Post #751150

After the DLSS 5 fiasco, I’ve been thinking about what it might be like to create a genAI rendering pipeline that gives artists more control, not less. I can imagine using meshes that are keyed to have the concept art team draw assets to fill these areas. Similar to how sketch-to-AI tools work, but with an artist controlling the sample data the models are built from. Low-poly meshes and vectors for velocity provide input to a custom model created by artists, producing truly wild visuals. I feel like DLSS5 could have been really cool and they could have introduced a whole new rendering pipeline, but they leaned so hard into AI slop instead.