🤖 Большое сравнение БЕСПЛАТНЫХ нейросетей. Кто лучше в 2024 году? (неожиданный финал)

The speaker, Nikita Bruskov, a web and neuro designer, introduces the video as a comparison of free AI models. He mentions that there was a previous video that included paid models, but since many AI models have updated and improved, they will focus only on the free ones. The AI models to be discussed include Dalle 3, Sberbank's Candinsky, Yandex's Rec AI, Playground A, GRAU, and Tensor Art. The video will compare these models based on several criteria such as price, censorship, additional features, speed, and quality.

The video discusses the limitations of the free AI models, such as the number of images that can be generated per day. It is noted that Dalle 3 has a limit of 15 images per day, while Candinsky and Shedevr (Masterpiece) have no such limitations. GRAU has a limit of 100 images per day, but it generates four images at a time, leading to 25 generations per day. The speaker also provides a subjective rating scale for each model based on the pricing and limitations.

The speaker evaluates the AI models' ability to understand and represent complex mechanisms. He uses a detailed prompt with many elements, such as a bald man in a black suit holding a red phone, to test the models' accuracy. The results vary, with some models accurately representing the prompt and others missing elements or introducing inaccuracies. The speaker also discusses the models' ability to handle complex technical details and provides subjective ratings based on their performance.

The video addresses the censorship and content control features of the AI models. The speaker tests the models' responses to generating images with certain characters and explicit content. It is shown that some models refuse to generate inappropriate content or replace it with more acceptable alternatives. The speaker also discusses the limitations in generating certain types of content and provides his subjective ratings on how well each model handles censorship.

The speaker discusses the control users have over the generated results and the additional tools provided by the AI models. He covers the ability to edit parts of an image, remix features, and other functionalities that allow users to modify the output. The speaker provides a detailed comparison of the tools available in each model and evaluates their usefulness and ease of use, offering subjective ratings for each.

The video focuses on the photorealism of the generated images by the AI models. The speaker compares the models' ability to create realistic human portraits and landscape photos. He provides examples and discusses the quality of the images, noting which models produce the most realistic results. The speaker also mentions the limitations of some models in generating high-quality images and gives his subjective ratings based on the photorealism of the outputs.

The speaker evaluates the AI models' performance in generating realistic landscape images. He uses a specific prompt featuring a view of a forest from above, taken with an 8K drone camera, and compares the results from different models. The speaker discusses the level of detail, the natural appearance of elements like trees and rivers, and the overall realism of the generated images, providing his subjective ratings on the quality and realism of the landscapes.

The video explores the AI models' ability to generate characters and styles in various categories, such as a cute character for a mobile game, a complex fantasy character, and a character in pixel art style. The speaker discusses the models' performance in capturing the essence of different styles and provides examples of the generated characters. He evaluates the quality, style, and overall appeal of the characters, offering subjective ratings for each model based on these criteria.

The speaker concludes the video by summarizing the overall performance of the AI models based on various criteria, including quality, censorship, additional features, and photorealism. He announces the winner of the comparison and expresses his intention to conduct a future battle between Craft (Крафт) and RNN, questioning which one, free or paid, would perform better for specific tasks. The speaker thanks the viewers for watching and encourages them to subscribe for more content.