100% agree on your Midjourney-killer comments. I'm getting pretty sick of all the click-bait titles where every new thing is supposedly the end of the competitor, and their whole team might as well jump off a bridge or whatever. But these content creators need content, so here we are.
@@kevnar Hey man, at 100,000 subscribers you got to dress for success - it is a business! :-) That said the normal garb is already iconic :-)…. Keep on being awesome! :-) The kids at my AI still summer camp did very well, thanks in large part to you! :-)
It's so tiring.. on Reddit things get especially bad, with vitriol aimed at Stability AI for releasing a 'broken' SD3 model.. people keep forgetting they're getting all this for free.
He is an image and video channel, AI explained and Bycloud are better at covering AI technology as a whole, and Mat Vid Pro is better at covering up scaling/music, while Mathew Bernard is better at covering open source LLM logic tests, they all have their specialties :)
Congratulations!!! It's a big achievement. For more to come! Flux Isn't the MJ's killer but I do think it's the best Open Source option at the moment. Flux Dev outputs can be used commercially, derivatives from the model can't. I thought the same but after someone pointed out, I re-read the license and was pleasantly surprised, as I will use it commercially. The difference is big between dev and schnell.
Yeah, I don't see it as a Midjourney killer either. As I mentioned in the video, MJ has at least one massive gamechanger in the pipeline: 3d. That's going to be huge. Ultimately: We can have two things. And two things are good!
Congratulations on the plaque! Well deserved. Nice to see someone highlighting positive and negative aspects of all available solutions, instead of using terms like "killer". Yes, more is better.
Congrats, Tim! I need all the new releases to slow down long enough to catch up on the old ones! I just started using Hedra today... looks like I'll be adding Flux to the mix soon. I barely sleep anymore since I fell down the AI creation tools rabbit hole. Really enjoying your videos and so happy to see your success!
grats on 100k, here's +1 - as for your non-clickbait content (not "bombastic" as you put it), it's pleasant to see you praise two good products, versus "midjourney KILLER has cometh"
Congrats Tim, been with you for awhile and you are my go to for AI and trying to figure things out, now clean that fingerprint off the polished part of the plaque please . . .
Hi...(ahhh I'm really sorry I forgot your name) Tim? First of all, the youtube award fully deserved. You create the perfect content for the subject matter you portray. This particular video I'm excited for. My friend uses midjourney and pays for it monthly. I've been wanting to try it myself for some time. Will download it today and have a go. Also, just an aside, I'm still using Scenario. That's a great bit of kit that I wouldn't know existed if it wasn't for you. Also, again thanks to you, I'm still not giving up on that comic creating AI software (although I'm not satsified with it yet). For commercial AI, you really have your finger on the pulse and we're really appreciative of it.
That's AWESOME to hear about Scenario! I need to do a follow up video on them soon. They're killing it over there! Flux-- yeah, if you're running a PC w/ 12gigs of VRAM or more, you should be good. If not, I might wait a week or so for someone to optimize the model. It's a bit heavy to run locally. But for sure, try it it out on Hugging Face and Fal. That said: If you DO have a beefy PC: You should be able to run Flux w/o a hitch!
Congrats on the 100k, started watching your channel awhile back and I don't tune in to every video but from what I do see... You never disappoint! Well deserved and can't wait for you to hit 1 Million!
FLUX is really doing a great job. People are complaining that the look isn't as good as MidJourney, but I get to put text in so I'm not really upset. Also, as you noted, hands look really good.
Congrats and good video! One thing you didn't look at is Flux vs. Niji Journey. In my tests, Flux falls apart more with hands and anatomy vs. niji. So Flux has some work to do in the anime/cartoon area there for those who need those styles. (I was using schnell for my tests). However, I feel that the composition and artistic arrangement over multiple variations were more interesting with Flux.
I think once we see it hit wider w/ the community, we'll see a lot more specially trained anime styles in Flux. I didn't test animation much, but I've seen some Gibili type images out there-- that said, yeah-- Niji has a MUCH wider range!
They've been talking about it. A few Office Hours back, David mentioned adding in ControlNet Pose features-- that'll be HUGE for MJ. Probably one of my biggest "wants"-- fingers crossed we see it in v6.2 (although, it'll probably be 7)
@@TheoreticallyMedia Hey man, at 100,000 subscribers you got to dress for success - it is a business! :-) That said the normal garb is already iconic :-)…. Keep on being awesome! :-) The kids at my AI still summer camp did very well, thanks in large part to you! :-)
Haha, I knew that line wasn't going to last long. Yeah, in the last few hours I've seen new Flux stuff popping up. Looking back, I probably shouldn't have even mentioned it! But, it was to my larger point: Since it's open source, all kinds of really amazing things are going to start popping up!
