[HN Gopher] Releasing open weights for FLUX.1 Krea
       ___________________________________________________________________
        
       Releasing open weights for FLUX.1 Krea
        
       Author : vmatsiiako
       Score  : 144 points
       Date   : 2025-07-31 13:41 UTC (9 hours ago)
        
 (HTM) web link (www.krea.ai)
 (TXT) w3m dump (www.krea.ai)
        
       | dvrp wrote:
       | Hello everyone.
       | 
       | I'm the Co-founder and CTO of Krea. We're excited because we
       | wanted to release the weights for our model and share it with the
       | HN community for a long time.
       | 
       | My team and I will try to be online and try to answer any
       | questions you may have throughout the day.
        
         | jackphilson wrote:
         | Hi. Thanks for this. What is your goal of doing so? From a
         | business standpoint. Or is it purely altruistic?
        
           | dvrp wrote:
           | Haha-classic!
           | 
           | It's simple: hackability and recruiting!
           | 
           | The open-source community hacking around it and playing with
           | it PLUS talented engineers who may be interested in working
           | with us already makes this release worth it. A single
           | talented distributed systems engineer has a lot of impact
           | here.
           | 
           | Also, the company ethos is around AI
           | hackability/controllability, high-bar for talent, and AI for
           | creatives - so this aligns perfectly.
           | 
           | The fact that Krea serves both in-house and 3rd-Party models
           | tells you that we are not that bullish on models being a
           | moat.
        
             | wjrb wrote:
             | I can say that it's definitely working on me! I hadn't
             | heard of Krea before, and this is a great introduction to
             | your work. Thanks for sharing it.
        
       | dvrp wrote:
       | For Dang or HN-mods:
       | 
       | I noticed that the URL for this submission is wrong: I tried to
       | submit the correct URL (https://www.krea.ai/blog/flux-krea-open-
       | source-release) but, for some reason, the submission gets flagged
       | as duplicated and then I can only find this item which has a URL
       | to our old blog post.
       | 
       | In the mean time, I'll setup a server-side redirect from the old
       | blog post to our new one, but it would be nice to fix the link
       | and I don't think I can do it on my side.
        
         | dang wrote:
         | It's because https://www.krea.ai/blog/flux-krea-open-source-
         | release contains this:                 <link rel="canonical"
         | href="https://www.krea.ai/blog/new-krea">
         | 
         | Our software follows canonical links when it finds them.
         | 
         | I've fixed the link above now (and rolled back the clock on the
         | submission, to make up for lost time) but you might want to fix
         | this for future pages.
        
           | dvrp wrote:
           | OMG. Thank you! I had to setup a CDN-level redirect and I was
           | so confused as to why when I asked others to help, their
           | submissions were flagged as [dupe] or [dead]
           | 
           | Thank you so much! I knew that HN software was advanced, but
           | I didn't know you guys used Canonical URLs like Google does.
           | Smart and thanks for helping us with this slip!!!
        
       | sangwulee wrote:
       | Hi! I'm lead researcher on Krea-1. FLUX.1 Krea is a 12B rectified
       | flow model distilled from Krea-1, designed to be compatible with
       | FLUX architecture. Happy to answer any technical questions :)
        
         | swyx wrote:
         | thanks for doing this!
         | 
         | what does " designed to be compatible with FLUX architecture"
         | mean and why is that important?
        
           | sangwulee wrote:
           | FLUX.1 is one of the most popular open weights text-to-image
           | models. We distilled Krea-1 to FLUX.1 [dev] model so that the
           | community can adopt it seamlessly into existing ecosystem.
           | Any finetuning code, workflows, etc that was built on top of
           | FLUX.1 [dev] can be reused with our model :)
        
             | swyx wrote:
             | do LoRAs conflict with your distillation?
        
         | oompty wrote:
         | The model looks incredible!
         | 
         | Regarding this part: > Since flux-dev-raw is a guidance
         | distilled model, we devise a custom loss to finetune the model
         | directly on a classifier-free guided distribution.
         | 
         | Could you go more into detail on the specific loss used for
         | this and any other possible tips for finetuning this that you
         | might have? I remember the general open source ai art community
         | had a hard time with finetuning the original distilled flux-dev
         | so I'm very curious about that.
        
       | vipermu wrote:
       | hey hn! I'm one of the founders at Krea.
       | 
       | we prepared a blogpost about how we trained FLUX Krea if you're
       | interested in learning more: https://www.krea.ai/blog/flux-krea-
       | open-source-release
        
       | erwannmillon wrote:
       | yoo i'm also a researcher on the krea 1 project and happy to
       | answer any questions :)
        
       | OsrsNeedsf2P wrote:
       | Cool to see an open weight model for this. But what's the
       | business use case? Is it for people who want to put fake faces on
       | their website that don't look AI generated?
        
