All articles tagged with #text to image
Microsoft has announced its first in-house developed text-to-image generator, MAI-Image-1, which excels at photorealistic imagery and has already ranked in the top 10 on the AI benchmark site LMArena. The model aims to produce faster, high-quality images and is part of Microsoft's broader AI product suite, with a focus on safety and responsible use.
Snapchat has introduced the 'Imagine Lens,' a new AI-powered feature for Snapchat+ subscribers that allows users to generate and edit images using text prompts, enhancing their AR and creative experiences.
Google has released its advanced text-to-image model, Imagen 4, for free on AI Studio for a limited time, offering significant quality improvements over previous models, with a premium Ultra version also available for better instruction-following results.
Stability AI has unveiled Stable Cascade, a new image-generating AI model that promises faster and more powerful performance than its predecessor, Stable Diffusion. The model offers various text-to-image editing features, including inpainting, outpainting, and canny edge, and is available for researchers on GitHub. Unlike Stable Diffusion, Stable Cascade consists of three different models using the Würstchen architecture, resulting in faster processing and improved aesthetic quality. However, the company faces legal challenges over copyright issues and has started offering commercial licenses to fund its research.
OpenAI has announced the release of DALL-E 3, the latest version of its text-to-image generator. This update incorporates integration with OpenAI's AI chatbot, ChatGPT, allowing users to generate images using natural language prompts. The integration works both ways, enabling users to generate an image and then ask ChatGPT to write text about it. DALL-E 3 also promises improved accuracy in interpreting context and overall image quality. OpenAI has added safety options, such as blocking the use of public figures' names in prompts and allowing artists to opt out of future versions of the model. The release will be staggered, with ChatGPT Plus and ChatGPT Enterprise users gaining access in October, followed by research labs and the API service.
OpenAI is set to release DALL-E 3, an improved version of its text-to-image AI system, which can generate results within the ChatGPT app. The new iteration integrates with ChatGPT to help users write detailed prompts for the image AI. DALL-E 3 is better at understanding user intentions and can create elements that previous AI generators struggle with. It also includes enhanced security measures and safeguards against explicit or hateful images. OpenAI plans to release DALL-E 3 next month to select customers, with wider availability to be announced later.
Adobe has launched a beta version of its text-to-image model Firefly, which is built into Photoshop and allows users to slowly build an image with prompts and stock elements. Firefly is not limited by resolution restrictions and matches the quality and resolution of the image being worked on. Adobe hopes that Firefly, its copyright-safe generative AI model, can help the company get its groove back after its $20 billion Figma acquisition was imperiled.
DeepFloyd IF, a new AI model, can create images from text prompts using advanced text-to-image generation techniques. The model was trained on a dataset of over a billion images and text and requires a GPU with at least 16GB of RAM to run. In other news, Apple introduced 20 new games to Apple Arcade, and down rounds are no longer seen as a failure for founders.
DeepFloyd, a research group backed by Stability AI, has unveiled DeepFloyd IF, an open-source text-to-image model that can integrate text into images. Trained on a dataset of over a billion images and text, DeepFloyd IF uses multiple different processes stacked together in a modular architecture to generate images. The model is particularly good at understanding complex prompts and even spatial relationships described in prompts. However, the model may suffer from biases, and its base model doesn't generate images that are quite as aesthetically pleasing as some diffusion models.