Congratulations on the subscriptions! Flux seems great though IMO not a Midjourney killer. I tested Flux a bit, comparing it prompt-to-prompt, personally I prefer Midjourney aesthetically. Though a bigger drawback to Flux for me is having to run it on my own computer. Additionally I really quite like MJ's --sref, I use it in almost every generation.
Running it on your own PC isn't a requirement, it is optional, and can be done so with SwarmiUI which, from what I can tell, runs on top of Stable Diffusion 3. I'm just trying to decide if I even want to run it locally vs. the web. You might find this vid helpful if you want to go down that road: th-cam.com/video/HKX8_F1Er_w/w-d-xo.html
1000% agree. That said, I think the important part with Flux is the open source aspect. It's looking REALLY good as a base model (not quite MJ, but the closest we've seen in my opinion)-- so, it'll really be once the community digs in and starts fine tuning that we're going to see some really remarkable results. MJ Killer? Nah-- MJ has it's own thing going on!
I've been with Midjourney for a couple years at 30/mo. Do you have to use that cmfy UI ? I'm either too old or too lazy to get into that interface. Even on the fusion page in Davinci Resolve I keep the node trees minimal.
You don't! Fal.AI for right now-- but this is early days-- I'd say within a week or so, Flux will be popping up on a bunch of platforms. That's sort of the beauty of open source. Trust me: That local run is a pretty big headache without doing a ton of research beforehand. And you'll likely drop a lot of money on new hardware to run it as well! It can be done, but time/money and all that...
@@TheoreticallyMedia Thanx! I'll see what comes up. Kling,Luma and Pixverse with a little Hedra when needed suits me for the present.I'll wait for the new platforms.
Just saw your vid here, love it! I did a back to back comparison of about 10 prompts with Midjourney (MJ) vs. Flux, and my first impression is that MJ is far better at imagination, thereby, giving a result that seems intuitive in its results, very artistic. With Flux, I would say that its clarity/resolution of a requested "color photograph" style is a lot better than MJ is. I can see how both AI tools could be used in a given project. A question for you? There are certain terms used on the Input form of Flux, that I don't understand, the help text for those settings are not helpful. Where I am confused is with "Num Inference Steps" and "Seed". Can you help with these, please? *Edit: I still need to test the Pro mode, my previous testing was on Schnell.
I just did some research and this is what I found: Number of inference steps: A larger number of steps increases the quality of the output but it takes longer to generate. For vanilla SDXL and Stable Diffusion 1.5 you should start with a value of about 20 steps. Don't go too high though, because after a point each step helps less and less. 50 steps is a good maximum. Seeds: Are essentially random numbers or vectors that are used to initialize various processes, ensuring that the results are not deterministic and exhibit desirable properties. (I'm just not sure when or under what possible conditions might utilizing the # of seeds would be good or not? Is this just a hit or miss, do you think?
Yup, you nailed it. I have noticed that if you take the inference steps up too high, you'll also prone to more weird distortions (too many arms etc)-- Seeds are all about you initial burst of "noise" that your image is created from. The best analogy here is when you look at a cloud and see a shape in it. A Bunny, a Man with a Hat, etc-- When you change your seed, you are changing the shape of that cloud, so the Image Generator will have a better chance of "seeing" something different. If you prompt for a Bunny, or a Man with a Hat, you'll still get those thing, but since you changed the shape of your cloud, they will appear in different positions and take on different characteristic. Hope that was helpful!
@@TheoreticallyMedia Thanks for the feedback. Have you installed SwarmyUI, and it appears that one can add the Flux part to that? I'm a rookie when it comes to having my own standalone AI....I guess the advantage is # of credits? Are there other considerations here? Although, I see I need 100-150 GB of HD space, Ouch! :) I would have to do some house cleaning on my 2 TB HDD to make it fit. Might be better to put this on my 2TB SDD as I have the space for that. Any feedback on going down this road would be appreciated! Again, thanks for the vid, and all your efforts!
Congratulations! So well deserved 👏👏✨ Could you also post a tutorial on its installation process for those who have failed miserably at keeping up with complex stable diffusion installations n updates process.. And had to give up 😅
Auraflow excels at creating images that blend text and faces - it can make pictures where half is text and half is a face, all in one seamless image. Flux.1, on the other hand, generates images and text separately. So if you want cool text-face hybrid images, Auraflow is the better choice, but if you need separate text and images, Flux.1 might be more suitable.
