Professional Rendering Tips - Ep. 1 Masking and Multi-Phase Ai Enhancement

[Renerabbitt]

Tools:

• Chief Architect X17
• Photoshop or Photopea
• Krea as of 2026, their enhancement tab and specifically the Krea Legacy LLM

Goal: iterative output for controlling specific elements of the original design intent while enhancing using Ai in a efficient way
We will be skipping some basics such as FOV settings, angle of camera, and composition in this first episode and we are jumping straight to the hot-topic of Ai image enhancement
Start by Exporting an image to disk. This is Important as we need to maintain an exact base resolution. we will be exporting a second image for material based mask selection

Next, we will be changing all of the vector based colors of our materials in camera and switching to a vector based rendering. This is specifically for material mask selections in Photoshop

We will then use Photoshops function to change line thickness to 0

When the Vector mask ID image is placed underneath the source image in the Photoshop Layers panel and is the targeted or only displayed layer it makes it very easy to select a material base on color.
Here is an example, the selection lines and active color suggest that I used select by color on the purple material ID. Notice the walls have a hatched selection line indicating their selection

Here is the mask of the walls

This entire sequence up until now has taken 5 minutes
The purpose of the material ID mask is to mask out problem areas that an Ai output changes
So lets send our image to our Ai Enhancement software
I made a pass with a HIGH level of creativity Low Level of fidelity and one with the opposite settings. Sometimes You may want to mask in a feature from a very interpretive output.
Here is our base ai enhancement

Notice a sprinkler head was added, that the photos were changed, as well as the chopping block and a few other features such as the highly reflective white surface tricking the ai into creating floating LED's
The major reason for utilizing the Ai enhancement software at the top of this post is because I have recognized that this software DOES NOT change the pixel-to-pixel perspective and orientation of the source image. 
For example, notice that the backsplash and photos are in exactly the same spot when revealing background layers with a mask:

So for our first pass, in Photoshop, I will create a raster based mask layer on the Ai enhanced image. The Ai enhanced image will be higher in the order of the layers panel, sitting on top of our source render. 
I will mask with a brush, shown as pink in this example:

Giving us:

Next I will add some lens flare, and using the material ID mask, I am going to add a hue adjustment layer and adjust the color of the walls and cabinets. Also using the material ID mask I am going to change the exterior by masking out the windows to reveal a layer underneath which will be a landscaped yard
At this point I can easily change any finish colors I need to very quickly

For example, I introduce a repeating pattern for texture. A blend mode set to color burn, masked to the walls, to give the walls some texture

At this stage we have spent 15 minutes total, and have change our output significantly and set ourselves up to make new edits very quickly. Make sure to save your camera settings in Chief Architect in case a design element has changed, because you can drop that source image back into your layer stack
Let's send this to Ai for another round of enhancement
Our 2nd round did offer some improvements. There is a 2nd round of masking and then we are ready for a camera raw filter.

The camera raw filter can correct some common pitfalls of ai enhancement, such as heavy contrast, white balance, white levels black levels, texture, sharpening, noise and more
The great thing about camera raw is that we can save our settings and they can provide a base edit that is often times all you need...a single button press for major improvement to the Ai output

This series of edits all in all took 20 minutes, and although we would have gotten a similar output from ai in just a single click, this final output is controlled and does not depict anything that was not specified in the original design