Hierarchical Text-Conditional Image Generation with CLIP Latents (DALL-E 2 / unCLIP)
OpenAI published research on hierarchical text-conditional image generation using CLIP latents, the technique underlying DALL-E 2. The approach uses a prior network to map text embeddings to image embeddings, then a diffusion decoder to generate images from those embeddings. This represented a significant advance in text-to-image generation quality and semantic fidelity at the time of release.
Related guides (3)
Related events (8)
DALL·E: Creating Images from Text
OpenAI announced DALL·E, a neural network capable of generating images from natural language text captions across a wide range of concepts. The model represents an early milestone in text-to-image generation using transformer-based architectures. This January 2021 announcement predates the broader diffusion-model wave and marks a foundational step in multimodal generative AI.
CLIP: Connecting Text and Images
OpenAI introduced CLIP (Contrastive Language-Image Pre-training), a neural network that learns visual concepts from natural language supervision. CLIP enables zero-shot visual classification by accepting natural language descriptions of categories rather than requiring task-specific training data. The approach mirrors the zero-shot transfer capabilities demonstrated by GPT-2 and GPT-3 in the language domain.
TEVI: Sparse autoencoders for text-conditioned editing of CLIP image embeddings to improve vision-language alignment
TEVI is a framework that uses sparse autoencoders to disentangle CLIP image embeddings and a learned masking module to selectively reconstruct embeddings conditioned on a given caption, addressing the information imbalance between images and their captions. The approach improves image-text retrieval on both coarse-grained benchmarks (MS COCO, Flickr) and fine-grained long-caption benchmarks (IIW, DOCCI), with larger gains on richer captions. The work also shows improved robustness on the RoCOCO benchmark.
Squeezing Capacity from MLLMs for Subject-driven Image Generation via Dual Layer Aggregation
This paper proposes conditioning diffusion models on Multimodal Large Language Models (MLLMs) that jointly encode text and reference images, augmented with VAE-based identity conditioning to address copy-paste artifacts and identity preservation failures in subject-driven image generation. A Dual Layer Aggregation (DLA) module aggregates multi-level MLLM features, and a multi-stage denoising strategy progressively balances semantic and fine-detail identity signals during inference. Experiments show improved human preference scores on subject-driven generation benchmarks compared to prior approaches that encode text and reference images separately.
Representation-Conditioned Diffusion Models for Controllable Image Generation
This paper explores conditioning diffusion models on representations from pre-trained self-supervised models as an alternative to text prompts or semantic maps, which require large annotated datasets. The self-conditioning mechanism improves unconditional image generation quality and provides a controllable representation space. The authors identify directions of variation in this space and demonstrate smoothness and disentanglement properties, suggesting potential for fine-grained generative control without heavy annotation overhead.
DALL·E 2 Pre-Training Mitigations
OpenAI describes the safety mitigations applied during pre-training of DALL·E 2 to reduce risks from powerful image generation models. The post outlines guardrails designed to prevent generated images from violating content policy. This represents an early public disclosure of safety-by-design approaches for large generative image models.
Fine-Tuning CLIP with Remote Sensing Satellite Images and Captions
This Hugging Face blog post describes fine-tuning OpenAI's CLIP model on the RSICD (Remote Sensing Image Captioning Dataset) to improve vision-language alignment for satellite and aerial imagery. The work demonstrates domain adaptation of a general-purpose contrastive vision-language model to a specialized remote sensing context. It serves as a practical tutorial and case study for transfer learning with CLIP on narrow domains.
Pose-ICL: 3D-aware in-context learning for pose-controllable image generation of custom subjects
Researchers introduce Pose-ICL, a tuning-free framework for generating images of user-specified subjects with accurate pose control. The method uses Surface-Anchored Position Embedding (SAPE) to give 2D diffusion models explicit 3D awareness by anchoring image tokens to volumetric bounding box surface coordinates. Evaluations on 3D assets and real-world subjects show improvements over existing methods in both pose accuracy and identity consistency. The framework is designed for compatibility with existing Diffusion Transformer (DiT) models.