Yup-- I wasn't going to go over all that for a basics tour-- well, that and I couldn't get it running on Comfy on my end yet! haha-- I'll be futzing with it tonight. But, that's the overall point as well: What's so exciting about Flux is how everyone is going to use it in these amazing ways. Can't wait to see the next 3 weeks in Image gen!
in comfyUI it's quite easy to get img2img running, I built a workflow with two steps flux and 1 step sdxl. basically, first step of flux is responsible for the composition, so don't really care about the details, 2nd step is adding all the details, and then that output gets into sdxl with my fav loras and controlnets and gives me an image that will give midjourney run away with tail between it's legs.
Yup! I didn't want to go too far down the Comfy Rabbit hole, as this was more of an "intro" video-- and, haha, I didn't get it running in Comfy. But, for sure I'll do a follow up where I address it once the dust settles on Flux. That said: Man, did Flux kick up a lot of Dust!! If anything, what I was hoping to nail in this video is how important this model is going to be over the next year. We're going to see some pretty amazing stuff.
@@TheoreticallyMedia am really waiting to see the community fine tune the model on laion-aesthetics datasets. the only place where I think this model is losing to midjourney v6.1 is in the aesthetic department. But if someone fine tune it on that dataset, then this weakness can be solved. This is what we all were expecting sd3 to be like, I finally got my hope on open source art models back. And btw they did make scaled down versions of flux-s and flux-d. That fp8 ones. I tested it and as long as you increase the steps by a bit and use an fp16 clip and dtype with eular sampler and beta scheduler, then you actually get way better results than the original or base versions of flux models. Quite surprising (especially eular+beta, the dark horse for this model, no other combinations give better results than this combination for this model, while for all other models till now this specific combination used to give horrible results) Also you can run this on swarmUI too if you don't wanna go down the comfyUI rabbit hole, but it will cost more VRAM on swarmUI (idk why)
1:48 I saw a couple videos ago that you were fast coming up upon 100K....... what's an amazing milestone........and completely deserved! Looking forward to the 1 Million Sub Milestone! 🏆
I think that Flux has the better text and anatomy while Midjourney has the better aesthetic and skin texture. Also for the local part: I downloaded the FP8 Dev/Schnell Model and used ComfyUI (Not from Pinokio, the official one) and the Schnell model takes 5-10 seconds per image depending on my settings and the dev model takes 25-60 seconds per image. I have a 4070 TI Super 16GB VRAM and 64GB RAM, but there are people that could run the model with 6GB VRAM and 16GB RAM.
I went from being 100% ai ignorant to producing great images over one weekend. I installed it locally and I’m using the dev version. Takes about 1 or 2 minutes per image. I think it’s amazing. So easy to use.
@@TomLally I used forgeUI. I’ll dig up the video later when I’m home. It was pretty much a one click installation process. I’ve tried others previously but there was always other steps that were never explained in layman’s terms. This one worked a charm. I now have the foundations to learn the intricate and ‘autism level’ stuff
Thanks for another great video, and that's awesome you have past 100K subs! This is by far my favorite source for what up on the bleeding edge of AI software.
Id like to take the time to share my appreciation for your videos and showing us new ai stuff. You’re awesome, and congratulations on the 100k subscribers. I hope to see you reach 500k then 1 million. Thank you very much for keeping us updated. With that said, do you know if there’s any ai program that can take a song, and can recreate the instrumental?
Wait. You were wearing the BLUE SUIT for your acceptance speech! Does that mean the man in the blue suit will have to carry around a silver plaque from now on? Congratulations! You produce good stuff. Thanks for the explaining the Flux stuff. I was getting overwhelmed with incomplete info.
Can you please make detailed video of walkthrough of pinokio in mac? I can't use it in mac mini because of space issue and don't know how to use ssd for it. Also no idea how to use models and all
Please try this for me most image generator miss the detail "a painting of a young woman in a garden. The woman is wearing a white, Victorian-style dress adorned with ruffles and blue ribbons. She has brown hair styled in an updo, and she wears a small black hat with dark blue Feather. The woman is bending forward slightly, smelling a pink rose from a rose bush. Her hands are clasped behind her back, holding a closed parasol.She is wearing white shoes. The garden is lush and green, with various rose bushes in bloom and a stone wall in the background partially covered with vines."
