Currently, the AI only intervenes in the process once the image has been fully rendered by D5.
This can make it difficult for the AI to interpret the image correctly.
This is especially true for details that are farther away and therefore very small, yet still important to the scene.
I’ve posted plenty of examples of this in this forum in the past.
The AI in D5 has significant problems with this, as do other well-known AI systems.
However, D5 has the advantage that it could rely on the actual 3D data when calculating images using AI.
This should prevent errors in the interpretation of shapes from occurring.
Is work already underway on this approach?
Am I the first to have this idea?
I can’t imagine that.
Or is this fundamentally not possible?
I completely and absolutely agree with you. If D5’s AI could, for example, read the types of materials used, instead of simply distinguishing them as layers, it could open up the possibility for retouching that does not change the materials, or makes them more realistic by knowing exactly what it is working with.
You’ve hit the nail on the head: our current AI tools operate on a ‘rendering + AI’ model, whereas our ultimate goal is to evolve into a true ‘AI Renderer’—where the AI deeply understands scene geometry, material properties, and lighting behavior for intelligent, cohesive processing.
In fact, our internal R&D team is already actively exploring this direction. It involves a significant architectural overhaul of our existing rendering pipeline to enable deeper integration.
Given the complexity and depth of this transition, it is a long-term development objective rather than a short-term fix. While we aren’t there yet, your suggestions align perfectly with our long-term roadmap. We truly appreciate you recognizing the direction we’re heading in—it’s exactly this kind of feedback that drives our development forward.
We do have internal ideas to create a native AI renderer, but this plan has just begun.