Please check out the Quickstart Guide to Flux for all the info you need to get started!
FLUX.1 [dev] is a 12 billion parameter rectified flow transformer capable of generating images from text descriptions. For more information, please read our blog post.
Key Features
Cutting-edge output quality, second only to our state-of-the-art model
FLUX.1 [pro].Competitive prompt following, matching the performance of closed source alternatives .
Trained using guidance distillation, making
FLUX.1 [dev]more efficient.Open weights to drive new scientific research, and empower artists to develop innovative workflows.
Generated outputs can be used for personal, scientific, and commercial purposes as described in the flux-1-dev-non-commercial-license.
Usage
We provide a reference implementation of FLUX.1 [dev], as well as sampling code, in a dedicated github repository. Developers and creatives looking to build on top of FLUX.1 [dev] are encouraged to use this as a starting point.
Learn More Here:
https://huggingface.co/black-forest-labs/FLUX.1-dev
Description
FAQ
Comments (91)
you are theally
civitai search recovery reminds me of yahoo from the early 90's ...i search for newest post's...get post from 25 days ago... so i then go search most reactions.... get post from yesterday. :-/
Yep, it's weird search.
Flux.1 Krea Dev - Excellent checkpoint, it follows prompts better than anything I've used so far. (and I've used a lot), it also plays well with LoRas. I'm running it on Forge and it works perfectly, and runs just like Flux.1 [dev] with no additional things needed.
my loras only half-work. probably have to retrain them for optimal use
cutetodeath78409597 Yeah, I've noticed that too, after playing with it for a day, I've also noticed that the human characters tend to be a lot more "cartoonish" looking compared to other models.
I'm so new to all of this and that I know right after that but flux is the gold standard when it comes to image generation
maybe the bronze standard but you're catching on
Krea actually does EXPRESSIONS! Human facial expressions!!! Original flux dev was terrible with this.
why do images often have horizontal semi-visible stripes in noisy areas? for example, https://civitai.com/images/91751003 from the examples (they are also on my generations), did I do something wrong (ae.safetensors files)/destilated cfg settings, steps or is this a feature of the model? I looked at the photo on the main page on civit.ai (https://civitai.com/models/618692?modelVersionId=2068000) and there you can also see the stripes on the noise, sometimes barely visible, sometimes more, but it is annoying. my settings - https://i.imgur.com/K01jDFX.png, ae file seems to be original from flux dev
Try running your image through the upscaler. No need to actually upscale, leave it at a 1:1 ratio. That'll get rid of the banding. I don't use forge so I don't know where that option is.
two things usually cause this for flux; loras or weird resolution sizes -- start there.
Try different guidance, samplers, resolutions, and LoRAs. Each of these things can affect the noise banding. Krea is especially pone to this issue but it also happens with the normal Flux Dev.
try increasing CFG and the number of iteration steps. It helps a little. Someone also already mentioned using an upscaler.
Because the message/image feed is sick, I don't understand when my generations can be seen by others((
Does it seem to you, too, that the model really likes to add bangs?
:looks at all the image with large foreheads: not really, no.
AmaaAaazing !!! 👌
Flux Pro is listed, but clicking "Create" doesn't work. Any idea when this might be fixed?
day late on this but... probably just server side issues they've been having "issues" for day or two
It was replaced by Krea. And there is little reason to use Pro when Pro 1.1 is available, it was there just for legacy API reasons.
Flux Pro was a legacy model and has now been dropped, superseded by Pro 1.1 which is faster, and cheaper, with better results.
worked very well for me, love it
很棒!!!
Flux images went into this experiment with WAN22...
The Chocolate Weapon
this model is piece of shit that almost no different from its predecessor , it knows no shit about art or fantasy
Not even close to true lmao
Someone who resorts to such nasty language without any solid arguments makes me not want to imagine what things must be like in their underwear.
@Silencer wake the hell up, you pay today for API and you pay everywhere, yet we spent hours upon hours testing and generating images for Stable Diffusion OR Black Forest. paid extra $ to be able to use the tech, but were left with nothing.... all this: Flux A, Flux B, or whatever free Flux is a huge BS. no model today can be a really good free alternative to what we already had for years. and it has nothing to do with a language. this platform had a bright future yet its all got reduced to what it is today. dont roll your eyes as a princess. languge is not about person but a model. hard work is not only building the model but also testing it. get real!
