I’ve used the AI render tool in the past with success (mostly), but lately it's been really bad. It keeps changing my design quite drastically, completely leaving out pieces or making different objects look identical when there are clear differences. Here's an example, screenshot from Shapr3D, and the AI render below. It made the middle and right planter boxes match the left one, even though they aren't the same thing in the model.
Here’s the prompt I gave it:
“These are cedar planter boxes. Do NOT change dimensions. They are all 48” x 96" x 12". Put these on grass but don’t show plants in them. Also note that these are 3 different designs, so they should not be identical, and DO NOT CHANGE MY DESIGNS. They should appear exactly as I created them in Shapr3D. AND BE CAREFUL WITH DIMENSIONS. Your last render made them look like they are 24" x 48", when in reality they are 48" x 96" x 12"".
The sizing is also a bit off, unless those blades of grass are 2" wide, but I could forgive that if it hadn’t changed two of the three objects in the render.
Am I doing something wrong, or is the AI render actually getting worse?
It’s AI, innit. It’s going to mess up, that’s just the nature of the beast.
If you really need to use it, I’d advise to keep the prompts as simple as possible. I’m not sure what model Shapr is using, but it seems to not do very well with natural language. Try more bullet point-style prompts, describing the environment and maybe lighting. I don’t think you should have to specify dimensions, it should have that from the actual model one would think. I assume all those 3 planters are in the same shapr file, layed out like your first image? It’s weird how it gets confused by that and makes all 3 the same version. Maybe it really isn’t very good at parsing natural language prompts.
HI, I faced the same problems. What I found from experience is that the Ai render is not really based on your 3D Model but an interpretation of what the Ai sees in the Visualization Window.
Usually it helps to change the perspective a bit, to get another result.
In the beginning I believed that the Ai is capable of reading the dimensions from your 3D Model and place it in a realistic manner in the rendered Environment. Turns out it just makes an interpretation of the visualization it sees. So it helps to find a viewing angle that shows ever important detail of your model. It will also help to just show one and give the prompt to copy the model 3 times. So you can show just on e model in the visualization window and the Ai “sees” more detail.
In the end it’s a nice tool for quick and “dirty” visualization. I usually need 3 or more tries to get a sufficient result. For that use case it’s really good and time saving, but if you need full control you will still need to make a render image.
Hi, what i usually do is make a first render in the app, save the photo, then upload the photo into chat gpt (for example), then tell chatgpt what the actual dimensions are and what else needs to be changed. You can also upload the stl or step file. The results then usually come out pretty close to what i had had in mind.
The prompt could be it, but I’ve used similar prompts with other designs, and it worked well. I’m thinking it’s not great when there are multiple objects in the same file. All my other uses of the AI had only one object.
