At 1 a.m., the Star count for Black Forest Lab's GitHub repository started to surge—after a year, Flux.2 released all code, weights, and ComfyUI workflows into the public domain at once. Within ten minutes, the first test images flooded Twitter: generating a 4MP fashion photo took 8 seconds and cost $0.003, while Google's Nano Banana Pro required a $3000 enterprise key for a 2K sample. The comments were all the same: "Google, you're way too expensive."

image.png

Flux.2 wasn't just a single shot—it was a four-pronged attack: the [pro] version cut the sampling steps to 8, improving speed by 2.3 times over the previous generation; the [flex] version opened up temperature, CFG, and noise scheduling, allowing developers to adjust the balance between quality and latency themselves; the [dev] version reduced parameters by 40%, making it possible to run on a 3060 laptop; and the upcoming [klein] version is even more aggressive, claiming to directly output 2K images on a Raspberry Pi-level edge device. In short: Google only sells one ace, while Black Forest throws out four straight flushes.

This time, fine control has been taken to the "pixel level": upload a pose stick figure and then overlay a reference leather jacket, and Flux.2 can map folds, zippers, and highlights onto the new generated model in one go. Fingers are no longer twelve, and background text is no longer nonsense. One user fed [flex] with character sketches from Cyberpunk 2077 and generated 12 4MP posters in three minutes, which were directly sold on Redbubble, equivalent to an hourly wage of $500.

Of course, Google still holds a fortress: when prompts include complex semantic layers like "Van Gogh landscape + rainy night neon + long exposure car flow," Nano Banana Pro still offers smoother color transitions and higher detail density. But the gap has shrunk to a "eye-picking" level, while the price difference is two orders of magnitude—clearly deciding the outcome for budget-limited independent creators, e-commerce small sellers, and students.