AI images just got WAY too real. FLUX 1.1 deep dive

The video script introduces Flux 1.1 Pro, an advanced AI image generator that has recently been released and is claimed to be the best in the market. The speaker plans to compare it with other leading image generators and provides benchmark scores for viewers interested in the technical aspects. The platforms where Flux 1.1 Pro can be used are mentioned, including Together AI, Rubber Band, Ever Art, Foul, Replicates, and Free.com, each with different credit systems. The speaker chooses Together AI for demonstration due to the free $5 credits offered. The script also explains how simple the process is, requiring only a positive prompt without complex settings like CFG scale or negative prompts. The video demonstrates the generation of images starting with a simple prompt like 'a woman in the city' and discusses the typical AI-generated image characteristics, such as polished faces and blurry backgrounds. A trick to generate more realistic images is shared, which involves using specific file format prompts like 'IMG uncore' followed by numbers and 'CR2', simulating a Canon camera file structure to produce images that mimic amateur photography.

The paragraph delves into the ability of Flux 1.1 Pro to generate photos that closely resemble amateur, non-AI generated images. The speaker tests the generator with various prompts, including 'selfie' and 'party', and observes the resulting images' graininess and lack of professional polish. The script highlights how these images can be made to look more realistic by using specific file format prompts like 'heic' and numbers, which can produce photos that are hard to distinguish from real ones. The video also demonstrates how adding simple prompts can yield surprisingly normal-looking images, such as a cat or a selfie, and how the generator handles more complex prompts, showing that the effectiveness of the file format trick diminishes with longer or more detailed descriptions.

This section of the script focuses on testing Flux 1.1 Pro with more complex and challenging prompts, such as generating images of a woman doing a warrior 1 yoga pose at home. The speaker notes that Flux 1.1 Pro struggles with understanding the pose and anatomy, but compares it favorably to other top image generators like Ideogram version 2 and Mid Journey version 6.1. The video shows side-by-side comparisons of the same prompts across different generators, allowing viewers to assess the realism and accuracy of each. The speaker also comments on the occasional flaws in hand and finger generation, despite Flux Pro's generally good performance in these areas.

The paragraph discusses the generation of low-quality selfies using Flux 1.1 Pro and compares it with other image generators. The speaker tests the generator with a prompt for a low-quality Snapchat photo of a teenage man taking a mirror selfie, shot on a phone and posted in 2015. The result is praised for its graininess and authenticity, resembling a real phone selfie. The video presents a side-by-side comparison with other generators, with Flux 1.1 Pro and Flux 1 Pro appearing quite similar, and Mid Journey struggling with the mirror selfie concept. The speaker expresses a preference for Flux 1.1 Pro's generation over Flux 1 Pro for this prompt.

This section tests Flux 1.1 Pro's ability to generate images of hands making a heart symbol and a woman showing her palms and soles of her feet. The speaker notes that while the generator performs well with the heart symbol prompt, it struggles with the complexity of showing palms and soles of feet, resulting in images that do not meet the prompt's requirements. A comparison with other generators shows that Ideogram comes closest to the prompt, while both Flux generations and Mid Journey fail to accurately represent the requested images.

The script explores Flux 1.1 Pro's capabilities in generating different art styles, such as watercolor paintings and anime. The speaker is impressed with the generator's ability to create a watercolor painting of a whale in the sky, with Flux 1.1 Pro producing the most accurate watercolor effect and whale representation. When testing anime generation, the speaker notes that while Flux 1.1 Pro produces a good image, the text on signs remains unrealistic, an issue common to all current image generators. The video compares the results with other generators, with Ideogram having a more anime-style background, and Mid Journey's result being less anime-like and more abstract.

The paragraph discusses the difficulties Flux 1.1 Pro faces in generating images of uncommon animals, using the example of a kodo dragon. The speaker notes that Flux 1.1 Pro fails to generate a kodo dragon that resembles the real animal, which is expected due to the lack of such images in its training data. A comparison with other generators shows Mid Journey as the clear winner in generating a kodo dragon, although it still fails to accurately depict the forked tongue. The speaker also attempts a trick with a date and file format prompt, but it does not yield a realistic kodo dragon image.

This section summarizes the performance of Flux 1.1 Pro and provides updates on its capabilities. The speaker references a third-party evaluator's leaderboard, which ranks Flux 1.1 Pro highest with a win rate of 69%. It is revealed that Flux 1.1 Pro, previously known as the mysterious 'blueberry' model, outperforms all other models. The video highlights that Flux 1.1 Pro generates images six times faster than Flux 1 Pro with improved quality and diversity. It also mentions that Flux 1 Pro has been updated to generate images twice as fast as before. The speaker anticipates that Flux 1.1 Pro will soon be able to generate ultra-high-resolution images at a lower cost than Flux 1 Pro. The video concludes by encouraging viewers to try the CR2 and JPEG prompts for realistic, amateur-looking photos and to share their experiences and techniques with Flux 1.1 Pro.