Show HN: FLUX.3 Video 1k generation test (180 mins of footage)

  • Posted 2 hours ago by niwrad
  • 1 points
https://www.martini.film/p/35b02ee5-2488-4dc6-a9d9-c403211cfede/flux-3-15ryn2n
I've always been a fan of Black Forest Labs' FLUX image models. Specifically how "un-AI" their look is. (One might even describe it the opposite of Midjourney which feels like it excels at a surreal look.)

Seedance 2.5 is likely taking the spotlight today but I wanted to do a deep dive into FLUX.3 so it can get some attention.

My general observation is that FLUX.3 video is the most "clay-like" video model I've played with.

Specifically: - Adding and rerunning video prompts feels additive/subtractive and not completely random generations (I feel much more in control of the medium). - "Draft" and "Enhance" modes (rerunning generations with the same seed) feel like getting to work with clay before committing in a kiln. Certainly a faster and cheaper way of working with generative video.

Excited to share results here.

The canvas has all generations with the prompts included.

Spent ≈$1,000 on this test.

Disclosure: I'm the co-founder of Martini (W26), which is where I ran the test

0 comments