Realiz3D: 3D Generation Made Photorealistic via Domain-Aware Learning

Ido Sobol, Kihyuk Sohn, Yoav Blum, Egor Zakharov, Max Bluvstein, Andrea Vedaldi, Or Litany

arXiv:2605.13852 · 2026-05-15 공개 · arXiv · PDF

diffusion-models domain-gap control-signal text-to-multiview residual-adapters diffusion-based-generators domain-aware-learning photorealistic-images

Abstract

We often aim to generate images that are both photorealistic and 3D-consistent, adhering to precise geometry, material, and viewpoint controls. Typically, this is achieved by fine-tuning an image generator, pre-trained on billions of real images, using renders of synthetic 3D assets, where annotations for control signals are available. While this approach can learn the desired controls, it often compromises the realism of the images due to domain gap between photographs and renders. We observe that this issue largely arises from the model learning an unintended association between the presence of control signals and the synthetic appearance of the images. To address this, we introduce Realiz3D, a lightweight framework for training diffusion models, that decouples controls and visual domain. The key idea is to explicitly learn visual domain, real or synthetic, separately from other control signals by introducing a co-variate that, fed into small residual adapters, shifts the domain. Then, the generator can be trained to gain controllability, without fitting to specific visual domain. In this way, the model can be guided to produce realistic images even when controls are applied. We enhance control transferability to the real domain by leveraging insights about roles of different layers and denoising steps in diffusion-based generators, informing new training and inference strategies that further mitigate the gap. We demonstrate the advantages of Realiz3D in tasks as text-to-multiview generation and texturing from 3D inputs, producing outputs that are 3D-consistent and photorealistic.

한국어 요약

📋 한 줄 요약

**[3D 생성 / 도메인 분리 학습]** 합성 렌더와 실제 사진 사이 도메인 갭을 명시적으로 분리해 제어 신호를 학습하면서도 사실적 출력을 유지하는 diffusion 학습 프레임워크 Realiz3D 제안.

🎯 핵심 기여도

💡 핵심 아이디어

"제어 가능성"과 "시각 도메인"을 동일 채널에서 학습하면 둘이 얽혀서 사실성을 잃지만, 도메인을 별도 covariate로 분리하면 같은 모델이 합성·실 도메인을 모두 학습한 뒤 제어 신호를 양쪽에 자유롭게 이식할 수 있다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: 합성 데이터가 풍부한 제어 학습 영역(3D, depth, normal, material)에서 사실성 손실 없이 제어를 학습하는 보편적 처방 제시. **한계**: 도메인 covariate가 양극화(real/synthetic)에 가까울수록 효과적이며, 미세한 sub-domain 분리에서의 효과는 검증 필요.

🚀 실용적 활용