Search

Explore-and-exploit tradeoffs play a key role in recommendation systems (RSs), aiming at serving users better by learning from previous interactions. Despite their commercial success, the societal effects of explore-and-exploit mechanisms are not well understood, especially regarding the utility discrepancy they generate between different users. In this work, we measure such discrepancy using the economic notion of envy. We present a multi-armed bandit-like model in which every round consists of several sessions, and rewards are realized once per round. We call the latter property reward consistency, and show that the RS can leverage this property for better societal outcomes. On the downside, doing so also generates envy, as late-to-arrive users enjoy the information gathered by early-to-arrive users. We examine the generated envy under several arrival order mechanisms and virtually any anonymous algorithm, i.e., any algorithm that treats all similar users similarly without leveraging their identities. We provide tight envy bounds on uniform arrival and upper bound the envy for nudged arrival, in which the RS can affect the order of arrival by nudging its users. Furthermore, we study the efficiency-fairness trade-off by devising an algorithm that allows constant envy and approximates the optimal welfare in restricted settings. Finally, we validate our theoretical results empirically using simulations.

We introduce a general class $F_0$ of additive functions $f$ such that $f(p) = 1$ and prove a tight bound for exponential sums of the form $\sum_{n \le x} f(n) e(\alpha n)$ where $f \in F_0$ and $e(\theta) = \exp(2\pi i \theta)$. Both $\omega$, the number of distinct primes of $n$, and $\Omega$, the total number primes of $n$, are members of $F_0$. As an application of the exponential sum result, we use the Hardy-Littlewood circle method to find the asymptotics of the Goldbach-Vinogradov ternary problem associated to $\Omega$, namely we show the behavior of $r_\Omega(N) = \sum_{n_1+n_2+n_3=N}\Omega(n_1)\Omega(n_2)\Omega(n_3)$, as $N \to \infty$. Lastly, we end with a discussion of further applications of the main result., Comment: 17 pages

We consider a refinement to the notions of collusion-resistance in transaction fee mechanisms. In particular, we require that the collusion is by itself incentive-compatible and individually rational to all of its participants. We then study the structural properties of these notions, and importantly, characterize the class of collusion-resistant and incentive-compatible transaction fee mechanisms in the single bidder case, and show that this is exactly the class of posted-price where the price is not too prohibitive. We analyze welfare and revenue implications, as well as the shape of the solution space, for both regular and non-regular distributions.

We address the problem of searching for a change point in an anomalous process among a finite set of M processes. Specifically, we address a composite hypothesis model in which each process generates measurements following a common distribution with an unknown parameter (vector). This parameter belongs to either a normal or abnormal space depending on the current state of the process. Before the change point, all processes, including the anomalous one, are in a normal state; after the change point, the anomalous process transitions to an abnormal state. Our goal is to design a sequential search strategy that minimizes the Bayes risk by balancing sample complexity and detection accuracy. We propose a deterministic search algorithm with the following notable properties. First, we analytically demonstrate that when the distributions of both normal and abnormal processes are unknown, the algorithm is asymptotically optimal in minimizing the Bayes risk as the error probability approaches zero. In the second setting, where the parameter under the null hypothesis is known, the algorithm achieves asymptotic optimality with improved detection time based on the true normal state. Simulation results are presented to validate the theoretical findings., Comment: 13 pages, 6 figures

Perpetual voting studies fair collective decision-making in settings where many decisions are to be made, and is a natural framework for settings such as parliaments and the running of blockchain Decentralized Autonomous Organizations (DAOs). We focus our attention on the binary case (YES/NO decisions) and \textit{individual} guarantees for each of the participating agents. We introduce a novel notion, inspired by the popular maxi-min-share (MMS) for fair allocation. The agent expects to get as many decisions as if they were to optimally partition the decisions among the agents, with an adversary deciding which of the agents decides on what bundle. We show an online algorithm that guarantees the MMS notion for $n=3$ agents, an offline algorithm for $n=4$ agents, and show that no online algorithm can guarantee the $MMS^{adapt}$ for $n\geq 7$ agents. We also show that the Maximum Nash Welfare (MNW) outcome can only guarantee $O(\frac{1}{n})$ of the MMS notion in the worst case.

