>>106288122
Because you're lying about the capabilities of whatever model you're trying to shill. Qwen (pic rel) has worse visual artifacts and nonsense objects if I try a LoRA with the same photoreal style. Wan can only do cinematic photoreal style and it would take forever to get an image as sharp as Chroma anyways. Visual artifacts that are typically found on a real camera or photograph aren't a flaw, just a model that captures that kind of detail.

>>106265115
>Everyone seems obsessed with Qwen right now
Because they have been fooled. For art and realism the model is heavily overbaked on the SOTA for API (Seedream 3.0), also has some GPT 4o mixed in there as well. This allows them to lie on benchmarks and pretend Qwen is good. It looks nice to the average user who is not too concerned about realism but wants to compare base models because that's what they know, Flux base (no exposure nor good awareness of Chroma). I'm sure it blows Flux dev out of the water based on having better styles alone. But it's not necessarily technically better. That is the big fat lie that the average user isn't aware of. But for some reason, it's working for a good chunk of them.

It's obvious, that the uncensored model will prevail. Qwen has a lot of work to do (pic rel) if it wants to stand a chance against Chroma, and it's prohibitively expensive to properly train.