A text-to-image (T2I or TTI) model is a machine learning model which takes an input natural language prompt and produces an image matching that description. Text-to-image models gradually began to be developed in the mid-2010s during the beginnings of the AI boom, as a result of advances in deep neural networks. In 2022, the output of state-of-the-art text-to-image models—such as OpenAI's DALL-E 2, Google Brain's Imagen, Stability AI's Stable Diffusion, Midjourney, and Runway's Gen-4—began to be considered to approach the quality of real photographs and human-drawn art. Text-to-image models are generally latent diffusion models, which perform the diffusion process in a compressed latent space rather than directly in pixel space. An autoencoder (often a variational autoencoder) is used to convert between pixel space and this latent representation. These systems typically use a pretrained language or vision–language model to convert the input prompt into a text embedding, and a diffusion-based generative image model that produces images conditioned on that embedding. The most effective models have generally been trained on massive amounts of image and text data scraped from the web.
History Before the rise of deep learning in the 2010s, attempts to build text-to-image models were limited to collages by arranging existing component images, such as from a database of clip art. The inverse task, image captioning, was more tractable, and a number of image captioning deep learning models came prior to the first text-to-image models.
2015–2019
The first modern text-to-image model, alignDRAW, was introduced in 2015 by researchers from the University of Toronto. alignDRAW extended the previously introduced DRAW architecture (which used a recurrent variational autoencoder with an attention mechanism) to be conditioned on text sequences. Images generated by alignDRAW were in small resolution (32×32 pixels, attained from resizing) and were considered to be 'low in diversity'. The model was able to generalize to objects not represented in the training data (such as a red school bus) and appropriately handled novel prompts such as "a stop sign is flying in blue skies", exhibiting output that it was not merely "memorizing" data from the training set. In 2016, Reed, Akata, Yan et al. became the first to use generative adversarial networks for the text-to-image task. With models trained on narrow, domain-specific datasets, they were able to generate "visually plausible" images of birds and flowers from text captions like "an all black bird with a distinct thick, rounded bill". A model trained on the more diverse COCO (Common Objects in Context) dataset produced images which were "from a distance... encouraging", but which lacked coherence in their details. Later systems include VQGAN-CLIP, XMC-GAN, and GauGAN2.
2020s One of the first text-to-image models to capture widespread public attention was OpenAI's DALL-E, a transformer system announced in January 2021. A successor capable of generating more complex and realistic images, DALL-E 2, was unveiled in April 2022, followed by Stable Diffusion that was publicly released in August 2022. In August 2022, text-to-image personalization allows to teach the model a new concept using a small set of images of a new object that was not included in the training set of the text-to-image foundation model. This is achieved by textual inversion, namely, finding a new text term that correspond to these images. Additional text-to-image models appeared with Adobe in March 2023 and Black Forest Labs in August 2024. Following other text-to-image models, language model-powered text-to-video platforms such as Runway, Make-A-Video, Imagen Video, Midjourney, and Phenaki can generate video from text and/or text/image prompts.
Architecture and training
Text-to-image models have been built using a variety of architectures. The text encoding step may be performed with a recurrent neural network such as a long short-term memory (LSTM) network, though transformer models have since become a more popular option. For the image generation step, conditional generative adversarial networks (GANs) was once widely used in early days. Since 2020, diffusion models have become the popular option. Rather than directly training a model to output a high-resolution image conditioned on a text embedding, a popular technique is to train a model to generate low-resolution images or latent space, and use one or more auxiliary deep learning models to upscale or decode it, filling in finer details. Text-to-image models are trained on large datasets of (text, image) pairs, often scraped from the web. With their 2022 Imagen model, Google Brain reported positive results from using a large language model trained separately on a text-only corpus (with its weights subsequently frozen), a departure from the theretofore standard approach.
Datasets
Training a text-to-image model requires a dataset of images paired with text captions. One dataset commonly used for this purpose is the COCO dataset. Released by Microsoft in 2014, COCO consists of around 123,000 images depicting a diversity of objects with five captions per image, generated by human annotators. Originally, the main focus of COCO was on the recognition of objects and scenes in images. Oxford-120 Flowers and CUB-200 Birds are smaller datasets of around 10,000 images each, restricted to flowers and birds, respectively. It is considered less difficult to train a high-quality text-to-image model with these datasets because of their narrow range of subject matter. One of the largest open datasets for training text-to-image models is LAION-5B, containing more than 5 billion image-text pairs. This dataset was created using web scraping and automatic filtering based on similarity to high-quality artwork and professional photographs. Because of this, however, it also contains controversial content, which has led to discussions about the ethics of its use. Some modern AI platforms not only generate images from text but also create synthetic datasets to improve model training and fine-tuning. These datasets help avoid copyright issues and expand the diversity of training data.
… excerpt ends here. Continue reading the full article.