Hi, Please Can I know, Pinokio version is work for my M1 Mac computer without any issue. I didn'tr install it yet. I need to get idea installing it to my local computer
Congrats on your 100k. Much deserved. Mind you, while I understand your pride in showcasing it, the focus was on TH-cam's plaque rather than your face the whole time. 😆
I tried Flux after watching this video, tried for a fantasy-realism scene with anime flavour that I had created in ChatGPT 4o's DALL-E and also with MidJourney. Flux was not able to add any sort of art style, no matter how I worded the prompt. It only created boring, generic, poorly rendered photo style image that was so boring without any dynamic angle, movement, creativity, uniqueness. I tried all sorts of prompts that MidJourney and DALL-E easily understand, but all I got was the very same pose in each slightly altered image, as though it was using the same identical seed each time. I also played around with the various slider settings ... results were the same boring ubiquitous drab crap. I think these models are trained on faces and on typical static posed images of people, maybe landscapes ... but if you throw anything creative at it, it defaults to static boring repetitive non-detailed images that are mind-numbingly boring.
100% agree on your Midjourney-killer comments. I'm getting pretty sick of all the click-bait titles where every new thing is supposedly the end of the competitor, and their whole team might as well jump off a bridge or whatever. But these content creators need content, so here we are.
@@kevnar Hey man, at 100,000 subscribers you got to dress for success - it is a business! :-) That said the normal garb is already iconic :-)…. Keep on being awesome! :-) The kids at my AI still summer camp did very well, thanks in large part to you! :-)
@@DirectorDanielMason It's not my channel, but I'm sure the boss appreciates the sentiment.
It's so tiring.. on Reddit things get especially bad, with vitriol aimed at Stability AI for releasing a 'broken' SD3 model.. people keep forgetting they're getting all this for free.
@@kevnar whoops- it responded to the wrong comment. Putting it back in the right place as well:)
100k fully deserved. You are the best AI channel on the internet bar none. Congrats.
Ahhh, bless you man!! Thank you so much!!
Couldn’t agree more!
Agreed, I only follow two AI people- Tim and Matt Wolfe
He is an image and video channel, AI explained and Bycloud are better at covering AI technology as a whole, and Mat Vid Pro is better at covering up scaling/music, while Mathew Bernard is better at covering open source LLM logic tests, they all have their specialties :)
You should check out ai explained and by cloud
Congratulations!!! It's a big achievement. For more to come! Flux Isn't the MJ's killer but I do think it's the best Open Source option at the moment. Flux Dev outputs can be used commercially, derivatives from the model can't. I thought the same but after someone pointed out, I re-read the license and was pleasantly surprised, as I will use it commercially. The difference is big between dev and schnell.
Yeah, I don't see it as a Midjourney killer either. As I mentioned in the video, MJ has at least one massive gamechanger in the pipeline: 3d.
That's going to be huge.
Ultimately: We can have two things. And two things are good!
BIG kudos to you for not using overly bombastic, click-bait titles, mate! And thanks for the great content. That plaque was well-earned!
Congratulations on the plaque! Well deserved.
Nice to see someone highlighting positive and negative aspects of all available solutions, instead of using terms like "killer". Yes, more is better.
2 is better than one! Unless it's a double order of pineapple on a pizza. That's a no!
I really appreciate the more relaxed and laid-back vibe of your video, as well as the non-ridiculous thumbnails, unlike some other channels.
Appreciate it! Haha, I'll say: I've had some fairly ridiculous thumbnails in the past! To be honest: I think they're ALL ridiculous!
Congrats on 100k subs!! :)
Thank you so much!!
Congrats, Tim! I need all the new releases to slow down long enough to catch up on the old ones! I just started using Hedra today... looks like I'll be adding Flux to the mix soon. I barely sleep anymore since I fell down the AI creation tools rabbit hole. Really enjoying your videos and so happy to see your success!
Congrats Tim! I love your content and appreciate the fine work you do.
grats on 100k, here's +1 - as for your non-clickbait content (not "bombastic" as you put it), it's pleasant to see you praise two good products, versus "midjourney KILLER has cometh"
Award winning vids here folks. Amazing, engaging and informative content.
Hi and congrats on 103,000! You got there fast, well done, good job on your content.
I wish a great generation for everyone.
Congratulations Tim!!!!! YOU so deserved it. There is NO one like you. Thank you for being there for us!
Congratz on the play button. You deserve it, truly the only straight forward AI channel on here.
Of course, the recognition is well earned, the reviews are always excellent. Keep up the good work, very grateful.Peace!!!
Thank you so much!! Truly means a lot!!
I enjoy your commentary on all the emerging media. Very helpful. Enjoy each of the vids when they pop up in my feed! Congrats on 100k
Congrats on the award! Well deserved!! 🎉
Thank you so much!!
Well deserved and well earned! Congrats on 100k! 😊
Can't wait to see you hit 250k Tim, won't be long. Thanks for always keeping me chuckling with your frequent humor. ;)
Hah! Thank you so much!! Man, 250 sounds insane...but, at one point 100k also did!
Congratulations, Tim! You have worked really hard for this. You deserve it. 🍸
Congrats on the 100k! Keep up the great work.
