ControlThinker: Unveiling Latent Semantics for Controllable Image Generation through Visual Reasoning

TL;DR

ControlThinker introduces a visual reasoning-based framework that enriches text prompts with latent semantics from control images, improving semantic consistency and visual quality in controllable image generation.

Contribution

It presents a novel comprehend-then-generate paradigm leveraging visual reasoning and a metric-based reward to bridge the semantic gap in controllable image synthesis.

Findings

Enhanced semantic consistency across benchmarks

Improved visual quality of generated images

Effective mitigation of semantic gap issues

Abstract

The field of controllable image generation has seen significant advancements, with various architectures improving generation layout consistency with control signals. However, contemporary methods still face challenges in bridging the semantic gap between input text prompts with sparse semantics and the target images, often over-relying on low-level control signals to infer regional details. To address this challenge, we propose ControlThinker, a novel framework that employs a "comprehend-then-generate" paradigm. Firstly, by incentivizing the visual reasoning capability of a MLLM, latent semantics from control images are mined to enrich text prompts. This enriched semantic understanding then seamlessly aids in image generation without the need for additional complex modifications. To further tackle the uncertainty arising from the ambiguity of control images, we encourage broader exploration of reasoning trajectories and select the optimal one using a metric-based output reward model (ORM). Extensive experimental results demonstrate that ControlThinker effectively mitigates the semantic gap between raw text prompts and target images, resulting in improved visual quality and semantic consistency across a wide range of benchmarks. The code and models are available at https://github.com/Maplebb/ControlThinker.