Any plan to make a flux2 klein aio model?
Now that flux2 kelin is out, I did some test, it's really fast. With 4b distilled model fp8 version, it's 5 second/image on my pc, compare to 40 seconds/image using 2511. Any plan to make an aio pack for it once there are loras out?
Will probably take some time for loras to come out to even make a AIO version
Perhaps a full finetune would be possible? The model is small compared to qwen edit.
I don't want to maintain multiple models that ultimately do the same thing, and Qwen Image Edit still appears to be better quality and more versatile, although slower (but still within seconds). Even if I did Flux2 Klein, I feel I'd still need Qwen Image Edit anyway for more quality or complex situations.
I don't want to maintain multiple models that ultimately do the same thing, and Qwen Image Edit still appears to be better quality and more versatile, although slower (but still within seconds). Even if I did Flux2 Klein, I feel I'd still need Qwen Image Edit anyway for more quality or complex situations.
bro actually I noticed in a few scenarios, klein-9b-distilled-fp8 is better than QEI2511
I don't want to maintain multiple models that ultimately do the same thing, and Qwen Image Edit still appears to be better quality and more versatile, although slower (but still within seconds). Even if I did Flux2 Klein, I feel I'd still need Qwen Image Edit anyway for more quality or complex situations.
bro actually I noticed in a few scenarios, klein-9b-distilled-fp8 is better than QEI2511
I've heard mixed reviews at best... we'll see how community support plays out. I also hear the license is pretty bad with the 9B model. I only want to "shift gears" when it is clearly a full replacement and better option.
I don't want to maintain multiple models that ultimately do the same thing, and Qwen Image Edit still appears to be better quality and more versatile, although slower (but still within seconds). Even if I did Flux2 Klein, I feel I'd still need Qwen Image Edit anyway for more quality or complex situations.
bro actually I noticed in a few scenarios, klein-9b-distilled-fp8 is better than QEI2511
I've heard mixed reviews at best... we'll see how community support plays out. I also hear the license is pretty bad with the 9B model. I only want to "shift gears" when it is clearly a full replacement and better option.
Positive points about Flux.2-Dev I do prefer the Flux.2-Dev model over Qwen-Image-Edit it supports up to 10 reference pictures and you can even provide it 10 reference pictures of a human subject and it can use all 10 images to reconstruct that subject. Qwen Image Edit usually only supports 1 reference picture of a human subject it does not understand or well enough multiple reference of the same subject. Also Flux.2-Dev has better image quality than Qwen-Image-Edit.
Positive points about Qwen-Image-Edit is faster than Flux.2-Dev for most people if you have 16GB VRAM or lower. It probably has more loras for SFW and NSFW than Flux.2-Dev.
I have not tested Flux.2-Dev Klein but if it is at least close enough to do as well Flux.2-Dev I would prefer an AIO for that.
Flux 2 Dev is undeniably impressive from a pure image-quality standpoint, especially for photorealism. No argument there.
That said, for anyone thinking beyond hobby use, the licensing is the elephant in the room. Flux 2 Dev is non-commercial by default, and the moment you want to repurpose outputs in a product, service, paid platform, or even a gray area (NSFW monetization, packs, commissions, Patreon, etc.), you’re stepping into licensing negotiations. That alone is a hard stop for a lot of artists and developers.
Qwen Image Edit (2511), on the other hand, is Apache-2.0. Fully open, commercially usable, no ambiguity. You can build on it, fine-tune, bundle, ship tools, sell outputs, and sleep at night. For creators who care about long-term viability, that matters more than squeezing the last 5–10% of visual polish.
On prompt handling, Qwen is often more faithful and deterministic. Flux can be stunning, but it sometimes “freestyles” in ways that aren’t ideal when precision or repeatability matters. For workflows that rely on controlled edits, consistency, or product-style outputs, Qwen behaves more like a tool and less like a mood swing.
About the common complaint that Qwen struggles with multiple reference images: in practice, it’s more about workflow than capability.
What works reliably:
• Use high-resolution, sharp references. Soft or low-res inputs hurt it badly.
• Don’t try to merge everything in one pass.
Generate a clean subject first (ideally with no background) using a single reference.
• Then iterate: reuse the generated image as a new reference and introduce the second character/object.
• Repeat as needed. Qwen rewards staged composition rather than brute-force prompting.
It’s slower, but far more controllable.
So yes — Flux 2 Dev may be “better” in raw output today. But unless the team (or anyone building an AIO) is willing to explicitly deal with licensing constraints and accept that many users won’t be able to monetize their work, it’s a risky foundation.
For artists and developers who want freedom, reuse, commercial safety, and open ecosystems, open-source models like Qwen are simply the smarter long-term choice. That’s why a lot of people are sticking with it — or moving toward it — despite Flux’s undeniable visual strengths.
I would stick with Qwen unless another open source will come out with better options and results
.