Been using FLUX.1 for a while now, let me talk about the details and how it handles prompts.

The Forge / 工坊 Models / LoRA / 模型与 LoRA
1

Black Forest Labs is basically the original SD team. After FLUX.1 went open-source, I ran it locally for a while. The most obvious thing is the detail density—when generating high-res landscapes or complex textures, the clarity is noticeably a tier above the models I used before.

Another thing is it understands prompts way better. That frustrating “I clearly wrote A but it gave me B” situation happens way less now, saving me a ton of time rewriting prompts.

Output diversity is decent too—run the same prompt a few times, and the composition varies more than I expected. Downsides? It eats VRAM; machines with lower specs really struggle. Overall, I’m using it as my go-to model now.

2

Yeah, VRAM is definitely the bottleneck here. You can pretty much forget about it with 8GB.

3

Yeah, the prompt adherence improvement is real. I’ve noticeably had to reword things way less often.

4

The schnell version runs fast and the quality is good enough.

5

BFL is literally the original SD team, the pedigree is there.

6

The details are great but sometimes it looks too “digital”

7

For local runs, you really need at least 24G to be comfortable.