We introduce Meta 3D Gen (3DGen), a new state-of-the-art, fast pipeline for text-to-3D asset generation. 3DGen offers 3D asset creation with high prompt fidelity and high-quality 3D shapes and textures in under a minute. It supports physically-based rendering (PBR), necessary for 3D asset relighting in real-world applications. Additionally, 3DGen supports generative retexturing of previously generated (or artist-created) 3D shapes using additional textual inputs provided by the user. 3DGen integrates key technical components, Meta 3D AssetGen and Meta 3D TextureGen, that we developed for text-to-3D and text-to-texture generation, respectively. By combining their strengths, 3DGen represents 3D objects simultaneously in three ways: in view space, in volumetric space, and in UV (or texture) space. The integration of these two techniques achieves a win rate of 68% with respect to the single-stage model. We compare 3DGen to numerous industry baselines, and show that it outperforms them in terms of prompt fidelity and visual quality for complex textual prompts, while being significantly faster.

We present Meta 3D AssetGen (AssetGen), a significant advancement in text-to-3D generation which produces faithful, high-quality meshes with texture and material control. Compared to works that bake shading in the 3D object's appearance, AssetGen outputs physically-based rendering (PBR) materials, supporting realistic relighting. AssetGen generates first several views of the object with factored shaded and albedo appearance channels, and then reconstructs colours, metalness and roughness in 3D, using a deferred shading loss for efficient supervision. It also uses a sign-distance function to represent 3D shape more reliably and introduces a corresponding loss for direct shape supervision. This is implemented using fused kernels for high memory efficiency. After mesh extraction, a texture refinement transformer operating in UV space significantly improves sharpness and details. AssetGen achieves 17% improvement in Chamfer Distance and 40% in LPIPS over the best concurrent work for few-view reconstruction, and a human preference of 72% over the best industry competitors of comparable speed, including those that support PBR. Project page with generated assets: https://assetgen.github.io, Comment: Project Page: https://assetgen.github.io

The recent availability and adaptability of text-to-image models has sparked a new era in many related domains that benefit from the learned text priors as well as high-quality and fast generation capabilities, one of which is texture generation for 3D objects. Although recent texture generation methods achieve impressive results by using text-to-image networks, the combination of global consistency, quality, and speed, which is crucial for advancing texture generation to real-world applications, remains elusive. To that end, we introduce Meta 3D TextureGen: a new feedforward method comprised of two sequential networks aimed at generating high-quality and globally consistent textures for arbitrary geometries of any complexity degree in less than 20 seconds. Our method achieves state-of-the-art results in quality and speed by conditioning a text-to-image model on 3D semantics in 2D space and fusing them into a complete and high-resolution UV texture map, as demonstrated by extensive qualitative and quantitative evaluations. In addition, we introduce a texture enhancement network that is capable of up-scaling any texture by an arbitrary ratio, producing 4k pixel resolution textures.

In the fully-anonymous (shared-memory) model, inspired by a biological setting, processors have no identifiers and memory locations are anonymous. This means that there is no pre-existing agreement among processors on any naming of the memory locations. In this work, we ask fundamental questions about the fully-anonymous model in the hope to obtain a better understanding of the role of naming and anonymity in distributed computing. First, we ask what it means to solve a task under processor anonymity. With tasks such as renaming, the traditional notion obviously does not apply. Instead of restricting ourselves to colorless tasks, we propose using the notion of group solvability, which allows transferring any task to processor-anonymous models. Second, the difficulty with anonymity is that processors can hardly avoid covering and then overwriting each other's writes, erasing information written by their predecessors. To get to the bottom of this phenomenon, we ask what system configurations are stable when processors keep reading and writing ad infinitum. Resolving this question leads us to a wait-free solution to the snapshot task, which then allows us to solve renaming and obstruction-free consensus.

Two firms are engaged in a competitive prediction task. Each firm has two sources of data -- labeled historical data and unlabeled inference-time data -- and uses the former to derive a prediction model, and the latter to make predictions on new instances. We study data-sharing contracts between the firms. The novelty of our study is to introduce and highlight the differences between contracts that share prediction models only, contracts to share inference-time predictions only, and contracts to share both. Our analysis proceeds on three levels. First, we develop a general Bayesian framework that facilitates our study. Second, we narrow our focus to two natural settings within this framework: (i) a setting in which the accuracy of each firm's prediction model is common knowledge, but the correlation between the respective models is unknown; and (ii) a setting in which two hypotheses exist regarding the optimal predictor, and one of the firms has a structural advantage in deducing it. Within these two settings we study optimal contract choice. More specifically, we find the individually rational and Pareto-optimal contracts for some notable cases, and describe specific settings where each of the different sharing contracts emerge as optimal. Finally, in the third level of our analysis we demonstrate the applicability of our concepts in a synthetic simulation using real loan data.

