Thinking-while-Generating: Interleaving Textual Reasoning throughout Visual Generation

Kari Jaaskelainen

21 Nov 2025 — 1 min read

AI that thinks while it draws

Meet TwiG (“Thinking‑while‑Generating”), a framework that lets an image model think in words as it generates pixels. Instead of only planning before or fixing after, TwiG interleaves short textual reasoning during generation—guiding upcoming regions and reflecting on what’s already made. The goal is more context‑aware, semantically rich visuals.

What they tried

Zero‑shot prompting: simple instructions make the model narrate and adjust on the fly.
Supervised fine‑tuning (TwiG‑50K): training on curated examples of images paired with interleaved “thoughts.”
Reinforcement learning (TwiG‑GRPO): rewards encourage helpful, timely reasoning.

Why it matters: letting models “talk through” their brushstrokes can reduce inconsistencies (e.g., mismatched objects or lighting) and improve local detail without losing the big picture.

Preliminary study. Paper: https://arxiv.org/abs/2511.16671v1 — Code (coming): https://github.com/ZiyuGuo99/Thinking-while-Generating

Paper: https://arxiv.org/abs/2511.16671v1

Register: https://www.AiFeta.com

#AI #GenAI #ComputerVision #ImageGeneration #Multimodal #MachineLearning #Research #OpenSource

Meet GRACE: a moral governor for safer, more transparent AI

AI agents are getting powerful—so how do we make sure they do the right thing, not just the effective thing? Meet GRACE, a reason-based moral governor that keeps AI behavior aligned with human norms by separating moral reasoning from goal-driven decision-making. * Moral Module: uses deontic logic and explicit reasons

Why some fine-tuned LLMs miss phishing—and how to fix it

Not all fine-tuned LLMs spot phishing equally. A new study tests Llama 3.1 8B, Gemma 2 9B, and Mistral on high-stakes phishing detection—and uses SHAP and mechanistic interpretability to reveal why models do (or don’t) generalize. * Architecture × data diversity matters: Gemma 2 9B hits state-of-the-art performance (F1

AI that explains itself—by following the science

AI that explains itself—by following the science Black-box AI can be powerful, but it often can’t tell us why it made a decision. Concept Bottleneck Models (CBMs) try to fix that by predicting human-understandable concepts first, then the final answer. The catch: standard CBMs ignore domain-specific cause-and-effect and

Meet GenomAgent: A Team of AI Specialists for Smarter Genomics Q&A

TL;DR: Finding reliable answers in genomics is hard. GenomAgent turns one big AI into a coordinating team of specialists—and beats the current leader by 12% on a key benchmark. Genomic facts live across many databases. Standard chatbots struggle because they can’t flexibly query those sources. GeneGPT added