The wait is over?
I don't really want to complain and I'm more than grateful for the zimage model... but are you also considering switching to Flux2Small9b? All this waiting is just annoying. We still haven't heard anything about when the basic model will finally be released, and there's no news anywhere. When I think about how long we've been waiting and don't even have a clue whether the model will be released at the beginning of the year, in the middle, or at the end... no information is being disclosed and we're all training on the turbo model. I tested flux2 small 9b today and the quality and speed of the distilled model in particular is overwhelming. The reference photo method is absolutely amazing. you can edit photos so realistically in a way that no other model has been able to do before. Qwen Image Edit is worlds behind and doesn't even come close to the quality and detail, etc. They released distilled and basis at the same time, and we'll probably be able to build decent loras tomorrow. So I honestly ask myself, why wait any longer when you now have such a great model with a reference photo function, abnormally good realistic quality, distilled, fast? I think the developers of Flux2 knew that everyone was waiting for the base model from zImage and have now given us what we've been waiting for since November :) EDIT: Today, January 18, 2026, official LoRa support for Fluxklein 9B and 4B is available after two days. UPDATE: The Lora training works so well with 9b and 4b that it makes absolutely no sense to wait for zimage base anymore. I've also deleted zimage and won't be using it anymore because Flux2-9b is simply so much better. Even the training is really easy and good, especially fast, it's crazy :) I'm so grateful to Black Forest for giving us this model so we no longer have to wait for zimage.
This is TOOOOOOOOOOO late and slower than slooooooooooooooooooth.
Flux klein edit is so good,much better than qwen.doubt z image edit can match or beat it.
@gt123 Absolutely, it's almost perfect. I also think that zimage edit won't be able to keep up, and qwen has probably already been deleted by 80% of people :-D . Qwen has really become more than just useless since Flux2klein.
Tell me you haven't actually tried Flux2Klein without telling me you haven't tried Flux2Klein. (that sounded better in my head)
Not sure what shrooms you guys are on, Flux2Klein is absolutely abysmal compared to Z-Image and Qwen 2511. It's no competition, it's not even in the same league. Flux2K has no understanding of anatomy, it is extremely censored and prompt restrictive. It has some extremely overtrained vectors which will lead some prompts to completely override others making it much less versatile. 9b is also much slower and more ram hungry than Z-image. Edit is also very restrictive and much, much worse than Qwen.
Not to glaze Z-image here, it also has a plethora of limitations that make it hard to work with in some scenarios, but I'd be very surprised if their Edit model will turn out worse than Flux2Klein.
That's why you closed the discussion? Worried people will agree with me? I encourage you to actually try Flux2 Klein, you'll see. It's not great, certainly not better than Z-Image-Turbo. I actually invite anyone to try it: Generate an image of a young woman with black hair, make the prompt as elaborate as you like. In both Flux2K and ZiT, then change the prompt to white hair. See for yourself how overtrained some vectors are in Flux, it's awful and makes working with it very unfun.
flux 2 klein is the next sdxl, assuming z-image doesn't instantly dethrone it with base's superior lora training/ finetuning capability. Competition is good for all sectors. Don't be a sperg, InvictusCreations.