formZ and Vray an upscaler of sorts....

[johnalexander1571]

Hi,

I verified yesterday that if you simply load an image as the background in VRay, set the image size you want (large) and render with no objects, it will give you a larger version of the image. I am finding this useful right now because the new AI enlargers (like Magnific.ai) add too much detail for my taste so I like to take the upscaled render and overlay it in photoshop and set the opacity to about 50 percent. I have found this adds a whole layer of detail not even possible with VRay. It adds wear and tear and makes rendered people look like photos. It isn't perfect (but damn near miraculous compared to what we had), but with some photoshop ingenuity I have found something I will likely apply to all renders going forward. Magnific.ai is 40 bucks a month right now, but I see this upscale while allowing AI to conjure up details over an image coming to Midjourney and Photoshop someday. They can't let this capability be only in the hands of other image apps.

[johnalexander1571]

Essentially, I am now using formZ and Vray to tell AI exactly what to generate with my regular old render. Then I go back and overlay the AI version over the render and then tune the opacity.

[Jaakko]

Hi!

That's interesting. Can you show some before and after examples?

[johnalexander1571]

Hey Jaakko,

I hope you can see this example well. Maybe I can put the files up somewhere if you can't. The top is my VRay render, then the bottom is the magnific.ai version. Now, I did overlay the AI version over the original and erase and details I didn't like. It didn't take too long to do this.

You can see how this upgraded the renderpeople in the scene to photo realistic and added detail to everything in the scene. It even appears to add some additional effects to the lighting. This will be coming soon to photoshop itself I am willing to bet.

I scaled from 2199 (the apps current start limit) to 4392 wide. Then I did one render at 4392 to get a nice sharp render to erase over.

I am guessing now the render larger in Vray trick is probably only as good as a PS upscale is now. I was just using that for photos. This really works nice with Midjourney images, which is how I found this.

 

[johnalexander1571]

https://www.artstation.com/artwork/BXKVxl

Artstation still doesn't let me get a full size post, but you can see the final here better I hope.

[johnalexander1571]

While this may not look like much at first, it is a big advance, I think. I was also able to use upscale on a midjourney image I asked for of corrugated metal. I was able to apply that 1 to 1 to the walls and ceiling, so they appear in the original render. It really replaced tons of lighting and texture work, because it addresses the whole image. It doesn't allow the AI mouse to run off with the cheese too far and do it's own thing. The new image fits right over the render, so it is a real way to harness AI for image finishing. I mean, I really don't do people in images because they suck so bad. But now, I think this gets people into renders. It will generate a new person every time, but that is part of the fun. It does have it's out of control moments at times, but I can truly use this as it brings realism to the finals.

Edited December 14, 2023 by johnalexander1571

[ZTEK]

Hey John, that's super interesting. Thanks for sharing the process. Can you also show some background images before rendering to help us better understand the process and the changes?

I recently used a scaling app for the first time called Upscayl (www.upscayl.org) to try to enhance some low-resolution, low-quality photos of a past project for my portfolio. I liked the result, which took away some realism, which I don't mind.

[johnalexander1571]

Hi ZTEK,

An example is the supervisor wanted a realistic background for the sound curtains (we sell). I snapped up this file from Turbosquid. I spent about half an hour just setting up colors. So just dragging materials out onto the cad file. Then I dropped it into the scene and let the render and the AI finisher do the real work. I think it looks convincing, but this is all it is, a 3d model with colors assigned.

You do need a 3D model for everything in your scene, this way you can ensure that the result is repeatable to the degree it is actually useful.

Yes, I have thought of this as a way to rescue some old work, and documentation of it as well.

 

[johnalexander1571]

I foresee Photoshop having a magic detail brush that you wave over your render or blurry photo and it adds this detail. It's possible now with what's out there. Just needs to be packaged into a tool.

[johnalexander1571]

Shaded work before Vray render.

[Jaakko]

John!

Thank you very much. This is interesting. Probably something I'll test later in a suitable project.

[johnalexander1571]

Hi, they let you do a few tests for free, so you can see if it's for you before you commit for a month. I am really expecting this to come out in something we already pay for like Photoshop so it won't carry it's own price tag.