Following a significant funding round that secured $140 million for the company, generative AI platform fal.ai has released a powerful new tool to the public. The company has open-sourced FLUX.2 [dev] Turbo, an image generation model engineered for extreme speed without compromising visual quality.
The Efficiency Breakthrough
The defining characteristic of the new Turbo model is its dramatic reduction in necessary compute power. Through the use of a lightweight LoRA adapter, fal.ai has successfully cut the required inference steps down from 50 to just eight.
Remarkably, this speed increase does not result in degraded output. According to fal.ai, the Turbo model matches or even exceeds the quality of the original FLUX.2 through a process called custom DMD2 distillation.
Also Read: Claude Is Now Inside Your Browser | Here’s What Changes for Everyday AI Users
Dominating Benchmarks
Independent testing has confirmed the model’s impressive capabilities. It has risen to the top of open-weights models, achieving an ELO score of 1,167 on the Artificial Analysis benchmark.
Furthermore, it now leads Yupp’s leaderboard for both speed and cost efficiency. The model is capable of generating a high-resolution 1024×1024 image in just 6.6 seconds.
Availability and Use Cases
The model is already drawing significant attention for its stunning results in standard text-to-image tasks, as well as complex edits like transforming comic book visuals into live-action scenes.
FLUX.2 [dev] Turbo has been released on the Hugging Face platform under a non-commercial license.
Also Read: ChatGPT Launches Its Own Year-End “Reflections” Feature
