Paper 2606.13679
InterleaveThinker: Reinforcing Agentic Interleaved Generation
- Published
- Jun 2026
- Research lab
- Independent
- Citations
- 0
- GitHub
- 208 stars
01 In brief
Summary
InterleaveThinker is a multi-agent framework that endows existing image generators with interleaved text-image generation capabilities, addressing visual over-reliance and step-wise error accumulation in Unified Multimodal Models (UMMs).
It uses a Planner agent to pre-plan the full instruction sequence, a Generator (e.g., FLUX.2-klein-9B) to execute steps, and a Critic agent to evaluate outputs and refine prompts iteratively.
Training data includes Interleave-Planner-SFT-80k, Interleave-Critic-SFT-112k, and Interleave-Critic-RL-13k, with a dual-reward strategy (accuracy and step-wise rewards) enabling efficient single-step RL via GRPO.
On benchmarks, InterleaveThinker+FLUX.2-klein achieves UEval average 66.3, CoMM IRS 5.2/8.2, WISE 0.73 (up from 0.47), and RISE 28.9 (up from 13.3), matching proprietary models like Nano Banana and GPT-5.
Ablations confirm the multi-agent design, data filtering, and RL rewards are essential.
Limitations include dependence on the base generator's knowledge, as shown in failing cases.
02 From the paper
Abstract
Recent image generators have demonstrated impressive photorealism and instruction-following capabilities in single-image generation and editing. However, constrained by their architectures, they cannot achieve interleaved generation (text-image sequence), which has crucial applications in visual narratives, guidance, and embodied manipulation. Even the latest open-source Unified Multimodal Models (UMMs) exhibit limited performance in this regard. In this paper, we introduce InterleaveThinker, the first multi-agent pipeline designed to endow any existing image generator with interleaved generation capabilities. Specifically, we employ a planner agent to organize the image-text input sequence, instructing the image generator on the required execution at each step. Subsequently, we introduce a critic agent to evaluate the generator's outputs, identify samples that deviate from the planned instructions, and refine the instructions for regeneration. To implement this pipeline, we construct the Interleave-Planner-SFT-80k and Interleave-Critic-SFT-112k to perform a format cold-start. Then we develop Interleave-Critic-RL-13k to reinforce the step-wise instruction correction capability within a generation trajectory using GRPO. Since a single interleaved generation trajectory may involve over 25 generator calls, optimizing the entire trajectory is computationally impractical. Therefore, we propose accuracy reward and step-wise reward, allowing single-step RL to effectively guide the entire generation trajectory. The results show that InterleaveThinker improves performance across various image generators. On interleaved generation benchmarks, it achieves performance comparable to Nano Banana and GPT-5. Surprisingly, it also significantly enhances the base model on reasoning-based benchmarks; for example, on 4-step FLUX.2-klein, we observe substantial gains on WISE and RISE.