We consider the problem of dynamic channel allocation (DCA) in cognitive communication networks with the goal of maximizing a global signal-to-interference-plus-noise ratio (SINR) measure under a specified target quality of service (QoS)-SINR for each network. The shared bandwidth is partitioned into K channels with frequency separation. In contrast to the majority of existing studies that assume perfect orthogonality or a one- to-one user-channel allocation mapping, this paper focuses on real-world systems experiencing inter-carrier interference (ICI) and channel reuse by multiple large-scale networks. This realistic scenario significantly increases the problem dimension, rendering existing algorithms inefficient. We propose a novel multi-agent reinforcement learning (RL) framework for distributed DCA, named Channel Allocation RL To Overlapped Networks (CARLTON). The CARLTON framework is based on the Centralized Training with Decentralized Execution (CTDE) paradigm, utilizing the DeepMellow value-based RL algorithm. To ensure robust performance in the interference-laden environment we address, CARLTON employs a low-dimensional representation of observations, generating a QoS-type measure while maximizing a global SINR measure and ensuring the target QoS-SINR for each network. Our results demonstrate exceptional performance and robust generalization, showcasing superior efficiency compared to alternative state-of-the-art methods, while achieving a marginally diminished performance relative to a fully centralized approach.

Most text-to-3D generators build upon off-the-shelf text-to-image models trained on billions of images. They use variants of Score Distillation Sampling (SDS), which is slow, somewhat unstable, and prone to artifacts. A mitigation is to fine-tune the 2D generator to be multi-view aware, which can help distillation or can be combined with reconstruction networks to output 3D objects directly. In this paper, we further explore the design space of text-to-3D models. We significantly improve multi-view generation by considering video instead of image generators. Combined with a 3D reconstruction algorithm which, by using Gaussian splatting, can optimize a robust image-based loss, we directly produce high-quality 3D outputs from the generated views. Our new method, IM-3D, reduces the number of evaluations of the 2D generator network 10-100x, resulting in a much more efficient pipeline, better quality, fewer geometric inconsistencies, and higher yield of usable 3D assets.

To allocate transactions to blocks, cryptocurrencies use an auction-like transaction fee mechanism (TFM). A conjecture of Roughgarden [44] asks whether there is a TFM that is incentive compatible for both the users and the miner, and is also resistant to off-chain agreements (OCAs) between these parties, a collusion notion that captures the ability of users and the miner to jointly deviate for profit. The work of Chung and Shi [12] tackles the problem using the different collusion resistance notion of side-channel proofness (SCP), and shows an impossibility given this notion. We show that OCA-proofness and SCP are different, with SCP being strictly stronger. We then fully characterize the intersection of deterministic dominant strategy incentive-compatible (DSIC) and OCA-proof mechanisms, as well as deterministic MMIC and OCA-proof ones, and use this characterization to show that only the trivial mechanism is DSIC, myopic miner incentive-compatible (MMIC) and OCA-proof. We also show that a randomized mechanism can be at most 0.842-efficient in the worst case, and that the impossibility of a non-trivial DSIC, MMIC and OCA-proof extends to a couple of natural classes of randomized mechanisms., Comment: 29 pages

We study a \emph{financial} version of the classic online problem of scheduling weighted packets with deadlines. The main novelty is that, while previous works assume packets have \emph{fixed} weights throughout their lifetime, this work considers packets with \emph{time-decaying} values. Such considerations naturally arise and have wide applications in financial environments, where the present value of future actions may be discounted. We analyze the competitive ratio guarantees of scheduling algorithms under a range of discount rates encompassing the ``traditional'' undiscounted case where weights are fixed (i.e., a discount rate of 1), the fully discounted ``myopic'' case (i.e., a rate of 0), and those in between. We show how existing methods from the literature perform suboptimally in the more general discounted setting. Notably, we devise a novel memoryless deterministic algorithm, and prove that it guarantees the best possible competitive ratio attainable by deterministic algorithms for discount factors up to $\approx 0.77$. Moreover, we develop a randomized algorithm and prove that it outperforms the best possible deterministic algorithm, for any discount rate. While we highlight the relevance of our framework and results to blockchain transaction scheduling in particular, our approach and analysis techniques are general and may be of independent interest., Comment: 43 pages, 9 figures, 5 tables

