[Feature] Do you have plans to support flux.2?

[102286545]

Checklist

• 1. If the issue you raised is not a feature but a question, please raise a discussion at https://github.com/mit-han-lab/nunchaku/discussions/new/choose. Otherwise, it will be closed.
• 2. I will do my best to describe the issue in English.

Motivation

Do you have plans to support flux.2?

Related resources

No response

[zwukong]

淘汰了已经😄

[nitro79sf]

Agreed - Flux Dev 2 support would probably be more valuable for faster generation,
as opposed to ZImage turbo, which is already fast. :-)

[pirateboogie]

Flux 2 dev is hands down the best open source model available at this time, both quality and knowledge\context understanding (prompt adherence) wise it's only real rival is Qwen edit 2511 and image 2512 but both are not equal to it, just a tad faster and have more useful task oriented loras due to the easier training (i guess) but still Flux2 has some good ones as well .
Speed wise they are actually nearly equal - Flux2 dev generation is not much slower than Qwen (either of them) especially at resolutions higher than 1mp - flux2 rocks at full hd and higher res generations and edits and it's recently released turbo lora is NOTHING like the 4 steps, quality crapping garbage currently available for recent Qwen models too - it is decent and works rather well without too much quality and prompt adherence sacrifice.
Yet, of course, if it can be Nunchakoed ))) and if the same can be expected of it's Nunchaku version as it is with Qwen and Flux1 models compared to their regular versions, it could become even better ). Imho Flux2 is a good and truly useful tool (Qwen IE 2511 too btw))) rather than a toy like Zimage Turbo which might dazzle with it's speed but has nothing else besides that.

[jarkevithwlad]

+1

[github-actions]

This issue has been automatically closed due to 30-day inactivity. Please feel free to reopen it with /reopen if needed.