Refinement via Regeneration: Enlarging Modification Space Boosts Image Refinement in Unified Multimodal Models
For practitioners of text-to-image generation using unified multimodal models, this work provides a more effective refinement method that outperforms existing editing-based approaches.
The paper proposes Refinement via Regeneration (RvR), a framework that reformulates image refinement as conditional regeneration instead of editing, achieving significant improvements on Geneval (0.78→0.91), DPGBench (84.02→87.21), and UniGenBench++ (61.53→77.41).
Unified multimodal models (UMMs) integrate visual understanding and generation within a single framework. For text-to-image (T2I) tasks, this unified capability allows UMMs to refine outputs after their initial generation, potentially extending the performance upper bound. Current UMM-based refinement methods primarily follow a refinement-via-editing (RvE) paradigm, where UMMs produce editing instructions to modify misaligned regions while preserving aligned content. However, editing instructions often describe prompt-image misalignment only coarsely, leading to incomplete refinement. Moreover, pixel-level preservation, though necessary for editing, unnecessarily restricts the effective modification space for refinement. To address these limitations, we propose Refinement via Regeneration (RvR), a novel framework that reformulates refinement as conditional image regeneration rather than editing. Instead of relying on editing instructions and enforcing strict content preservation, RvR regenerates images conditioned on the target prompt and the semantic tokens of the initial image, enabling more complete semantic alignment with a larger modification space. Extensive experiments demonstrate the effectiveness of RvR, improving Geneval from 0.78 to 0.91, DPGBench from 84.02 to 87.21, and UniGenBench++ from 61.53 to 77.41.