I have a long-term project of taking a hand-painted rendering done in the 1930’s, which was only ever an image, and making a bas-relief sculpture of it. I have tried several online free AI image-to-3D websites because Dammit Jim, I’m an engineer and not a sculptor, without success.
The (probably simplistic) AI algorithms fail because the main figure is a Greek mythological god flying through clouds, with primary lighting from above but bounce lighting from beneath. This dual-source lighting is what I think fools the AI.
Years ago I spent several months trying to sculpt in SolidWorks but didn’t get too far because, well, see above… The image is mid-1930’s so I do want to render it in an Art Deco/Streamline style.
Do any of you have suggestions of what generator I might try? thanks.
Thank you, @drakko . I did try reliefScuptureMakerV2, but that was also fooled by the dual lighting. When I get home I should list the tools I’ve tried so far.
One of the online tools does have a Gimp-style curves control, which I did use. It’s a more fundamental question (and one which I may need to have a tool that lets me tell it “compensate for bounce lighting from below”…)
If you stand on a snow surface and extend your arm, your arm, chin, cheeks and nose will all be lit by the sun from above but also by the bounce light off the snow below. This aspect of the illustration causes the cheeks, nose, chin, and outstretched arm as well as extended wings of the figure, to have a ridge along the upper edge where it should be receding.
@MZip, thank you, I am afraid that is what I might need to do. Brush off my long-dormant Gimp experience of masking and gradient-painting along curved contours to try to give a graded shadow where there is now a graded highlight. Having said that, I might be able to mask the problem areas and do a grayscale invert to make lighter, darker and then adjust the curve of those areas to match their transition into the adjacent surface at their merge…
If your graphics card and your PC are even less powerful, check out the link - you’ll find some online generators there, though they’re a little hard to spot. Sometimes unavailable; alternatives are sometimes listed in the text / discord.
You need to make sure you optimize the image for the AI - for example, by removing the background, etc. Another option might be to convert it to grayscale. This advice applies to all AIs, whether local or online.
Also, you should get this Blender tool. You don’t need to understand Blender, but you do need to use this one feature: Make your model “manifold” / fix model - which means closing all the holes - even the ones you can’t see.
Have a look at Hueforge. The latest release has added highly improved features and capabilities for converting 2D to 3D images. I’m using it personally for some time now, and I can assure you that it definitely delivers astonishing results.
Here’s an alternative suggestion: If you want, you could upload the photo - unless you don’t want anyone to see it. Then we can take a look at how to optimize it so it’s suitable for the AI.
@RetroSharky , thank you for the extensive reply. My PC is minimal–a N100 Nucbox with onboard video. The image is B/W (color wasn’t even invented until almost 1940, I think it was driven by the need of Wizard of Oz… ). Source image is a photo of a somewhat-curled print of a photo of the original artwork (has crop marks and all) and I have never been able to trace down the original artwork. I have already done de-warping and gray levels improvement 10 years ago (before a long sabbatical) and cannot find even the source photo online these days–I think it was back in the Usenet days I got it… When I get home from work I’ll upload a copy. It is nothing remotely secret, but a project and quest I have taken on for myself…
and here are the artifacts (of the grayscale scan with the dual lighting) on the top edge of the arms etc and nose/face when running through an image-to-3D program
The programs I’ve tried so far are “image to relief” by blair johnson (Hugging Face Space) and Hi3D, to which I needed to buy points to even get test renders. I have downloaded the portable version of Hunyuan but not tried it yet. I only have access to this forum from here at work, but not access to makerworld. At home I do not have access to this forum (at least without registering a new username) but do have access to makerworld… It is a somewhat fractured online existence these days for me…
[edit] I just noticed how to add an alternate email to my account… [/edit]
I don’t understand the “portable” aspect of this. It looks like a whole Python distribution with many scripts and supporting files. “Portable” Windows programs in my experience are self-contained executable files… I will likely try resuming my own modeling, possibly taking the old SolidWorks output and tweaking it in Blender (which is a whole new learning curve/rabbit-hole for me! )
I think that’s the idea in this case. In python there’s all sorts of libraries to handle special tasks of all sorts. If you don’t have Python installed then you likely will be missing the common libraries as well as the less common. Supplying everything makes it where you can still run Python programs but don’t have to install or maintain an installation. Plus versioning doesn’t matter since it’s all there.
MZip has already explained it perfectly. If you download the non-portable version, you’ll see the difference. You then have to install the correct version of PyTorch and be careful not to install the wrong one - it’s usually not the latest version, but a very specific one.
Next, download from GitHub and use the command line to specify the type of project you want to run. I’m deliberately not using the word ‘install’ here, because for now, you’re just copying it to your computer.
To make matters worse, these AI systems also require the correct graphics card drivers - and if you specify the wrong ones, you’ll have to hunt down the exact right ones.
If you choose the non-portable version, believe me, it can sometimes become a full-time job for a day.
The advantage of the portable versions is that you don’t have to worry about a thing, and everything happens automatically - most of the time, that is, when everything goes smoothly.
What you know as portable version is often similar, but you don’t notice it because they just pack it into a single file. All you see is some .exe file. The downside, though, is that it usually only runs on a specific system.
@MZip , @RetroSharky , thank you for your replies. I will need to give up on Hunyuan for a different reason–I had overlooked its need for a GPU, which my PC (N150 Nucbox) does not have. I do have a Python3 installation already, but am aware at the multitude of libraries that different scripts require. I think that my most solid course of action is picking up with my attempt of modeling from scratch. Thanks again.
). Source image is a photo of a somewhat-curled print of a photo of the original artwork (has crop marks and all) and I have never been able to trace down the original artwork. I have already done de-warping and gray levels improvement 10 years ago (before a long sabbatical) and cannot find even the source photo online these days–I think it was back in the Usenet days I got it… When I get home from work I’ll upload a copy. It is nothing remotely secret, but a project and quest I have taken on for myself… 
[/edit]