I've been looking for a creator that posts about AI a little more frequently than the ones I already follow. I'm glad I found this channel.
Finally!! great to see the live action version of the man in a blue biz suit. and congrats!!
Thank you so much!! I thought about doing the whole video in the blue suit...but--ehhh, I wanted to be comfortable!
Congrats on the 100k Tim! Keep up the good work!! :)
Congrats with your silver button. I love your channel. You do really great content. Hi from Kazakhstan.
an ai cjannel Without clickbait titles? Thank god! instant subscribe
Congrats Tim, been with you for awhile and you are my go to for AI and trying to figure things out, now clean that fingerprint off the polished part of the plaque please . . .
Hi...(ahhh I'm really sorry I forgot your name) Tim?
First of all, the youtube award fully deserved. You create the perfect content for the subject matter you portray.
This particular video I'm excited for. My friend uses midjourney and pays for it monthly. I've been wanting to try it myself for some time. Will download it today and have a go.
Also, just an aside, I'm still using Scenario. That's a great bit of kit that I wouldn't know existed if it wasn't for you.
Also, again thanks to you, I'm still not giving up on that comic creating AI software (although I'm not satsified with it yet).
For commercial AI, you really have your finger on the pulse and we're really appreciative of it.
That's AWESOME to hear about Scenario! I need to do a follow up video on them soon. They're killing it over there!
Flux-- yeah, if you're running a PC w/ 12gigs of VRAM or more, you should be good. If not, I might wait a week or so for someone to optimize the model. It's a bit heavy to run locally. But for sure, try it it out on Hugging Face and Fal.
That said: If you DO have a beefy PC: You should be able to run Flux w/o a hitch!
WOW! Congrats on 100K! Nice! You deserve it Tim
Bless the open source community. And congrats on 100k subscribers! Well deserved
Congrats man , You have always dropped the best game changers from an artists perspective
I appreciate that so much!!
Congratulations on the huge milestone!!! You earned it!
Congrats on crossing the 100k marker! Well earned and well deserved. Onward to the next. 👏👏
I think the next award is at 1 million-- so...I've got some time. Might need a new suit by then!
Great video as always. Please keep us posted on when flux gets tye features it is currently lacking. Love the Dev version output in your video!
Your channel is wonderful and one of my absolute favorites! Congratulations on 100k!!! ❤
Oooh got so jelly when I saw your silver plaque.
Congratulations!❤❤❤❤❤
Always an informative channel no bombastic needed. 😊
Not gonna lie, had the biggest, stupidest grin on my face when I saw the UPS guy!
Congrats on the 100k, started watching your channel awhile back and I don't tune in to every video but from what I do see... You never disappoint! Well deserved and can't wait for you to hit 1 Million!
Congrats on the award. I really appreciate your analysis on ai and your sense of humor
FLUX is really doing a great job. People are complaining that the look isn't as good as MidJourney, but I get to put text in so I'm not really upset. Also, as you noted, hands look really good.
Congratulations! Well deserved. You have a really good channel
anybody knows what is the service with inpainting mentioned at 8:20? something along "wan", "woan", "whan"??
Ah, that is Wand! Sorry, I forgot to put that in the Description! www.wand.app/
Note that the FLUX demo he showed is not live yet. We're working on adding it right now though. The App Store version is still SDXL
Congrats on hitting 100K mate well done, quality work showing where we are at. Looks like I'm gonna need to upgrade my hardware.
Congrats on the 100K, well deserved! And thanks for keeping us up to date!
You are amazing, man. I love you. You have one of the best channels, truly amazing. Please try to inform us about any free AI tools, and thank you.
Ah, thank you so much!! Hugely appreciated!
Hi, where are you getting the community examples? Would like to see more. Thanks.
Congrats and good video! One thing you didn't look at is Flux vs. Niji Journey. In my tests, Flux falls apart more with hands and anatomy vs. niji. So Flux has some work to do in the anime/cartoon area there for those who need those styles. (I was using schnell for my tests). However, I feel that the composition and artistic arrangement over multiple variations were more interesting with Flux.
I think once we see it hit wider w/ the community, we'll see a lot more specially trained anime styles in Flux. I didn't test animation much, but I've seen some Gibili type images out there-- that said, yeah-- Niji has a MUCH wider range!
Gratz on your plaque dude, you TOTALLY deserve it!!
Thank you so much! NGL: I was a little giddy when I saw the mailman roll up!
As it uses Comfy UI does that mean you can use Flux within Krita and Photoshop?
7:18 this is pretty incredible. two distinct stylings within one output, where there is no blurring or hallucination effects. that's pretty wild.
