Player not loading? Watch on YouTube
The original video is in Russian. The installation section explains an existing working setup. Sonya demonstrates a Windows workflow for installing Stable Diffusion and using it inside Photoshop to postprocess 3D renders. Her manual setup uses Python 3.10.6, Git and AUTOMATIC1111, with a Stable Diffusion 1.5 checkpoint. The supplied instructions recommend Photoshop 2024 or later; she reports that her installation works in Photoshop 2025. She also points to StabilityMatrix as an easier installation route, but says she has not tried it herself.
The integration uses Auto Photoshop SD plugin v1.4.1, the --api launch argument and WebUI extensions for the plugin and ControlNet. She selects AUTOMATIC1111 in the plugin settings and uses Realistic Vision V6.0 B1 for generation. Examples cover text-to-image and image-to-image, with lower strength keeping results closer to the original render.
This local AI workflow runs on her computer's GPU. She uses a GeForce RTX 3060 with 12 GB of VRAM and prefers 512-pixel generation because higher resolutions take too long for her workflow. That is her working choice, not a stated hardware minimum. She edits small regions to retain detail, then blends generated water and scratched metal through masks. Generation uses the visible Photoshop composite and covers the selection's bounding rectangle, so separate patches need care.
The finishing work combines ID masks and curves for clearer object boundaries, Topaz Photo AI for upscaling, and photobashing with color adjustments to match added sea and sky images to the render.