Current diffusion or flow-based generative models for 3D shapes divide to two: distilling pre-trained 2D image diffusion models, and training directly on 3D shapes. When training a diffusion or flow models on 3D shapes a crucial design choice is the shape representation. An effective shape representation needs to adhere three design principles: it should allow an efficient conversion of large 3D datasets to the representation form; it should provide a good tradeoff of approximation power versus number of parameters; and it should have a simple tensorial form that is compatible with existing powerful neural architectures. While standard 3D shape representations such as volumetric grids and point clouds do not adhere to all these principles simultaneously, we advocate in this paper a new representation that does. We introduce Mosaic-SDF (M-SDF): a simple 3D shape representation that approximates the Signed Distance Function (SDF) of a given shape by using a set of local grids spread near the shape's boundary. The M-SDF representation is fast to compute for each shape individually making it readily parallelizable; it is parameter efficient as it only covers the space around the shape's boundary; and it has a simple matrix form, compatible with Transformer-based architectures. We demonstrate the efficacy of the M-SDF representation by using it to train a 3D generative flow model including class-conditioned generation with the 3D Warehouse dataset, and text-to-3D generation using a dataset of about 600k caption-shape pairs., Comment: More results and details can be found at https://lioryariv.github.io/msdf

We present Semantify: a self-supervised method that utilizes the semantic power of CLIP language-vision foundation model to simplify the control of 3D morphable models. Given a parametric model, training data is created by randomly sampling the model's parameters, creating various shapes and rendering them. The similarity between the output images and a set of word descriptors is calculated in CLIP's latent space. Our key idea is first to choose a small set of semantically meaningful and disentangled descriptors that characterize the 3DMM, and then learn a non-linear mapping from scores across this set to the parametric coefficients of the given 3DMM. The non-linear mapping is defined by training a neural network without a human-in-the-loop. We present results on numerous 3DMMs: body shape models, face shape and expression models, as well as animal shapes. We demonstrate how our method defines a simple slider interface for intuitive modeling, and show how the mapping can be used to instantly fit a 3D parametric body shape to in-the-wild images.

In the 1980s, three related impossibility results emerged in the field of distributed computing. First, Fischer, Lynch, and Paterson demonstrated that deterministic consensus is unattainable in an asynchronous message-passing system when a single process may crash-stop. Subsequently, Loui and Abu-Amara showed the infeasibility of achieving consensus in asynchronous shared-memory systems, given the possibility of one crash-stop failure. Lastly, Santoro and Widmayer established the impossibility of consensus in synchronous message-passing systems with a single process per round experiencing send-omission faults. In this paper, we revisit these seminal results. First, we observe that all these systems are equivalent in the sense of implementing each other. Then, we prove the impossibility of consensus in the synchronous system of Santoro and Widmayer, which is the easiest to reason about. Taking inspiration from V\"olzer's proof pearl and from the Borowski-Gafni simulation, we obtain a remarkably simple proof. We believe that a contemporary pedagogical approach to teaching these results should first address the equivalence of the systems before proving the consensus impossibility within the system where the result is most evident., Comment: Added reference to conference version, and corrected typos and small errors

Federated learning (FL) is an emerging machine learning paradigm for training models across multiple edge devices holding local data sets, without explicitly exchanging the data. Recently, over-the-air (OTA) FL has been suggested to reduce the bandwidth and energy consumption, by allowing the users to transmit their data simultaneously over a Multiple Access Channel (MAC). However, this approach results in channel noise directly affecting the optimization procedure, which may degrade the accuracy of the trained model. In this paper we jointly exploit the prior distribution of local weights and the channel distribution, and develop an OTA FL algorithm based on a Bayesian approach for signal aggregation. Our proposed algorithm, dubbed Bayesian Air Aggregation Federated learning (BAAF), is shown to effectively mitigate noise and fading effects induced by the channel. To handle statistical heterogeneity of users data, which is a second major challenge in FL, we extend BAAF to allow for appropriate local updates by the users and develop the Controlled Bayesian Air Aggregation Federated-learning (COBAAF) algorithm. In addition to using a Bayesian approach to average the channel output, COBAAF controls the drift in local updates using a judicious design of correction terms. We analyze the convergence of the learned global model using BAAF and COBAAF in noisy and heterogeneous environment, showing their ability to achieve a convergence rate similar to that achieved over error-free channels. Simulation results demonstrate the improved convergence of BAAF and COBAAF over existing algorithms in machine learning tasks., Comment: A short version of this paper was presented at the 58th Annual Allerton Conference on Communication, Control, and Computing

