You are allowed to use Flux.1 Dev commercially, BFL have clarified what they mean by "non-commercial". The generated images are yours (or technically, like all AI generations, they're public domain, because there's no human authorship). What you're not allowed to do is sell access to the model. That's it. You can't use it to build a Midjourney competitor and ask for money, and you can't sell a Photoshop plugin that uses Flux. "Non-commercial" only refers to the middleman, not the end-user.
Looks great. I hope we get some custom models similar to Rev Animated (from SD 1.5) in style, because it is by far my favorite, and nothing else could scratch that itch for amazing 2.5D images.
The qualiy of the image output is mostly subjective but I'd say this has beaten MJ in all other areas. It's a local install, it's offline and private, it's free and it doesn't need the terrible Discord UI (which is also subjective). To me, that makes it far superior. The are flaws with the model itself, which was expected, but it's results are still fantastic. While it doesn't do celebs (no surprise) it seems to do a really great job with brands, logos and several fictional characters I tried. This was all on an RTX 3060 with 12Gb VRAM and after the initial one time load, each generation took about 2 - 2 1/2 mins.
Outputs of dev can be used commercially, so the average user isn't affected by license, only trying to sell finetunes of the models or things like that are probibited
@@enigmatic_e I saw a video and in the video a product photographer take a shot and tether it to his laptop with capture one, and in another screen comfyui take that photo and change the background, the lighting and basically is delivers an almost finished photo, it was kind of amazing
big fan of forge webui but i guess i will have to check out comfy again. I wonder if existing extensions like layer diffuse or roop would work with this or they would need to be remade for flux? in terms of dethroning midjourney i feel like a good amount of sdxl models already did that. Midjourney fell off a while back and continue to produce minimal or no updates or focus on other aspects that the community doesnt care for.
@HolidayAtHome not bad, but i prefer the full monty version, takes longer sure, but its better/best quality overall - can always adjust the size out renders to be smaller, and upscale the ones u like or wanna keep.
@@oggy3D I generate a 1280 x 1024 image in about 40 seconds. I run it on my RTX360 12gb just fine. Be sure to use the Schnell model and the t5xxl_fp8_e4m3fn file not the FP16 as fp8 is made for GPU's with less ram. Try it out buddy I'm sure you'll like it
@@oggy3D 1 image. but remember I'm only running a 12gb card where as Midjourney is running hundreds of high end cards. I'm happy to wait 40 seconds for a great image that can rival Midjourney and save $30+ a month for.
I wish you would’ve done more nuanced and novel prompts because it probably trained on ALOT of superhero related images as with a lot of other models because it’s so abundant; from comics, to movies and magazines, etc. So it might be very good at reproducing images in those contexts.
Great videos mate. Do you respond to messages on your discord still? I messaged you a while back with mo replies I guessed you might be busy. Just checking if youre still active on it, so i know if i should stay or go
@@enigmatic_e Got it! Thank you. The discord is not right for me then. I need places to learn. I only reach out after conducting significant research and attempts myself. Keep up the good work.
Great model, great results... But 40 seconds to 1 min for a image? Damn, its gonna take a while for this to be able to be adopted by the masses. But for those with high end cards it really can work well.
Can this be run on mobile? I've been running Midjourney and sometimes Leonardo from my Tab S9 Ultra with no issues, but I'm happy to switch to something else
I tried with the obvious prompts and so far I don't have a straight answer yet. I was able to get topless sometimes but bottomless is a no go so far unless those parts are hidden behind something. Horror and gore seems quite doable though with the few tests I did. Celebs are a no go but fctional characters might work. SD1.5 and SDXL custom checkpoints are still the best for that but it's very early days for Flux, so we'll have to see.
@@enigmatic_e actually is with dev. I noticed that it depends on prompt. Some scenes (not more elaborate or complex) require more time than others. First time it happens in four years of comfy or sd.
@@marcobelletz4734 you must be doing something wrong, it takes a few minutes or less depending on the size of render, try 512x512 and increase 768x768 etc, should be much faster - im using 3090 too. - failing that try schnell flux.
Can't be finetuned and can't do porn properly for what i was told Image gen models are usable to be finetuned into porn generators so this is pretty much dead on arival
flux is over rated it can do hands and poses better but i find sdxl lightning plus realvis 4.0 does WAY better with lifelike people and lighting. plus apparently you wont be able to fine tune flux which is terrible!
Thanks for the video! but I don't know why I get the error: module 'torch' has no attribute 'float8_e4m3fn' I updated comfy and my torch, placed all the models and vae but still get this error. my PC is super powerful (128ram and 24vrma, windows 10) maybe you know why? Thanks!