         | dvrp wrote:
         | Thanks!
         | 
         | From a business point of view, there are many use-cases. Here's
         | a list in no particular order:
         | 
         | - You can quickly generate assets that can be used _alongside_
         | more traditional tools such as Adobe Photoshop, After Effects,
         | or Maya/Blender/3ds Max. I've seen people creating diffuse maps
         | for 3D using a mix of diffusion models and manual tweaking with
         | Photoshop.
         | 
         | - Because this model is compatible with the FLUX architecture,
         | we've also seen people personalizing the model to keep products
         | or characters consistent across shots. This is useful in
         | e-commerce and fashion industry. We allow easy training in our
         | website -- we labeled it Krea 1 -- to do this, but the idea
         | with this release is to encourage people with local rigs and
         | more powerful GPUs to be able to tweak with LoRAs themselves
         | too.
         | 
         | - Then I've seen fascinating use-cases such as UI/UX designers
         | who prompt the model to create icons, illustrations, and
         | sometimes even whole layouts that then they use as a reference
         | (like Pinterest) to refine their designs on Figma. This reminds
         | me of people who have a raster image and then vectorize it
         | manually using the pen tool in Adobe Illustrator.
         | 
         | We also have seen big companies using it for both internal
         | presentations and external ads across marketing teams and big
         | agencies like Publicis.
         | 
         | EDIT: Then there's a more speculative use-case that I have in
         | mind: Generating realistic pictures of food.
         | 
         | While many restaurants have people who either make
         | illustrations of their menu items and others have
         | photographers, the big tail of restaurants do not have the
         | means/expertise to do this. The idea we have from the company
         | perspective is to make it as easy as snapping a few pictures of
         | all your dishes and being able to turn all your menu (in this
         | case) into a set of professional-looking pictures that
         | accurately represent your menu.
        
       | gshklovski wrote:
       | Helpful blog post for understanding what kind of data is needed
       | for these models!
       | 
       | Does this have any application for generating realistic scenes
       | for robotics training?
        
         | dvrp wrote:
         | Hey thanks! I'll ask Sangwu to hop here to answer this and give
         | a more research-oriented answer
        
         | sangwulee wrote:
         | Thank you! Glad you find it helpful. The model is focused on
         | photorealism so it should be able to generate most realistic
         | scenes. Although, I think using 3D engines would be more
         | suitable for typical cases for robotics training since it gives
         | you ground truth data on objects, location, etc.
         | 
         | One interesting use case would be if you are focusing on a
         | robotics task that would require perception of realistic
         | scenes.
        
         | erwannmillon wrote:
         | yeah data really is everything that was the number one lesson
         | from this whole project
        
       | kmavm wrote:
       | Amazing. I can practically _smell_ that owl it looks so darned
       | owl-like.
       | 
       | From the article it doesn't seem as though photorealism per se
       | was a goal in training; was that just emergent from human
       | preferences, or did it take some specific dataset construction
       | mojo?
        
         | sangwulee wrote:
         | I love owls. Photorealism was one of the focus areas for
         | training because "AI look" (e.g. plastic skin) was biggest
         | complaint for FLUX.1 model series. Photorealism was achieved
         | with both careful curation of finetuning and preference
         | dataset.
        
       | dvrp wrote:
       | Relevant links:
       | 
       | - GitHub repository: https://github.com/krea-ai/flux-krea
       | 
       | - Model Technical Report: https://www.krea.ai/blog/flux-krea-
       | open-source-release
       | 
       | - Huggingface model card: https://huggingface.co/black-forest-
       | labs/FLUX.1-Krea-dev
        
       | bangaladore wrote:
       | Can someone ELI5 why the safetensor file is 23.8 GB, given the
       | 12B parameter model? Does the model use closer to 24 GB of VRAM
       | or 12 GB of VRAM. I've always associated a 1 billion parameter =
       | 1 GB of VRAM. Is this estimate inaccurate?
        
         | petercooper wrote:
         | That's a good ballpark for something quantized to 8 bits per
         | parameter. But you can 2x/4x that for 16 and 32 bit.
        
           | 7734128 wrote:
           | I've never seen a 32 bit model. There's bound to be a few of
           | them, but it's hardly a normal precision.
        
             | zamadatix wrote:
             | Some of the most famous models were distributed as F32,
             | e.g. GPT-2. As things have shifted more towards mass
             | consumption of model weights it's become less and less
             | common to see.
        
               | petercooper wrote:
               | And on the topic of image generation models, I think all
               | the Stable Diffusion 1.x models were distributed in f32.
        
               | nodja wrote:
               | > As things have shifted more towards mass consumption of
               | model weights it's become less and less common to see.
               | 
               | Not the real reason. The real reason is that training has
               | moved to FP/BF16 over the years as NVIDIA made that more
               | efficient in their hardware, the same reason you're
               | starting to see some models being released in 8bit
               | formats (deepseek).
               | 
               | Of course people can always quantize the weights to
               | smaller sizes, but the master versions of the weights is
               | usually 16bit.
        
         | piperswe wrote:
         | A parameter can be any size float. Lots of downloadable models
         | are FP8 (8 bits per parameter), but it appears this model is
         | FP16 (16 bits per parameter)
         | 
         | Often, the training is done in FP16 then quantized down to FP8
         | or FP4 for distribution.
        
