controllable-generation

Shaer: a controllable Classical Arabic poetry generation framework that jointly conditions on natural-language descriptions, meter subforms, and target hemistich counts, built by QLoRA fine-tuning Yehia-7B on a 116,032-poem enriched corpus, achieving 95.17% base-meter accuracy and strong length/subform control.

SpaceFlow 提出了一种免训练的局部可控 3D 生成流程，用户可通过几何基元为不同区域指定控制强度，并分别路由文本或图像提示以合成外观，从而在几何保真与生成自由度之间取得平衡。

Black Forest Labs has released FLUX 3 Image, a controllable image generation model that supports synthesizing images from scratch, allowing precise layout of elements in the scene via bounding boxes and maximum control over every pixel.

This paper introduces SVEET, a framework for high-quality streaming video editing that leverages a pretrained video diffusion model to enable auto-regressive editing with real-time performance on a single GPU.

PAUSE introduces editable strategy artifacts to make cultural decisions in long-form story adaptation more inspectable and contestable by humans. Experiments show that human edits to the strategy effectively propagate into chapter-level prose, improving transparency in AI-mediated cultural adaptation.

Surg-UniWorld is a unified surgical world model with multimodal control experts, enabling controllable generation of coherent instrument-tissue interaction videos using edge, depth, and optical-flow inputs. It introduces a new benchmark (Cholec80-SurgWAM) and outperforms existing controllable video generation methods.

Introduces appearance pointers, compact tokens that guide Diffusion Transformers to apply correct appearance cues at specified spatial locations, enabling modality-agnostic localized multimodal control without retraining the base model.

Malleable Prompting is a novel interactive technique that reifies natural language preferences into GUI widgets (sliders, toggles, dropdowns) for direct manipulation, with a decoding algorithm that modulates token probabilities based on widget values to enable precise control over LLM generation. A user study shows it outperforms natural language prompting in precision, controllability, and transparency.

A comprehensive blog post by Volodymyr Kuleshov's Cornell group explains how to build diffusion language models, covering core techniques like masked diffusion, iterative refinement, variable-length generation, controllable generation, fast samplers, and RL post-training, using open-source models such as Mercury, Gemma Diffusion, and Nemotron Diffusion as examples.

CONFLUX is a 3D latent diffusion model for chest CT synthesis that achieves high-fidelity volumetric generation with controllable clinical attributes, enhanced by a reinforcement learning post-training stage to improve conditioning reliability. The model and a synthetic dataset of ~200k chest CT volumes are released.

This paper introduces a French OSCE dialogue dataset of 240 interactions and a controllable LLM-based pipeline for generating synthetic OSCE dialogues, enabling realistic virtual patient simulations for medical training with automatic feedback.

InstanceControl enables multi-instance controllable image generation without manual instance labeling by leveraging a vision-language model to establish instance-level correspondences between text prompts and visual conditions, with adaptive mask refinement for improved accuracy.

This paper introduces LISA, a regularization method that aligns the intermediate features of a side network with an approximated likelihood score to improve training efficiency and the quality of visual-condition controllable generation in score-based generative models.

Arbor introduces explicit geometric control for 3D asset generation by using constraint meshes (hull, avoidance, touch regions) to condition latent generation, improving spatial constraint adherence without sacrificing object quality.

Holo-World presents a unified controllable video world model that generates videos from a single image with explicit control over camera, object motion, and weather. It introduces a novel dataset and techniques to preserve scene structure while transferring to target weather states.

SciR is a new controllable benchmark for evaluating LLMs on scientific reasoning including deduction, induction, and causal abduction, with parametric control over extraction and inference difficulty. Tests show both axes degrade performance across models, with reasoning models like DeepSeek-R1 outperforming instruct models on inference.

This paper proposes AirfoilGen, a latent diffusion model for airfoil shape generation that ensures geometric validity via a circle sweeping representation and enables control over aerodynamic performance (lift/drag coefficients). Experiments show 98.41% performance-conditioning accuracy, using a new dataset of over 200,000 airfoils.

CogOmniControl is a reasoning-driven framework for controllable video generation that uses a specialized vision-language model (CogVLM) trained on anime production data to infer creative intent from sparse conditions, then guides a diffusion-based generator via reinforcement learning, achieving state-of-the-art results on new benchmarks.

Proposes CoMole, a controllable molecular generative foundation model using motif-aware graph diffusion and reinforcement learning, achieving superior controllability across materials and drug discovery benchmarks.

This paper introduces a method for controllable generation in flow matching by adjusting the conditional endpoint mean using a reference set, offering both training-free and semi-parametric guidance for style and content control.