Snapchat Launches Its First Free Open-Prompt AI Lens in the US

Snapchat Unveils Free “Imagine Lens” for All Users

Snapchat is excited to announce its first open AI image generation feature, the “Imagine Lens,” now available to all users at no cost. Initially launched in September exclusively for paid subscribers, this new Lens allows everyone to personalize their Snaps using creative prompts or even generate unique images.

Transform Your Snaps with Fun Prompts

Users can unleash their creativity by entering prompts like “Turn me into an alien” after capturing a selfie, or even request whimsical images of a “grumpy cat.” This innovative tool is perfect for exploring Halloween costume ideas or reimagining your friends’ appearances.

Share Your Creations With Friends

Once you’ve generated your masterpiece, you can easily share it with friends or post it directly to your Snapchat Story, expanding your audience beyond the app.

Competing in the AI Landscape

Snapchat’s broader rollout of the AI Lens comes in response to recent launches from competitors like Meta with Meta AI and OpenAI’s Sora, which offer even more advanced AI tools aimed at engaging the younger audience. Sora, for instance, allows users to create videos of themselves after providing a one-time video and audio recording, facilitating shared experiences through AI-generated personas called “cameos.”

A Strategic Move for Snapchat

Recognizing the competitive pressure, Snapchat’s decision to open the AI image Lens to all users appears to be a strategic investment. Previously, this feature was only available to Lens+ and Snapchat Platinum subscribers, but now a limited number of generations are accessible to free users as well.

Global Expansion Plans

Initially, free users in the U.S. can enjoy the new Lens, with plans to expand to other markets, starting with Canada, Great Britain, and Australia shortly thereafter.

How to Access the New Lens

To find the Lens, simply look for it at the front of the in-app Lens Carousel, or search by name. You can edit your prompt by tapping the caption or choose from preloaded suggestions for inspiration.

Lens Engagement Statistics

Snapchat reports that its users engage with Lenses over 8 billion times each day, highlighting the immense popularity and potential of this feature.

Sure! Here are five FAQs regarding Snapchat’s release of its first open prompt AI Lens:

FAQ 1: What is the new open prompt AI Lens on Snapchat?

Answer: The new open prompt AI Lens is an interactive feature that uses artificial intelligence to generate unique visual effects based on user input. By providing prompts, users can create personalized AR experiences and share them with friends.

FAQ 2: How can I access the open prompt AI Lens?

Answer: To access the open prompt AI Lens, simply open the Snapchat app, swipe to the camera screen, and then select the “Lens Explorer” option. You can find the AI Lens in the featured section and start using it for free.

FAQ 3: Are there any costs associated with using the AI Lens?

Answer: No, the open prompt AI Lens is available for free to all users in the United States. You can use it without any subscription or hidden fees.

FAQ 4: Can I create my own prompts for the AI Lens?

Answer: Yes! Users can input their own creative prompts to generate customized AR effects, allowing for a highly personalized experience tailored to individual tastes.

FAQ 5: Is the AI Lens available globally?

Answer: Currently, the open prompt AI Lens is only available for free users in the United States. There may be plans to expand its availability to other regions in the future. Keep an eye on updates from Snapchat for any changes!

Source link

MARKLLM: A Free Toolkit for LLM Watermarking

Title: Innovative LLM Watermarking Techniques for Ethical AI Use

LLM watermarking is a crucial tool in preventing the misuse of large language models, such as academic paper ghostwriting and the spread of fake news. This article explores two main families of watermarking techniques: KGW and Christ, each with unique approaches to embedding imperceptible signals in LLM outputs.

KGW Family: Enhancing Watermark Detection and Removal Resistance

The KGW Family focuses on modifying logits produced by LLMs to create watermarked text. By categorizing vocabulary into green and red lists and biasing the logits of green list tokens, this technique enhances watermark detectability. Improvements include better list partitioning, logit manipulation, and resistance to removal attacks.

Christ Family: Altering Sampling Processes for Unique Watermark Embedding

On the other hand, the Christ Family alters sampling processes during text generation to embed watermarks. This technique aims to balance watermark detectability with text quality, addressing challenges like robustness and increasing watermark capacity. Recent research focuses on refining list partitioning and logit manipulation.

MarkLLM Framework: A User-Friendly Approach to Watermarking

To simplify the experimentation with LLM watermarking frameworks, the open-source MarkLLM toolkit offers intuitive interfaces for implementing algorithms and visualizing their mechanisms. With a comprehensive suite of tools and automated evaluation pipelines, MarkLLM streamlines the evaluation process and provides in-depth insights into the performance of different watermarking algorithms.

Overall, LLM watermarking is essential for the responsible use of large language models, offering a reliable method to trace and verify text generated by AI models. The ongoing research and innovation in the field continue to evolve both the KGW and Christ Families, ensuring their effectiveness in combating misuse and ensuring ethical AI use.

  1. What is MARKLLM?
    MARKLLM is an open-source toolkit for LLM watermarking, which stands for Learned Layer Multiplexing. It is a method for embedding invisible watermarks into deep learning models to protect intellectual property.

  2. How does MARKLLM work?
    MARKLLM utilizes a technique called layer multiplexing, where multiple layers of a deep learning model are jointly trained to embed and extract watermarks. This allows for robust and imperceptible watermarking that can withstand various attacks.

  3. Is MARKLLM compatible with all types of deep learning models?
    MARKLLM is designed to work with a wide range of deep learning models, including neural networks, convolutional neural networks, and recurrent neural networks. It can be easily integrated into existing models for watermarking purposes.

  4. What are the benefits of using MARKLLM for watermarking?
    MARKLLM provides a secure and efficient way to protect deep learning models from unauthorized use or redistribution. By embedding watermarks directly into the model parameters, it ensures that the ownership of the model can be verified and protected.

  5. Is MARKLLM free to use?
    Yes, MARKLLM is an open-source toolkit, which means it is freely available for anyone to use and modify. Users are encouraged to contribute to the development of MARKLLM and share their improvements with the community.

Source link