That was SUPER impressive! Really showcases the prompt coherence as well!
Congrats on the 100k big guy!
Congrats on the 100k, you deserve it!
This is great news. Congratulations on 100K! 🙂
Congrats on the TH-cam Silver Creator Award! Woohoo! Well-deserved!
thank you so much!! I'm still kind of floored by it!
Been gone a while, but still supporting you, Tim. Congratulations! 💯 Got Travolta just right. 😆Must try Flux! Thank you as always, and keep at it. 👍
Looks awesome! Can't wait to try it out! Off topic; does Midjourney have image guidance like Leonardo and if not is it likely to maybe get it?
They've been talking about it. A few Office Hours back, David mentioned adding in ControlNet Pose features-- that'll be HUGE for MJ. Probably one of my biggest "wants"-- fingers crossed we see it in v6.2 (although, it'll probably be 7)
Congrats on the TH-cam milestone! 👏👏
appreciate that!! Thank you so much!!
Congratulations to 100k!
Fantastic info Tim! And congrats on the award, you deserve it!
Congrats on the Play button 👏🏾
appreciate it! Thank you so much!!
Congratulations!!🎉🎉🎉well deserved 👏🏽👏🏽👏🏽
Thank you so much!!
@@TheoreticallyMedia one day it’ll be my turn😊
Congratulations Brother, well earned...your content is classy and top-notch!
You now *are* the man in the blue business suit - love the costume change, and congrats! :-)
Haha, thank you so much!! that suit has sat in my closet for a LONG time, waiting for its moment!!
@@TheoreticallyMedia Hey man, at 100,000 subscribers you got to dress for success - it is a business! :-) That said the normal garb is already iconic :-)…. Keep on being awesome! :-) The kids at my AI still summer camp did very well, thanks in large part to you! :-)
You deserve more than 100k. Always great content
Image to Image works, There are some workflows available. They also work as an upscaler too.
Haha, I knew that line wasn't going to last long. Yeah, in the last few hours I've seen new Flux stuff popping up. Looking back, I probably shouldn't have even mentioned it! But, it was to my larger point: Since it's open source, all kinds of really amazing things are going to start popping up!
congrats on 100k
Thank you so much 😀
Having so much fun with Flux. There's something about running this stuff locally that makes me really happy.
Is there a difference running it locally rather than the demo?
Congratulations on the subscriptions! Flux seems great though IMO not a Midjourney killer. I tested Flux a bit, comparing it prompt-to-prompt, personally I prefer Midjourney aesthetically. Though a bigger drawback to Flux for me is having to run it on my own computer. Additionally I really quite like MJ's --sref, I use it in almost every generation.
Running it on your own PC isn't a requirement, it is optional, and can be done so with SwarmiUI which, from what I can tell, runs on top of Stable Diffusion 3. I'm just trying to decide if I even want to run it locally vs. the web. You might find this vid helpful if you want to go down that road: th-cam.com/video/HKX8_F1Er_w/w-d-xo.html
1000% agree. That said, I think the important part with Flux is the open source aspect. It's looking REALLY good as a base model (not quite MJ, but the closest we've seen in my opinion)-- so, it'll really be once the community digs in and starts fine tuning that we're going to see some really remarkable results.
MJ Killer? Nah-- MJ has it's own thing going on!
Congrats Tim ❤big love overdue well deserved
I've been with Midjourney for a couple years at 30/mo. Do you have to use that cmfy UI ? I'm either too old or too lazy to get into that interface. Even on the fusion page in Davinci Resolve I keep the node trees minimal.
You don't! Fal.AI for right now-- but this is early days-- I'd say within a week or so, Flux will be popping up on a bunch of platforms. That's sort of the beauty of open source.
Trust me: That local run is a pretty big headache without doing a ton of research beforehand. And you'll likely drop a lot of money on new hardware to run it as well!
It can be done, but time/money and all that...
@@TheoreticallyMedia Thanx! I'll see what comes up. Kling,Luma and Pixverse with a little Hedra when needed suits me for the present.I'll wait for the new platforms.
Well done man hard work pays off! Sometimes!
Just saw your vid here, love it!
I did a back to back comparison of about 10 prompts with Midjourney (MJ) vs. Flux, and my first impression is that MJ is far better at imagination, thereby, giving a result that seems intuitive in its results, very artistic. With Flux, I would say that its clarity/resolution of a requested "color photograph" style is a lot better than MJ is. I can see how both AI tools could be used in a given project.
A question for you? There are certain terms used on the Input form of Flux, that I don't understand, the help text for those settings are not helpful. Where I am confused is with "Num Inference Steps" and "Seed". Can you help with these, please?
