Heeding the Inner Voice: Aligning ControlNet Training via Intermediate Features Feedback

Despite significant progress in text-to-image diffusion models, achieving precise spatial control over generated outputs remains challenging. ControlNet addresses this by introducing an auxiliary conditioning module, while ControlNet++ further refines alignment through a cycle consistency loss applied only to the final denoising steps. However, this approach neglects intermediate generation stages, limiting its effectiveness. We propose InnerControl, a training strategy that enforces spatial consistency across all diffusion steps. Our method trains lightweight convolutional probes to reconstruct input control signals (e.g., edges, depth) from intermediate UNet features at every denoising step. These probes efficiently extract signals even from highly noisy latents, enabling pseudo ground truth controls for training. By minimizing the discrepancy between predicted and target conditions throughout the entire diffusion process, our alignment loss improves both control fidelity and generation quality. Combined with established techniques like ControlNet++, InnerControl achieves state-of-the-art performance across diverse conditioning methods (e.g., edges, depth).

A3D: Does Diffusion Dream about 3D Alignment?

We tackle the problem of text-driven 3D generation from a geometry alignment perspective. We aim at the generation of multiple objects which are consistent in terms of semantics and geometry. Recent methods based on Score Distillation have succeeded in distilling the knowledge from 2D diffusion models to high-quality objects represented by 3D neural radiance fields. These methods handle multiple text queries separately, and therefore, the resulting objects have a high variability in object pose and structure. However, in some applications such as geometry editing, it is desirable to obtain aligned objects. In order to achieve alignment, we propose to optimize the continuous trajectories between the aligned objects, by modeling a space of linear pairwise interpolations of the textual embeddings with a single NeRF representation. We demonstrate that similar objects, consisting of semantically corresponding parts, can be well aligned in 3D space without costly modifications to the generation process. We provide several practical scenarios including mesh editing and object hybridization that benefit from geometry alignment and experimentally demonstrate the efficiency of our method. https://voyleg.github.io/a3d/