You are allowed to use Flux.1 Dev commercially, BFL have clarified what they mean by "non-commercial". The generated images are yours (or technically, like all AI generations, they're public domain, because there's no human authorship). What you're not allowed to do is sell access to the model. That's it. You can't use it to build a Midjourney competitor and ask for money, and you can't sell a Photoshop plugin that uses Flux. "Non-commercial" only refers to the middleman, not the end-user.
Thanks for clarifying! 🙏🏽🙏🏽
I don't have no followers no way so I ain't worried about it I'm trying to sell or monetize
Looks great.
I hope we get some custom models similar to Rev Animated (from SD 1.5) in style, because it is by far my favorite, and nothing else could scratch that itch for amazing 2.5D images.
Can't wait to see Flux AnimateDiff
Literally my first thought lol
@@johntnguyen1976 it's happening!
you’ve come so far with your video quality great work
Yeah, I’ve learned a lot since I started. 😊
Fantastic video and results!
Thanks!!
The qualiy of the image output is mostly subjective but I'd say this has beaten MJ in all other areas. It's a local install, it's offline and private, it's free and it doesn't need the terrible Discord UI (which is also subjective). To me, that makes it far superior.
The are flaws with the model itself, which was expected, but it's results are still fantastic. While it doesn't do celebs (no surprise) it seems to do a really great job with brands, logos and several fictional characters I tried. This was all on an RTX 3060 with 12Gb VRAM and after the initial one time load, each generation took about 2 - 2 1/2 mins.
Yea, I think the whole running it locally and the fact that’s free is a big plus to a lot of us.
dude, as usual, giving and showing the best tools available. U da best. Eres el mejor
Gracias!!! 🙏🙏🙏🙏🙏🙏
Outputs of dev can be used commercially, so the average user isn't affected by license, only trying to sell finetunes of the models or things like that are probibited
They saw his demo of his crazy video model and to think that it will be open source, there are no limits to what you can do
Definitely exciting times!
Flux dev's non-commercial license allows for selling images made by Flux dev, however it doesn't allow for selling Flux dev or any finetunes of it
I got the crystools GPU and CPU resource monitor and it makes me feel extra nerdy like MR. Robot
what is this?
This looks awesome. I got out the AI game for a bit to work on a project but excited to explore some of these new tools!
Welcome back!
Thanks for this in deep review! Die quality is amazing!
awesome video.
Is there any workflow that involves tethering so that i can take a photo and automatically edit it in comfyui?
Edit it in what way?
@@enigmatic_e I saw a video and in the video a product photographer take a shot and tether it to his laptop with capture one, and in another screen comfyui take that photo and change the background, the lighting and basically is delivers an almost finished photo, it was kind of amazing
great video as always man , quick question, the flux1-dev.safetensors is the same as flux1-dev.sft right ?
Thanks mate for the great video
im hoping this gets incorporated into Stability Matrix.
Mind sharing the link for that Flux prompt enhancer? I don't see if in the description.
Yes, my bad, check description again.
where the "Load diffusion model" node to download?
I don't have the node:(
Really appreciate you and your videos.
🙏🏽🙏🏽🙏🏽 ♥️♥️♥️
big fan of forge webui but i guess i will have to check out comfy again. I wonder if existing extensions like layer diffuse or roop would work with this or they would need to be remade for flux? in terms of dethroning midjourney i feel like a good amount of sdxl models already did that. Midjourney fell off a while back and continue to produce minimal or no updates or focus on other aspects that the community doesnt care for.
Awesome. How can we incorporate real image in picture?
great video thank you!
Is it possible to make animations using flux?
the elo of this surpasses mj 6.1/6 so it's already been dethroned, espec since Dev is so easy to run on a local machine.
Allow me to comment on your page what's the name of it quick put a timer for product comes out it's already outdated😢😮
14..15 guys created wonder 😂😂😂
fp8 version is out. Takes 2 Seconds with the schnell model. 13 Seconds with the dev model ( on a 4090 )
On a 4090...
@HolidayAtHome not bad, but i prefer the full monty version, takes longer sure, but its better/best quality overall - can always adjust the size out renders to be smaller, and upscale the ones u like or wanna keep.
mind BLOWN! this is amazing!
RIP Midjourney. This follows my prompts way better than midjourney does and runs locally on my 3060 RTX 12GB Very cool
Can it run in my 4070 12gb ?
And how much time it takes to show results? Reply would be appreciate
@@oggy3D I generate a 1280 x 1024 image in about 40 seconds. I run it on my RTX360 12gb just fine. Be sure to use the Schnell model and the t5xxl_fp8_e4m3fn file not the FP16 as fp8 is made for GPU's with less ram. Try it out buddy I'm sure you'll like it
@@TPCDAZ 40 sec to render 1 image or 4 ?
@@oggy3D 1 image. but remember I'm only running a 12gb card where as Midjourney is running hundreds of high end cards. I'm happy to wait 40 seconds for a great image that can rival Midjourney and save $30+ a month for.