*Edit: I still need to test the Pro mode, my previous testing was on Schnell.
I just did some research and this is what I found:
Number of inference steps:
A larger number of steps increases the quality of the output but it takes longer to generate. For vanilla SDXL and Stable Diffusion 1.5 you should start with a value of about 20 steps. Don't go too high though, because after a point each step helps less and less. 50 steps is a good maximum.
Seeds:
Are essentially random numbers or vectors that are used to initialize various processes, ensuring that the results are not deterministic and exhibit desirable properties. (I'm just not sure when or under what possible conditions might utilizing the # of seeds would be good or not? Is this just a hit or miss, do you think?
Yup, you nailed it. I have noticed that if you take the inference steps up too high, you'll also prone to more weird distortions (too many arms etc)-- Seeds are all about you initial burst of "noise" that your image is created from. The best analogy here is when you look at a cloud and see a shape in it. A Bunny, a Man with a Hat, etc--
When you change your seed, you are changing the shape of that cloud, so the Image Generator will have a better chance of "seeing" something different. If you prompt for a Bunny, or a Man with a Hat, you'll still get those thing, but since you changed the shape of your cloud, they will appear in different positions and take on different characteristic.
Hope that was helpful!
@@TheoreticallyMedia Thanks for the feedback. Have you installed SwarmyUI, and it appears that one can add the Flux part to that? I'm a rookie when it comes to having my own standalone AI....I guess the advantage is # of credits? Are there other considerations here? Although, I see I need 100-150 GB of HD space, Ouch! :) I would have to do some house cleaning on my 2 TB HDD to make it fit. Might be better to put this on my 2TB SDD as I have the space for that. Any feedback on going down this road would be appreciated! Again, thanks for the vid, and all your efforts!
Congratulations! So well deserved 👏👏✨ Could you also post a tutorial on its installation process for those who have failed miserably at keeping up with complex stable diffusion installations n updates process.. And had to give up 😅
Congratulation for getting the Playbutton!
appreciate that!! I'm still kind of stunned by it!
@@TheoreticallyMedia whenever you have any doubt in what you do, you can always turn and look at it and think "I deserved that thing" ;)
Auraflow excels at creating images that blend text and faces - it can make pictures where half is text and half is a face, all in one seamless image. Flux.1, on the other hand, generates images and text separately. So if you want cool text-face hybrid images, Auraflow is the better choice, but if you need separate text and images, Flux.1 might be more suitable.
in Comfy you could aways do image to image and upscale using flux
Yup-- I wasn't going to go over all that for a basics tour-- well, that and I couldn't get it running on Comfy on my end yet! haha-- I'll be futzing with it tonight.
But, that's the overall point as well: What's so exciting about Flux is how everyone is going to use it in these amazing ways. Can't wait to see the next 3 weeks in Image gen!
Congrats Tim!!🎉❤ I have Felt like we are brothers from another Mother! I’m trying to be right behind you! Heh! But may be year away the way I’m going!
I am already using it, and I am super impressed
in comfyUI it's quite easy to get img2img running, I built a workflow with two steps flux and 1 step sdxl. basically, first step of flux is responsible for the composition, so don't really care about the details, 2nd step is adding all the details, and then that output gets into sdxl with my fav loras and controlnets and gives me an image that will give midjourney run away with tail between it's legs.
Yup! I didn't want to go too far down the Comfy Rabbit hole, as this was more of an "intro" video-- and, haha, I didn't get it running in Comfy. But, for sure I'll do a follow up where I address it once the dust settles on Flux.
That said: Man, did Flux kick up a lot of Dust!! If anything, what I was hoping to nail in this video is how important this model is going to be over the next year. We're going to see some pretty amazing stuff.
@@TheoreticallyMedia am really waiting to see the community fine tune the model on laion-aesthetics datasets. the only place where I think this model is losing to midjourney v6.1 is in the aesthetic department. But if someone fine tune it on that dataset, then this weakness can be solved.
This is what we all were expecting sd3 to be like, I finally got my hope on open source art models back.
And btw they did make scaled down versions of flux-s and flux-d. That fp8 ones. I tested it and as long as you increase the steps by a bit and use an fp16 clip and dtype with eular sampler and beta scheduler, then you actually get way better results than the original or base versions of flux models. Quite surprising (especially eular+beta, the dark horse for this model, no other combinations give better results than this combination for this model, while for all other models till now this specific combination used to give horrible results)
Also you can run this on swarmUI too if you don't wanna go down the comfyUI rabbit hole, but it will cost more VRAM on swarmUI (idk why)
With the flux safetensors file it works on my macbook M1. But takes about 15 minutes to create an image.
