon, more plants upstairs ?
Kudos to the work on the inside, I haven’t dealt with it yet, but it made something correct-ish.
Final one on Veras, I asked for the people to walk around, if you forget the car door situation, it’s not bad. I find the behaviour of the woman upstairs quite wow. the way she touches the handrail while on the phone ? c’mon.
Nano Banana. just asked to make it more real. it went full throttle on the hydrangeas. People look good though. none of them looks like the enscape version.
GPT. sharper shadows, better lighting, I find GPT tends to be good at adding a layer of “grey dirt” on the ground. Off course, I had to explicitly tell it to respect all the geometry, the perspective, and only make the model look more realistic.
AI render from sketchup. Scroll back up, I used the SU version for this, so plants and people were blobs. I didn’t expect too much on that front.
Puzzled as to why the fiat 500 changed colour.
If you watched my tedtalk, or read my essays here, you’ll know I’m ambivalent toward ai. it’s a tool, and it can be a good tool to bridge the extra gap. and it’s also a massive source of slop that (already) lowers the overall level.
the more details your model / render has, the easier it is to coerce AI to respect your vision.
the AI tool that enhances vegetation and people in Enscape is a good idea, it lowers the need of a powerful machine and time to make a good image.
and since I’m a millenial hipster, I’ll keep on doing stuff “by hand” because it has more soul.
Funny how Gemini changed a dark woman into a white male. In the Gemini image all the people seem “white”.
gemini made everyone and everything whiter. the whole image is so luminous 
You are fast becoming the AI Rendering Guru. 
I really hope I don’t…
it’s just that unlike most people talking about it on social network, I’m taking a more cautious and conservative stance.
you won’t me publish stuff like “Woah, this new AI is changing the game, click here to see how gpt is killing rendering tools”
I’m an autistic frenchman, meaning my heightened power of pattern recognition and thirst for knowledge is coupled with a contrarian mind. and a strong accent. and the number of hills I’m willing to kill on is high.
If you break AI down and simplify it, its really just next gen software based on data learning, not written rules.
You’re just the guy willing to explain how it can and cant be used.
Also, I live near a Hill that has a local myth of a dragon slaying on top of it. Perhaps we could call on you to fight for us in case we get invaded by AI Dragons 
Out of curiosity, I gave 2 images to chat GPT.
the raw result from enscape, and the image enhanced using gpt. Then, I asked if it could tell me if image A (enscape) or B (GPT) had been AI-enhanced.
it maintained that the raw enscape image was clearly AI, while the GPT-enhanced one was made in a rendering tool, and maybe some post processing.
Then I told the truth and after a long 4-5 sec of computing…
morality ?
again, it’s a tool. don’t trust it more than you need to.
I trust my hammer to hammer nails. Eventually, also, to hold a stack of paper, break ice cubes and bend metal stuff into shape.
I don’t trust my hammer to do my accounting.
edit : one thing these LLM are teaching us is that if you sound confident enough, people will blindly follow you.
Yep, AI is a bare faced liar.
And when you call it out, it thanks you, then explains the lie away.
I found this out whilst researching new cars. I was fed up looking online at all the different trim levels and features on various makes and models in my price range so I asked Copilot to summarise into a nice table for me to compare.
And it totally messed up the features available on one model. When I pointed out it was wrong, it apologised and then blamed it’s mistake on an initial press release from 4 years ago.
But If I hadn’t already known it had made an error I could have easily taken it at face value and then wondered why my new car didn’t actually have an ejector seat.
This has been true forever… add a safety vest, clipboard and hard hat and you can get into places you likely shouldn’t be, especially if you act like you belong there.
The scary part - LLMs have the entire internet to build their model of the world from… and knowing that the internet is a bastion of truth, logic, scientific thought and honesty we are in good hands.
Right?
…
Right?