AI NEWS    25 Nov – 9 Dec

 

1. Hedra Ai has received a major update. Now the neural network can generate 12-minute videos in a single rendering – in fact, a full-fledged short film. There is also a new noise removal system (sound quality will become much better and clearer) and the ability to clone and create new voices.
2. A neural engine has been added to expand the boundaries of video. Runway Gen-3 can now adjust the video to any aspect ratio. AI doesn’t just finish the frame but also brings it to life by adding realistic zooms and camera movements.
3. AI video editing in two clicks. The result is the AI Video Composer neural network, which works very simply: we upload images/videos/audio, and then write a promo with a wish. The service will do everything by itself in a couple of minutes, leaving the user only to make corrections.
4. OtioAI’s learning neuron has become even more powerful. Otio AI is now many times better than ChatGPT. It can quickly analyze large materials such as books, lectures, and research and create detailed summaries with key points. The neural also finds trusted sources on any topic and provides links to scientific theses. It will also edit your texts to the level of a professor in a minute.
5. Online fitting room Kling. Virtual Try-on is a tool for trying on clothes on a model, you just need to upload a photo of the item and the model. The neural network can also animate the finished generation. Regular generation costs 10 credits, while animation costs 20.
6. Stable Diffusion 3.5 update was released. The neural network now has three new features: Blur: which raises the quality of images up to 8-16k, Canny: which analyses the contour of images to create a full-fledged photo, Depth: allows you to control the composition with the help of new depth maps.
7. NVIDIA introduced a neural network for creating 3D models in 4K. The Edify 3D neural network generates 3D models from textual descriptions, providing high resolution, detailed geometry, and high-quality textures.
8. Improving the quality of photos with InstantIR. We found a useful feature from Huggingface that can restore even the most shaky photos. To process it, you need to upload the original photo and select the settings: the number of editing steps, quality, and prompt as desired. The speed depends on the settings; on average, the rendering takes a minute.
9. AI will choose a gift for anyone. The Lutra service searches for a person on the Internet analyses their interests, and offers gift ideas with stores where they can be bought at a favourable price. Optionally, you can add additional information about the person.
10. The new neural network from Vrai AI allows you to verify the authenticity of any luxury item in seconds. The algorithm has been trained to recognize even the most subtle differences between the original and a fake. For example, it can detect incorrect seam placement or poor-quality material.
11. Minimax works wonders. The developers have shown access to the new model – it can bring static images to life and does a great job with comics and photos. There is an unlimited free tariff but with a rather long rendering time.
12. Meta decided to interrupt OpenAI’s announcements and released LLaMa 3.3 70B. This is a revolution: the medium-sized model (70B) is almost as good as, and sometimes even better than, the giants at 405B and even GPT-4o.
13. The new Photon image generator. The developers from Luma AI claim that the image quality outperforms competitors (Midjourney V6, Flux 1.1 Pro, Recraft V3, and Ideogram V2), and the generation price will be lower. The company is also going to release a faster and updated version of Photon Flash in the near future.