1:48
I saw a couple videos ago that you were fast coming up upon 100K....... what's an amazing milestone........and completely deserved!
Looking forward to the 1 Million Sub Milestone! 🏆
I think that Flux has the better text and anatomy while Midjourney has the better aesthetic and skin texture. Also for the local part: I downloaded the FP8 Dev/Schnell Model and used ComfyUI (Not from Pinokio, the official one) and the Schnell model takes 5-10 seconds per image depending on my settings and the dev model takes 25-60 seconds per image. I have a 4070 TI Super 16GB VRAM and 64GB RAM, but there are people that could run the model with 6GB VRAM and 16GB RAM.
Is it better downloading it locally rather than using it on their website? The free versions, Dev and Schenl.
@@TomLally It depends on your PC.
@@risphereeditor yeah I’ve only got a 1070 8gb😅 16gb ram.
@@TomLally it may run!
I went from being 100% ai ignorant to producing great images over one weekend. I installed it locally and I’m using the dev version. Takes about 1 or 2 minutes per image. I think it’s amazing. So easy to use.
How do you install it locally? Is it comfy Ui or something? As I’m using Fooocus currently. Much appreciated.
@@TomLally I used forgeUI. I’ll dig up the video later when I’m home. It was pretty much a one click installation process. I’ve tried others previously but there was always other steps that were never explained in layman’s terms.
This one worked a charm.
I now have the foundations to learn the intricate and ‘autism level’ stuff
Thanks for another great video, and that's awesome you have past 100K subs! This is by far my favorite source for what up on the bleeding edge of AI software.
I don't think the Mac local Flux download link works anymore? Is there an updated link available?
🎉Congratulations.
Id like to take the time to share my appreciation for your videos and showing us new ai stuff. You’re awesome, and congratulations on the 100k subscribers. I hope to see you reach 500k then 1 million. Thank you very much for keeping us updated.
With that said, do you know if there’s any ai program that can take a song, and can recreate the instrumental?
Great review for the FLUX model, I am running it Locally, by far the hype is real.
Wait. You were wearing the BLUE SUIT for your acceptance speech! Does that mean the man in the blue suit will have to carry around a silver plaque from now on? Congratulations! You produce good stuff. Thanks for the explaining the Flux stuff. I was getting overwhelmed with incomplete info.
@theoreticallyMedia Great review! What are the titles of the songs that are playing in the background?
Try hiting refresh and selecting a models in comfyui
Can you please make detailed video of walkthrough of pinokio in mac? I can't use it in mac mini because of space issue and don't know how to use ssd for it. Also no idea how to use models and all
once I get it up and running on my end, I'll do that!
Great video! Thanks for sharing !
Congratulations on your 100k 🎉🎉🎉
Please try this for me most image generator miss the detail "a painting of a young woman in a garden. The woman is wearing a white, Victorian-style dress adorned with ruffles and blue ribbons. She has brown hair styled in an updo, and she wears a small black hat with dark blue Feather. The woman is bending forward slightly, smelling a pink rose from a rose bush. Her hands are clasped behind her back, holding a closed parasol.She is wearing white shoes. The garden is lush and green, with various rose bushes in bloom and a stone wall in the background partially covered with vines."
Hi, Please Can I know, Pinokio version is work for my M1 Mac computer without any issue. I didn'tr install it yet. I need to get idea installing it to my local computer
100K achievement unlocked! Now for 500K!!
(will be quicker than you think! Congrats!)
That boggles the mind!! Thanks so much!!
Congrats on your 100k.
Much deserved.
Mind you, while I understand your pride in showcasing it, the focus was on TH-cam's plaque rather than your face the whole time.
😆
Haha, my stupid camera changed autofocus modes! I'm actually picking up a small monitor today. This is the third time that's happened!
Thanks for the intro to Flux! My two interests are character consistency and comic book characters. Any idea how it works with that?
Keep doing what you love. You are doing a great job and your videos are amazing!!!!!
I tried Flux after watching this video, tried for a fantasy-realism scene with anime flavour that I had created in ChatGPT 4o's DALL-E and also with MidJourney. Flux was not able to add any sort of art style, no matter how I worded the prompt. It only created boring, generic, poorly rendered photo style image that was so boring without any dynamic angle, movement, creativity, uniqueness. I tried all sorts of prompts that MidJourney and DALL-E easily understand, but all I got was the very same pose in each slightly altered image, as though it was using the same identical seed each time. I also played around with the various slider settings ... results were the same boring ubiquitous drab crap.
I think these models are trained on faces and on typical static posed images of people, maybe landscapes ... but if you throw anything creative at it, it defaults to static boring repetitive non-detailed images that are mind-numbingly boring.