Black Forest Labs, the studio behind the Fluxfamily of AI image generators, recently unveiled the latest addition to their lineup – Flux 1.1 [Pro]. This new release comes only two months after the debut of the original models which included Flux 1 Pro, a closed-source model with cutting-edge capabilities, Flux 1 Dev, an open-source noncommercial model, and Flux Schnell, a completely open-source model.
The Flux models marked a significant advancement in generative AI technology with their text generation capabilities, prompt adherence, and overall image quality. Even the smaller models like Flux Dev and Flux Schnell delivered results comparable to those from MidJourney and far superior to outputs from SD3, Stability’s highly anticipated evolution over SDXL, which unfortunately fell short of expectations.
The latest model has already made a noticeable impact by securing the top Elo score in the Artificial Analysis image arena, a leading benchmarking platform for AI models. It has outperformed all other text-to-image models on the market while maintaining a speed comparable to its smallest model.
The graph below illustrates the Elo score (image quality) on the Y-axis and the generation speeds on the X-axis. MidJourney enthusiasts may notice the absence of their model—it’s so slow that it’s literally off the chart. Nonetheless, its Elo Score hovers around 1100 points, just below Ideogram V2.
Elo scores for AI image generators. Image: Black Forest Labs
The new Flux Pro stands out with its pricing, offering Flux 1.1 Pro at $0.04 per image, lower than many other models in the market, including the original Flux 1 Pro. This pricing strategy positions it as a strong competitor against other paid services like MidJourney and Ideogram, which are priced at $96 and $84 per year respectively. Additionally, MidJourney and Ideogram are slower and have a higher cost per token.
Unfortunately, Flux 1.1 Pro cannot be run locally. Unlike its less powerful open-source counterparts such as FLUX1 [Dev] and FLUX1 [Schnell], this new pro version is a closed-source model, restricting users to access it through platforms like Together AI, Replicate, Fal AI, and Freepik. It cannot be customized or fine-tuned.
For those interested in trying out the model, some platforms offer free credits for initial generations, but once those are used up, Freepik stands out as the best service according to our criteria. Its Mystic workflow significantly enhances generations by providing higher details and improved aesthetics.
IT’S FINALLY HERE!
🔥 Freepik Mystic 🔥
“Any sufficiently advanced technology is indistinguishable from magic.” — Arthur C. Clarke ✨ Mystic is the most advanced AI generator to date with outputs directly in Full HD.
But what’s really Mystic? Let’s dive in 🧵👇 pic.twitter.com/nrlPTi0OWo
— Javi Lopez ⛩️ (@javilopen) August 27, 2024
There have been no announcements regarding an open-source version 1.1 of the FLUX1 [Dev] or FLUX1 [Schnell] models, but it’s evident that Black Forest Labs is dedicated to developing high-quality models for image and video creators.
Hands-On Testing and Review
We put the new Flux model to the test and the results were satisfactory. While it may not represent a revolutionary leap like the transition from SDXL to Flux, it is certainly a welcomed upgrade.
It is highly realistic overall, showcasing excellent text generation capabilities and creativity in artistic tasks and styles. The model is versatile, offering fast generations without compromising quality.
Realism
Prompt: “Polaroid photo with VSCO filter, 1990, woman, night, flash photo, blonde, young face, beautiful shadows, tropical plants, inside an apartment, DSLR, camera flash, holding a handwritten sign on a notebook saying ‘Verification for Decrypt October 7, 2024.’ The woman is doing the peace sign with her other hand.”
Test images generated with Flux 1.1 Pro
The model excels at producing realistic images, surpassing the airbrushed appearance of the initial Flux models. While not flawless, the results are highly convincing, especially with appropriate prompting. At first glance, these images—both generated with Flux 1.1 Pro—could easily be mistaken for real without scrutinizing minor details.
The text is in line with the prompt, and hand rendering has improved, though not perfect. It’s worth noting that these are not hand-picked samples but the initial two generations. Typically, working with generative AI yields better results after several iterations and adjustments.
The lighting mimics a camera flash, focusing on the subject without flooding the entire room with light. The VSCO filter enhances realism, and the model demonstrates exceptional prompt adherence.
Comparing Flux 1.1 to Flux 1 reveals comparable realism at first glance. However, when provided with the same prompt, the new model delivers a more natural pose and a consistent body structure. For example, Flux 1 generated what appeared to be