@BzzzDarklord wat?
@BzzzDarklord With all due respect, that may well be the case, and your opinion may indeed be as you say—or not. But it is based on opinion, and only on opinion. I am unimpressed by that, and to be perfectly frank, I neither asked for it, nor doubted it, nor encouraged it. What I do find objectionable is your tone. Fine, I can disregard it; and surely you can too (to cut it short: you most likely couldn’t care less about how you come across). Still, when something smells like nonsense, it most probably is. And you can stand on your head, wiggle your ears, rant, stay silent, sulk, blame me or half the internet, send out countless wake the hell up calls - it makes no difference.
Therefore, this model (or in generel Flux) is great. I speak only for myself and don't force you or others to my opinion. Beyond that, I paid nothing and no one. Except with my own free time to give some respectful feedback to the creators, and my electricity bill (since everything runs locally).
tl;dr: if (in any kind) you want to be taken a bit of serious, just imagine you would try another more kindly approach.
@Silencer You get down to personals, again, yet i criticize the model- you will get over it when grow up, or maybe not...i will describe 'your model greatness' in one sentence: 'ass-chicks replaced the chin'- basically, cant get non-cleft chin, nasolabial folds of 40 yo male, instant 2 giant front hare teeth, cant get rounded figure forms- all on diet, limited motions and poses, limited styles, limited fantasy sets and this is only for starters. where its pretty good? in photo realism. now, I do not force my opinion on anyone, yet, you can agree or disagree, that's why its being called 'comments'. once you get down to personals, you will get kicked off by admins.
@BzzzDarklord I value respectful communication and there is nothing personal in that. It is regrettable if this was perceived otherwise. Speaking of 'personal' in this context is misleading; should you feel disrespected, the admins are the right point of contact. I take note of your opinion. However, it is my free opinion that I say again - I will take someone who uses filthy language not for serious and I speak only for myself. You may deal with it or not.
@Silencer stop trolling! I don't care what you think, period. Respect needs to be earned first, not to be given by default
@BzzzDarklord Ok, I tried politely and I’ll remain polite, but also clear on the matter: this is not trolling. The issue is that you are not even showing basic courtesy. Instead, your language has turned crude and vulgar, and from the beginning you offered no real arguments; your attempt to create a discussion never rose above the excremental tone of an adolescent.
Now you take offense at someone pointing this out. Deal with it, simple as that. Or not. It just seems unfortunate that this clearly troubles you, as you now turn up here uninvited (see previous posts) with confused stories that are of little real interest. I certainly never asked you for them. And yet you think it appropriate to burden the forum with such foul language? Please don’t reply to this… see above. This is meant as a constructive reminder, though even the suggestion to place a little more value on courtesy seems to bounce right off you.
So stay where you are - but preferably at a respectful distance.
Feel free to report anytime to an admin and make sure you make your point clear.
However, if you once again mess up another section here in such a manner and I happen to see it, then deal with it or not that I will call you what you appear to be. And with this, I will politely withdraw from this discussion and wish you all the best.
@Akalabeth i love your images, probably the only one here, on these comments, that would respect their opinion, good job! keep on doing.
No commercial use. ILLUSTRIOUS / PONY / SDXL with LORAs are very good too, not sure about that huge amount of Go...
EDIT: My bad, the license specifies it is possible to use the output for commercial projects. That being said, it lacks a lot compared to the older but better SDXL :D
Reading is hard:
d. Outputs. We claim no ownership rights in and to the Outputs. You are solely responsible for the Outputs you generate and their subsequent uses in accordance with this License. You may use Output for any purpose (including for commercial purposes), except as expressly prohibited herein. You may not use the Output to train, fine-tune or distill a model that is competitive with the FLUX.1 [dev] Model or the FLUX.1 Kontext [dev] Model.
Or do you intend to offer use of the model and/or a derivative for momentary gain?
Also, define better. The whole SDXL community support is immense. Comparing base to base, flux wins each and every time, except for speed.
@Mittows Thank you. My bad, you are right about the output. There's a lot of misinterpretation about this license, that seems completely wrong. Thanks again for quoting the source.
https://github.com/black-forest-labs/flux/blob/main/model_licenses/LICENSE-FLUX1-dev
@_8_I_8_ You're welcome. I thought the same at first. It still doesn't make much sense to prohibit using the model itself to make money, but not the output. I'd consider it the same thing if it wasn't mentioned separately.
