Intrinsic-ControlNet : A Generative Rendering Approach to Render Any Real and Unreal
Keywords: generative rendering, diffusion model, photorealistic images
Abstract: Rendering highly realistic images from 3D assets is one of the most persistent challenges of the graphics community, which is procedurally conducted by simulating real-world geometry, material, and light transportation. However, such simulations are both burdensome and expensive. Recently, diffusion models have seen great success in realistic image generation by leveraging priors from large datasets of real-world images. Nonetheless, these generative models provide limited control over the output and, unlike graphic pipelines, cannot accurately integrate materials and geometric information for precise image synthesis. In this work, we propose a generative rendering framework, Intrinsic-ControlNet, that enables the generation of corresponding RGB images from 3D assets like a rendering engine by taking intrinsic images, e.g., material, normal, and structural information, as network inputs. We propose a novel multi-conditional control method that allows the model to accept any number of intrinsic images as input conditions. To mitigate bias from synthetic training data, we propose a new model architecture that allows appearance and structural conditions to be input separately into ControlNet, preserving the realism of appearance generation from real data while maintaining structural control capabilities from synthetic data. Experiments and user studies demonstrate that our method can generate controllable, highly realistic images based on the input intrinsic images.
Primary Area: generative models
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 6973
Loading