FrameDiffuser: G-Buffer-Conditioned Diffusion for Neural Forward Frame Rendering
This addresses the need for efficient, real-time neural rendering in interactive applications like gaming, where existing methods lack temporal consistency or are too computationally expensive.
The paper tackles the problem of generating temporally consistent and photorealistic frames for interactive neural rendering by introducing FrameDiffuser, an autoregressive framework that conditions on G-buffer data and previous outputs, achieving stable generation over hundreds to thousands of frames with superior quality in lighting, shadows, and reflections compared to generalized methods.
Neural rendering for interactive applications requires translating geometric and material properties (G-buffer) to photorealistic images with realistic lighting on a frame-by-frame basis. While recent diffusion-based approaches show promise for G-buffer-conditioned image synthesis, they face critical limitations: single-image models like RGBX generate frames independently without temporal consistency, while video models like DiffusionRenderer are too computationally expensive for most consumer gaming sets ups and require complete sequences upfront, making them unsuitable for interactive applications where future frames depend on user input. We introduce FrameDiffuser, an autoregressive neural rendering framework that generates temporally consistent, photorealistic frames by conditioning on G-buffer data and the models own previous output. After an initial frame, FrameDiffuser operates purely on incoming G-buffer data, comprising geometry, materials, and surface properties, while using its previously generated frame for temporal guidance, maintaining stable, temporal consistent generation over hundreds to thousands of frames. Our dual-conditioning architecture combines ControlNet for structural guidance with ControlLoRA for temporal coherence. A three-stage training strategy enables stable autoregressive generation. We specialize our model to individual environments, prioritizing consistency and inference speed over broad generalization, demonstrating that environment-specific training achieves superior photorealistic quality with accurate lighting, shadows, and reflections compared to generalized approaches.