Generative models typically sample outputs independently, and recent inference-time guidance and scaling algorithms focus on improving the quality of individual samples. However, in real-world applications, users are often presented with a set of multiple images (e.g., 4-8) for each prompt, where independent sampling tends to lead to redundant results, limiting user choices and hindering idea exploration. In this work, we introduce a scalable group inference method that improves both the diversity and quality of a group of samples. We formulate group inference as a quadratic integer assignment problem: candidate outputs are modeled as graph nodes, and a subset is selected to optimize sample quality (unary term) while maximizing group diversity (binary term). To substantially improve runtime efficiency, we progressively prune the candidate set using intermediate predictions, allowing our method to scale up to large candidate sets. Extensive experiments show that our method significantly improves group diversity and quality compared to independent sampling baselines and recent inference algorithms. Our framework generalizes across a wide range of tasks, including text-to-image, image-to-image, image prompting, and video generation, enabling generative models to treat multiple outputs as cohesive groups rather than independent samples.

Jays react angrily when shown a cup-and-balls-style magic trick in which their favourite snack is swapped for a less appealing one. Their responses show cognitive abilities that may come into play when they pilfer food caches hidden by other birds. Eurasian jays (Garrulus glandarius) have impressive memories and show some capacity for imagining the beliefs and intentions of others, known as theory of mind. As such, Alexandra Schnell and her colleagues at the University of Cambridge wondered whether jays would be sensitive to cognitive illusions designed to fool humans. First, they tested six birds to find out which food each one preferred from a choice of worms, cheese and peanuts.

In an amazing show of self-control, cuttlefish can resist the impulse to eat a morsel of food if it means getting to eat two morsels later on, a new study shows. In experiments, the marine molluscs passed a variation of the'marshmallow test' – originally used in the 1970s to measure a child's ability to delay gratification. In the original Stanford experiment, pre-school kids were given one marshmallow and told they could eat it straight away, or, if they waited 20 minutes, have two marshmallows instead. For this new study, scientists performed a'fishy version' of the legendary experiment using shrimp instead of marshmallows. They found the creatures could wait over two minutes to get their preferred type of shrimp – and that the cuttlefish that could delay gratification the longest were the most intelligent, as determined by a another learning task.