Like Asynchrony, Mobility of faults precludes consensus. Yet, a model M in which Consensus is solvable, has an analogue relaxed model in which Consensus is not solvable and for which we can ask, whether Consensus is solvable if the system initially behaves like the relaxed analogue model, but eventually morphs into M. We consider two relaxed analogues of M. The first is the traditional Asynchronous model, and the second to be defined, the Mobile analogue. While for some M we show that Consensus is not solvable in the Asynchronous analogue, it is solvable in all the Mobile analogues. Hence, from this perspective Mobility is superior to Asynchrony. The pie in the sky relationship we envision is: Consensus is solvable in M, if and only if binary Commit-Adopt is solvable in the mobile analogue. The ``only if'' is easy. Here we show case by case that the ``if'' holds for all the common faults types.

We propose a new distributed-computing model, inspired by permissionless distributed systems such as Bitcoin and Ethereum, that allows studying permissionless consensus in a mathematically regular setting. Like in the sleepy model of Pass and Shi, we consider a synchronous, round-by-round message-passing system in which the set of online processors changes each round. Unlike the sleepy model, the set of processors may be infinite. Moreover, processors never fail; instead, an adversary can temporarily or permanently impersonate some processors. Finally, processors have access to a strong form of message-authentication that authenticates not only the sender of a message but also the round in which the message was sent. Assuming that, each round, the adversary impersonates less than 1/2 of the online processors, we present two consensus algorithms. The first ensures deterministic safety and constant latency in expectation, assuming a probabilistic leader-election oracle. The second ensures deterministic safety and deterministic liveness assuming irrevocable impersonation and eventually-stabilizing participation. The model is unrealistic in full generality. However, if we assume finitely many processes and that the set of faulty processes remains constant, the model coincides with a practically-motivated model: the static version of the sleepy model.

Recent text-to-image diffusion models are able to generate convincing results of unprecedented quality. However, it is nearly impossible to control the shapes of different regions/objects or their layout in a fine-grained fashion. Previous attempts to provide such controls were hindered by their reliance on a fixed set of labels. To this end, we present SpaText - a new method for text-to-image generation using open-vocabulary scene control. In addition to a global text prompt that describes the entire scene, the user provides a segmentation map where each region of interest is annotated by a free-form natural language description. Due to lack of large-scale datasets that have a detailed textual description for each region in the image, we choose to leverage the current large-scale text-to-image datasets and base our approach on a novel CLIP-based spatio-textual representation, and show its effectiveness on two state-of-the-art diffusion models: pixel-based and latent-based. In addition, we show how to extend the classifier-free guidance method in diffusion models to the multi-conditional case and present an alternative accelerated inference algorithm. Finally, we offer several automatic evaluation metrics and use them, in addition to FID scores and a user study, to evaluate our method and show that it achieves state-of-the-art results on image generation with free-form textual scene control., Comment: CVPR 2023. Project page available at: https://omriavrahami.com/spatext

In robust decision making under uncertainty, a natural choice is to go with safety (aka security) level strategies. However, in many important cases, most notably auctions, there is a large multitude of safety level strategies, thus making the choice unclear. We consider two refined notions: (i) a term we call DSL (distinguishable safety level), and is based on the notion of ``discrimin'', which uses a pairwise comparison of actions while removing trivial equivalencies. This captures the fact that when comparing two actions an agent should not care about payoffs in situations where they lead to identical payoffs. (ii) The well-known Leximin notion from social choice theory, which we apply for robust decision-making. In particular, the leximin is always DSL but not vice-versa. We study the relations of these notions to other robust notions, and illustrate the results of their use in auctions and other settings. Economic design aims to maximize social welfare when facing self-motivated participants. In online environments, such as the Web, participants' incentives take a novel form originating from the lack of clear agent identity -- the ability to create Sybil attacks, i.e., the ability of each participant to act using multiple identities. It is well-known that Sybil attacks are a major obstacle for welfare-maximization. Our main result proves that when DSL attackers face uncertainty over the auction's bids, the celebrated VCG mechanism is welfare-maximizing even under Sybil attacks. Altogether, our work shows a successful fundamental synergy between robustness under uncertainty, economic design, and agents' strategic manipulations in online multi-agent systems., Comment: In Proceedings TARK 2023, arXiv:2307.04005

