Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Flux 3

bfl.ai

Post reply on HN
Print view View original

141–150 of 151 posts

Re: Flux 3

#141

Post by jarjoura »

jdthedisciple wrote: post #20

It's incredible how negative and dismissive the comments in here are while here I am thinking the model actually looks impressively capable. But then again I heard the downers have always been the first to leave their dung comments here so let's see...

Given that Martin Scorsese is now an "advisor", I suspect BFL will continue to position itself as the research lab for filmmakers. This video model, if comparable to Seedance 2.0 won't really make sense as a slop generator, as it's too expensive for that, but it would be part of a projects VFX pipeline.

Re: Flux 3

#142

Post by potsandpans »

vunderba wrote: post #134
potsandpans wrote: post #133

Earlier quoted context omitted.

Flux2.dev and even klein 9b are extremely close to sota. People who are saying otherwise probably haven't used them very much.

I run a fairly high-traffic site for generative image models focusing on complex prompt adherence. Flux.2 doesn’t score anywhere near SOTA proprietary models. Klein 9b is decent for image-to-image, but when used for pure generative purposes brings back SDXL levels of body horror (have some Gattica pianists). If you can get past the annoying JSON structuring, Ideogram 4 is probably the best option in the open‑weights…

I suppose "close" could be interpreted subjectively.

I use flux2.dev on my 5090 with a prompt upsampler (that I host on my 3080ti, it's been so long since I looked at it, its either some qwen or Mistral model). I use the upsampler to produce json structure based on this prompting guide: https://docs.bfl.ml/guides/prompting_guide_flux2

For being able to run on my 5090, it produces incredibly consistent and stable results that work for my use case so well, I find myself reaching for it over proprietary models.

I'm always interested to see objective comparisons though. It's clear from yours that flux lags. But the fact that I can use it locally over closed source sota models is very compelling (to me).

You've convinced me to checkout ideogram4 though!

Re: Flux 3

#143

Post by vunderba »

potsandpans wrote: post #142
vunderba wrote: post #134

Earlier quoted context omitted.

I run a fairly high-traffic site for generative image models focusing on complex prompt adherence. Flux.2 doesn’t score anywhere near SOTA proprietary models. Klein 9b is decent for image-to-image, but when used for pure generative purposes brings back SDXL levels of body horror (have some Gattica pianists). If you can get past the annoying JSON structuring, Ideogram 4 is probably the best option in the open‑weights…

I suppose "close" could be interpreted subjectively. I use flux2.dev on my 5090 with a prompt upsampler (that I host on my 3080ti, it's been so long since I looked at it, its either some qwen or Mistral model). I use the upsampler to produce json structure based on this prompting guide: https://docs.bfl.ml/guides/prompting_guide_flux2 For being able to run on my 5090, it produces incredibly consistent and stable resu…

Ideogram4 is great! Just make sure that you either use some kind of preprocessor to convert your natural language prompt to the strict structured JSON format otherwise the results can suffer.

I personally use Qwen3 8b LLM for this purpose, but I think that the KJ prompt builder node handles this out of the box in ComfyUI.

https://github.com/kijai/ComfyUI-KJNodes

Re: Flux 3

#144

Post by dang »

kensai wrote: post #63
tormeh wrote: post #57

These people are hiring in... Freiburg im Breisgau? Wonder how hiring is working out for them there.

You are completely ignorant. This is one of the best places to live and work on earth. Nonetheless, they also have positions in San Francisco for what is worth.

I'm sure it's a great place to live and work, but can you please edit out swipes from your comments here, as the site guidelines ask (https://news.ycombinator.com/newsguidelines.html)?

Your post would have been just fine without that first bit.

Re: Flux 3

#145

Post by taneq »

CamperBob2 wrote: post #139
deadbabe wrote: post #132

Earlier quoted context omitted.

filmmakers have the most to gain here, now anyone with a good idea for a story can just prompt out their own full length movie or tv show.

Before AI: 90% of everything was crap After AI: 99% of everything is crap ... but there will be >10x more of it, so we'll ultimately end up with more good stuff.

But since our capacity to experience stuff can't keep up with the sheer quantity of stuff, 99% of the stuff we see will be crap, even if there's technically a larger amount of good stuff in absolute terms.

Re: Flux 3

#147

Post by teiferer »

cpldcpu wrote: post #37
teiferer wrote: post #3

Lots of words about multi-modal but then this: > our mission to develop real-world visual intelligence Visual is mono-modal, isn't it?

its doing video, audio, images and motion. I think that counts as multimodal.

Indeed, but audio isn't exactly "visual", is it?

And I'm not sure video, images and motion are actually 3 different things. Images are just still motion and videos capture motion, so it's really just "video and audio" of which one is visual, the other is not, thus my confused/surprised comment.

Re: Flux 3

#148

Post by deadbabe »

taneq wrote: post #145
CamperBob2 wrote: post #139

Earlier quoted context omitted.

Before AI: 90% of everything was crap After AI: 99% of everything is crap ... but there will be >10x more of it, so we'll ultimately end up with more good stuff.

But since our capacity to experience stuff can't keep up with the sheer quantity of stuff, 99% of the stuff we see will be crap, even if there's technically a larger amount of good stuff in absolute terms.

But not all AI stuff is crap.

There will come a day where some piece of AI content brings a smile to your lips before a tear to your eye.

Re: Flux 3

#149

Post by casey2 »

It's wild that something like this isn't top of all major news sites. I think people are in shock with how fast world models are progressing.
Post reply on HN