@Mittows They want to keep the model and license it for AI sites, it is pretty logical. As they are a business company and not an open source entity. However, as you said, it makes to sense to forbid using the output of the model.
Hurray! I managed to run Flux Dev fp32 on Geforce 1070 8 GB
uhhhh.. k.
Bro, tell me the steps if you can, I have a GTX 1650 4gb hehe :/
@6vidit9 I know you guys are joking, but seriously, some Forge versions claim they can manage it with only 6 GB. I doubt they can handle the FP32 model, and even if they could, it would probably be without loading any LoRA.
nah, you'll need Forge if you want something that looks almost exactly like A1111. It hasn't been updated in a while but it still works fine
I used Flux 1 Dev (Q8 quant) to create ultrawide DQHD wallpapers (first rendering at 2560x720, then Upscaling 2x to 5120x1440) and had excellent results. But with KREA (Q6, Q8 and FP8), with same resolution, even using the Comfy template, got pretty blurry and grainy images. Anybody else having issues with grainy and blurry generations at wider higher resolutions than ~1536? Thanks!
I have a similar issue when I try to use USO with Krea, the output image has a grey filter with a blurry / grainy rendering
Krea likes higher guidance. Try 4.5 with Euler Beta.
@ZootAllures9111 Thanks for the tip. Gonna try it again and update with results.
always gets great results
Isn’t struggling part of the fun of generating images?
Krea is the best all-round model for realism in terms of quality-to-speed ratio IMO
In general great but after trying for a while, I am bored of the ever same looking common average, bland pretty faces without own character.
KREA IS A MASTERPIECE 🥂🤍
how you all run this model on your computer , its too heavy . you are all so rich bro :/
try 'nunchaku' or 'gguf'
lol it works fine on my 3060, just needs a bit longer then normal models
the gguf variants lower the barrier to entry a little, but yeah. Not everyone is playing with entry level hardware, even if a gen or 3 older lol
I am using 4gb gpu , with i5 3rd gen processor with 16 gb ram . that much poor I am@halfBurn @xFanexx @D3monCeps
@proudbeing90415 Ultimately, you've gotta save-up for some better hardware if you really wanna play with this stuff, or a lot of things really
@D3monCeps i know bro , my computer is from dinosaur era . i think in this community i am the only one who is still using sd 1.5 and sdxl 1.0 model 😂
@proudbeing90415 Just save up money and you will be able to get a nice High End GPU In the future <3
The current iron market is absurd because kidneys are more affordable than their component parts.
I just have an i7 + RTX 4070 12GB and 32GB DDR4 - generates well on Win10 with a dedicated swapfile on a decent 2TB SSD.
Please stop posting images created with “Midjourney” in the “FLUX Dev” gallery.
What are you referring to exactly? Midjourney V7 isn't even better than Flux Dev at all, also, it's not anywhere remotely close to to the top of any recent text-to-image model user preference leaderboard.
@ZootAllures9111 I’m not talking about something like that. Can’t you tell who I’m referring to by looking at my image gallery?
It's an Midjourney LORA typically
Still one of my favourites
она очень много весит, у меня памяти нет в облаке
Any timeline on the Flux.2 models becoming available for image generation and Lora training?
I got it to work amazingly one time, my pod crashed, been at it for 48 hours straight now, shits coming out so blurry , having issues organizing the support files in the right place
aaand none of that has to do with civitai lol..
Actually Blurry is a classic sign of a corrupted model file. You should always crc check any model before loading it .. or yes you'll crash or at best be blurry
how to run flux krea on Stable Diffusion Forge?
The realism is insane. I'm impressed with how well it handles lighting and complex poses. Definitely one of my favorite models to work with right now. Keep up the great work!
I know theally posted the model and the images, but why it's not on his account?
will i be able to run this model on EasyDiffusion 3.0?
Good quality and very fast, I like it!
Stupid Model dont download!
care to explain why before i get it ?
Very complete checkpoint, it lends itself to doing many things!
Helped me to create this cool video!
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.



















