Stable Diffusion v2 - by Jesus Rodriguez
Stable Diffusion has been one of those few machine learning (ML) models that have transcended to mainstream culture. A few months ago, Stability AI shocked the ML community by open-sourcing Stable Diffusion, taking a different approach from big AI labs like OpenAI, Google Brain, and Meta AI. Since its initial release, Stable Diffusion has become the most widely used generative AI model, with applications across different domains. Just three months after the original release, the significant level of adoption, together with the rapid iteration native of open source distribution models, has led Stability AI to release a second version of Stable Diffusion.
Stable Diffusion v2 is a significant upgrade to its predecessor. The new version was trained using a new text encoder called OpenCLIP, which improves the quality of images relative to the previous latent diffusion encoder. Also related to image quality is the addition of an Upscaler Diffusion model that enhances the resolution of output images by 4x. Depth2Img is another interesting addition to Stable Diffusion that can infer depth from an input image and represent that in the generated outputs. The new release also includes a text-guided inpainting model that simplifies the experience of modifying parts of a given image.
Stable Diffusion v2 is another release that pushes the boundaries of the generative AI space. The speed and depth of innovation in this area of deep learning are nothing short of remarkable.
🔺🔻TheSequence Scope – our Sunday edition with the industry’s development overview – is free. To receive high-quality content about the most relevant developments in the ML world every Tuesday and Thursday, please subscribe to TheSequence Edge 🔺🔻
🗓 Next week in TheSequence Edge:
Edge#247: we classify ML interpretability methods; explore the building blocks of interpretability by Google Research; explain TensorWatch, an open-source framework for debugging ML models.
Edge#248: we dive deep into a fascinating topic: Foundation Models that create the industrial era of AI.
In these uncertain times for Twitter, we’d like to introduce TheSequence’s LinkedIn account. We are building a unique resource and support system for all ML&AI aficionados. Let’s connect!
Now, let’s review the most important developments in the AI industry this week
CICERO
Meta AI published a paper detailing CICERO, an AI agent able to negotiate and cooperate in conversational environments →read more
Interactive Agents in Video Game Environments
DeepMind published a paper introducing a framework for AI agents that can follow human instructions in open-ended video game environments →read more
NEVIS 22
DeepMind published a paper introducing NEVIS 22, a new benchmark to evaluate the tradeoff between efficiency and performance in large computer vision models →read more
Neural Attentive Circuits
Amazon Science published a paper proposing Neural Attentive Circuits, a technique that trains a neural network to mimic the structure of a causal network →read more
Stable Diffusion 2.0
Stability AI unveiled the second release of its groundbreaking Stable Diffusion models with all sorts of interesting quality improvements →read more
TensorFlow 2.11
TensorFlow released a new version that includes improvements to DTensor, a new embedding utility for Keras called StructuredTensor and the completion of the migration of the Keras Optimizer →read more
Airbnb Categories
Airbnb discusses the ML behind its Categories feature to improve search →read more
Conversations Summaries in Google Chat
Google Brain discusses the techniques used to implement conversation summaries in Google Chat →read more
AI-powered
SaaS startup Taktile raised a $20 million Series A investment round co-led by Index Ventures and Tiger Global. Hiring remote.
Insights company QualSights raised a $7.7 million Series A funding round, led by 4490 Ventures. Hiring in Chicago, US.
Lawtech startup Harvey raised $5 million in a funding round led by the OpenAI Startup Fund. Hiring in San Francisco, US.
Healthtech platform RedBrick AI raised $4.6 million in a seed funding round led by Sequoia India and Southeast Asia’s Surge. Hiring in India.
Data-platform-as-a-service startup Keboola raised a $4.5 million seed financing round led by Presto Ventures. Hiring globally.
ncG1vNJzZmismJrApr3UnqWcnV6owqO%2F05qapGaTpLpwvI6sq5qanJp6pbXFn6ysoZ%2Bjerd%2B