@@TPCDAZ yeah you are right, but concerned about my gpu life i mean it wont burn my gpu right if i use to much?
How did you import .png as workflow?
There is workflow data in images that are generated in comfyui
I wish you would’ve done more nuanced and novel prompts because it probably trained on ALOT of superhero related images as with a lot of other models because it’s so abundant; from comics, to movies and magazines, etc. So it might be very good at reproducing images in those contexts.
I agree, I will definitely keep that in mind with future tests.
I will say this, when I tried famous characters in SD3, it was still giving me bad results.
Amazing!
We need controlnet for this asap :-)
Yes!!!
Great videos mate. Do you respond to messages on your discord still?
I messaged you a while back with mo replies
I guessed you might be busy. Just checking if youre still active on it, so i know if i should stay or go
@@SejalDatta-l9u I respond sometimes, don’t always have the time to respond especially for trouble shooting issues.
@@enigmatic_e Got it! Thank you. The discord is not right for me then.
I need places to learn.
I only reach out after conducting significant research and attempts myself.
Keep up the good work.
Great model, great results... But 40 seconds to 1 min for a image? Damn, its gonna take a while for this to be able to be adopted by the masses. But for those with high end cards it really can work well.
Yeah, the generation time can take pretty long, I hope people will find ways to optimize it somehow.
@@enigmatic_e can just lower the size like 512x512 with 3-4 batch images, much faster, just upscale the ones you wanna keep etc
Anything free will always be better then anything that costs money period.
Can this be run on mobile? I've been running Midjourney and sometimes Leonardo from my Tab S9 Ultra with no issues, but I'm happy to switch to something else
for schnell sometimes it takes 40 seconds and sometimes it takes 100 seconds, on a 4070
Will it work in swarm?
The real question is : is it a true uncensored model? If it is, we finally have an upgrade to SD1.5!
I tried with the obvious prompts and so far I don't have a straight answer yet. I was able to get topless sometimes but bottomless is a no go so far unless those parts are hidden behind something. Horror and gore seems quite doable though with the few tests I did. Celebs are a no go but fctional characters might work.
SD1.5 and SDXL custom checkpoints are still the best for that but it's very early days for Flux, so we'll have to see.
@@Elwaves2925 customs tunes and checkpoints are happening, in the meantime, sd3 is being lowered 12ft under for good! ;)
@@2PeteShakur As Status Quo sang - Down, down, deeper and down.
i was waiting for you to throw the hulk prompt into midjourney for comparison...
@@madshader you remember the sd3 version? 🤣
Python Has Stopped Working. I get this everytime.
not strong enough pc.
insert *shut up and take my like and subscribe* meme
cant wait to see.....
Can work on 8GB VRAM ?
I doubt it, or it will take forever to generate.
@@enigmatic_e Thank you
strange... just generating black images.... maybe I moved something I shouldn't have hahaha
Have you got the correct VAE for this in the correct folder?
prompt enhancer?
Oh my bad, I just put it in the description.
u need to try Flexy
What is Flexy?
25 minutes for a single generation on 3090 RTX 24GB? That's weird
Is that with Schnell?
@@enigmatic_e actually is with dev. I noticed that it depends on prompt. Some scenes (not more elaborate or complex) require more time than others. First time it happens in four years of comfy or sd.
@@marcobelletz4734 you must be doing something wrong, it takes a few minutes or less depending on the size of render, try 512x512 and increase 768x768 etc, should be much faster - im using 3090 too. - failing that try schnell flux.
I doubt this will run on a RTX 4060 GPU
Hmm not sure. You might have test and see.
The Schnell and Dev models are running for me on a 3060 with 12GB VRAM.
it is paid now??
It’s not.
👍👍❤❤
Can't be finetuned and can't do porn properly for what i was told
Image gen models are usable to be finetuned into porn generators so this is pretty much dead on arival
flux is over rated it can do hands and poses better but i find sdxl lightning plus realvis 4.0 does WAY better with lifelike people and lighting. plus apparently you wont be able to fine tune flux which is terrible!
I would want to wait and see how things develop, I wouldn’t want to dismiss what Black Forest Labs or the community can do.
you can finetune flux, you've been mislead. a lot of initial training scripts are already out
@@xbon1 🎉🎉🎉🎉
Better than Midjourney? Are you serious? 🤣😂😅
literally been proven by ELO
Oh ! There you are !
@@cowlevelcrypto2346 🙋🏽♂️
Thanks for the video! but I don't know why I get the error: module 'torch' has no attribute 'float8_e4m3fn'
I updated comfy and my torch, placed all the models and vae but still get this error.
my PC is super powerful (128ram and 24vrma, windows 10) maybe you know why?
Thanks!
Paste your error into GPT and it will walk you through how to fix it.
@@skycladsquirrel yeah i tried that, no go. but thanks