Cryptocurrencies employ auction-esque transaction fee mechanisms (TFMs) to allocate transactions to blocks, and to determine how much fees miners can collect from transactions. Several impossibility results show that TFMs that satisfy a standard set of "good" properties obtain low revenue, and in certain cases, no revenue at all. In this work, we circumvent previous impossibilities by showing that when desired TFM properties are reasonably relaxed, simple mechanisms can obtain strictly positive revenue. By discretizing fees, we design a TFM that satisfies the extended TFM desiderata: it is dominant strategy incentive-compatible (DSIC), myopic miner incentive-compatible (MMIC), side-contract-proof (SCP) and obtains asymptotically optimal revenue (i.e., linear in the number of allocated bids), and optimal revenue when considering separable TFMs. If instead of discretizing fees we relax the DSIC and SCP properties, we show that Bitcoin's TFM, after applying the revelation principle, is Bayesian incentive-compatible (BIC), MMIC, off-chain-agreement (OCA) proof, and approximately revenue-optimal. We reach our results by characterizing the class of multi-item OCA-proof mechanisms, which may be of independent interest., Comment: 28 pages

We propose Make-A-Video -- an approach for directly translating the tremendous recent progress in Text-to-Image (T2I) generation to Text-to-Video (T2V). Our intuition is simple: learn what the world looks like and how it is described from paired text-image data, and learn how the world moves from unsupervised video footage. Make-A-Video has three advantages: (1) it accelerates training of the T2V model (it does not need to learn visual and multimodal representations from scratch), (2) it does not require paired text-video data, and (3) the generated videos inherit the vastness (diversity in aesthetic, fantastical depictions, etc.) of today's image generation models. We design a simple yet effective way to build on T2I models with novel and effective spatial-temporal modules. First, we decompose the full temporal U-Net and attention tensors and approximate them in space and time. Second, we design a spatial temporal pipeline to generate high resolution and frame rate videos with a video decoder, interpolation model and two super resolution models that can enable various applications besides T2V. In all aspects, spatial and temporal resolution, faithfulness to text, and quality, Make-A-Video sets the new state-of-the-art in text-to-video generation, as determined by both qualitative and quantitative measures.

Recent text-to-image models have achieved impressive results. However, since they require large-scale datasets of text-image pairs, it is impractical to train them on new domains where data is scarce or not labeled. In this work, we propose using large-scale retrieval methods, in particular, efficient k-Nearest-Neighbors (kNN), which offers novel capabilities: (1) training a substantially small and efficient text-to-image diffusion model without any text, (2) generating out-of-distribution images by simply swapping the retrieval database at inference time, and (3) performing text-driven local semantic manipulations while preserving object identity. To demonstrate the robustness of our method, we apply our kNN approach on two state-of-the-art diffusion backbones, and show results on several different datasets. As evaluated by human studies and automatic metrics, our method achieves state-of-the-art results compared to existing approaches that train text-to-image generation models using images only (without paired text data)

We study the number of degree $n$ number fields with discriminant bounded by $X$. In this article, we improve an upper bound due to Schmidt on the number of such fields that was previously the best known upper bound for $6 \leq n \leq 94$., Comment: 24 pages; official version for Discrete Analysis

Recent text-to-image generation methods provide a simple yet exciting conversion capability between text and image domains. While these methods have incrementally improved the generated image fidelity and text relevancy, several pivotal gaps remain unanswered, limiting applicability and quality. We propose a novel text-to-image method that addresses these gaps by (i) enabling a simple control mechanism complementary to text in the form of a scene, (ii) introducing elements that substantially improve the tokenization process by employing domain-specific knowledge over key image regions (faces and salient objects), and (iii) adapting classifier-free guidance for the transformer use case. Our model achieves state-of-the-art FID and human evaluation results, unlocking the ability to generate high fidelity images in a resolution of 512x512 pixels, significantly improving visual quality. Through scene controllability, we introduce several new capabilities: (i) Scene editing, (ii) text editing with anchor scenes, (iii) overcoming out-of-distribution text prompts, and (iv) story illustration generation, as demonstrated in the story we wrote.