#flux2

We’ve decided to still post Flux 2.0 fanworks tomorrow in case anyone needs a distraction. We post to Ao3 at 11am GMT/6am EST ✨

forgot to post yesterday but happy first birthday to flux 2.0, the full-length novelisation of doctor who s13 💙💙

eight authors (plus an amazing team of editors and artists), 160k+ words, and so much heart.

forever proud of @the13archives team and all we accomplished last year. ✨

Now that I’ve spent some time actually playing with Flux 2 Dev, my feelings about it are… complicated.

My first impression was pure wow. The model follows references insanely well, it listens to the prompt like a very well-trained golden retriever, and it can generate exactly what you ask for with almost no artifacts. As long as you’re clear about what you want, from which reference, and in what style, Flux 2 Dev just delivers. It understands you frighteningly well.

But that same obedience turned out to be a downside for me.

I suddenly realised I’d lost that “Christmas present” feeling I usually get from image generation with Illustrious/Flux 1 Dev+LoRas. With Flux 2 Dev, the model is so precise and so loyal to the reference that it almost always gives me exactly the character I asked for — with barely any deviation, barely any “happy accidents”.

I’m used to generation as a kind of game with unpredictability, almost like neural network tarot:

I put in an intention and then see what the universe-model sends back.

Flux 2, on the other hand, feels like it’s saying:

“No, sweetheart, we’re not divining the future here.

You asked for this guy — here he is: stable, high-quality, and instantly recognisable. Almost. Every. Single. Time.”

And that feels like less magic of randomness,

more production of polished but predictable content.

In other words, I’ve lost a piece of that little ritual of surprise — and for a creative person, that’s actually a big part of the fun.