         | sangwulee wrote:
         | Quick napkin math assuming bfloat16 format : 1B * 16 bits = 16B
         | bits = 2GB. Since it's a 12B parameter model, you get around
         | ~24GB. Downcasting to bfloat16 from float32 comes with pretty
         | minimal performance degradation, so we uploaded the weights in
         | bfloat16 format.
        
       | TuringNYC wrote:
       | I usually use https://github.com/axolotl-ai-cloud/axolotl on
       | Lambda/Together for working with these types of models. Curious
       | what others are using? What is the quickest way to get started?
       | They mention Pre-training and Post-training but sadly didnt
       | provide any reference starter scripts.
        
         | dvrp wrote:
         | We actually have a GitHub repository to help with inference
         | code.
         | 
         | Check this out: https://github.com/krea-ai/flux-krea
         | 
         | Let me see if we can add more details on the blog post and
         | thanks for the flag!
        
           | TuringNYC wrote:
           | Thanks! Yes, the inference is pretty straightforward, but the
           | real opportunity IMHO is the custom pre-training and post-
           | training opportunities given the open weights.
        
       | sergiotapia wrote:
       | For the Krea team that might be reading: I was trying to evaluate
       | Krea for my image gen use case, and couldn't find:
       | 
       | - cost per image - latency per image
       | 
       | Hope you guys can add it somewhere!
        
         | dvrp wrote:
         | Yup... We have the following: https://www.krea.ai/pricing
         | 
         | Though we wanted to keep this technical blogpost free from
         | marketing fluff, but maybe we over-did it.
         | 
         | However, sometimes it's hard to give an exact price per image,
         | as it depends on resolution, number of steps, whether a LoRA is
         | being used or not, etc.
        
       | bluehark wrote:
       | Do you have an NVIDIA optimized version? Similar to how RTX
       | accelerated FLUX.1 Kontext: https://blogs.nvidia.com/blog/rtx-ai-
       | garage-flux-kontext-nim...
        
         | sangwulee wrote:
         | We have not added a separate RTX accelerated version for FLUX.1
         | Krea, but the model is fully compatible with existing FLUX.1
         | dev codebase. I don't think we made a separate onnx export for
         | it though. Doing 4~8 bit quantized version with SVDQuant would
         | be a nice follow up so that the checkpoint is more friendly for
         | consumer grade hardware.
        
       | bluehark wrote:
       | How large was the dataset used for post-training?
        
         | sangwulee wrote:
         | We used two types of datasets for post-training. Supervised
         | finetuning data and preference data used for RLHF stage. You
         | can actually use less than < 1M samples to significantly boost
         | the aesthetics. Quality matters A LOT. Quantity helps with
         | generalisation and stability of the checkpoints though.
        
       | qlm wrote:
       | Images still look AI generated
        
         | dvrp wrote:
         | Yeah, there are still imperfections. But, it's surprising to us
         | how much the quality can be improved without the need of a
         | whole pre-training (re-) run.
        
       | artninja1988 wrote:
       | Thanks for the release to the team! Any plans on doing the same
       | for FLUX.1 Kontext?
        
       | leftstrokeviral wrote:
       | How much data is the model trained on?
        
         | dvrp wrote:
         | Copying and pasting Sangwu's answer:
         | 
         | We used two types of datasets for post-training. Supervised
         | finetuning data and preference data used for RLHF stage. You
         | can actually use less than < 1M samples to significantly boost
         | the aesthetics. Quality matters A LOT. Quantity helps with
         | generalisation and stability of the checkpoints though.
        
       | paparicio wrote:
       | Great, these people are crazy. CONGRATS!!!
        
         | dvrp wrote:
         | Haha thanks! Do you happen to have a use-case for it?
        
       | TuringNYC wrote:
       | I'd recommend you offer a clearly documented pathway for
       | companies to license commercial output usage rights if they get
       | the results they seek (i'll know soon enough!)
        
         | dvrp wrote:
         | You can find details about the license here:
         | https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/mai...
         | 
         | In a nutshell, it follows the same license as BFL Flux-dev
         | model.
        
       | ilc wrote:
       | Tried a simple prompt, and got some pretty interesting results:
       | 
       | "Octopus DJ spinning the turntables at a rave."
       | 
       | The human like hands the DJ sprouts are interesting, and no
       | amount of prompting seems to stop them.
       | 
       | Opinionated, as the paper says.
        
         | earthicus wrote:
         | Describing it as "Octopus DJ with no fingers" got rid of the
         | hands for me, but interestingly, also removed every
         | anthropomorphized element of the octopus, so that it was
         | literally just an octopus spinning turntables.
        
       | jacooper wrote:
       | Non-commercial license... What's the point even, I can't use it
       | for anything.
        
         | hhh wrote:
         | for the love of the game, not for corporate greed
        
       ___________________________________________________________________
       (page generated 2025-07-31 23:00 UTC)