In case you missed it, here's yesterday's runthrough of Pika 1.5, Kling's new lip-sync, a new player on that front, and Minimax is getting image to video! ua-cam.com/video/qW7uaxBz_-Y/v-deo.html
@@VisionOfFury Yes it's confirmed but I don't think you can just assume from how good the text to image is... video is substantially harder and I wouldn't be surprised to see them behind other players.
@@MiraPloyIt’s not harder necessarily, it’s more complex but they have a great dataset. You can literally have chatGPT build you a styleGAN 3 with VAE and then add more functionality to achieve great results The main issue for normies is finding, structuring, and processing insane amounts of data but the technology isn’t that hard to understand (not saying I can code like a pro) but like I said, I tested it with chatGPT.
You cannot use it on your own. And also I love charts that doesn't start from zero - you know the charts where you could think that the results are much better than they realy are 🤫
Haha, I literally just got back from Europe this week-- and I'll say, after a week of drinking tiny espressos-- I get it. It packs a punch! First time I've had that sweating from coffee overload feeling in awhile!
Tim, thank you so much for these videos. Its very difficult to keep up with everything thats going on. As for FLUX, I still love my MJ, but I do tend to gravitate to FLUX when I need a certain image fast! 90% of the time, when I prompt in FLUX, I get the image I want one the first or second roll. Its unreal!!
Prompt coherence is kind of crazy good with Flux! Still love MJ, and generate there at least one session a day. I was going to put a bit in here about the “Midjourney killer” aspect, but I didn’t want to add fuel to the fire. That said; I know MJ is working on some really interesting stuff!
@@TheoreticallyMedia I agree. Love MJ and I will never kill my Sub. FLUX is simply stunning and hits the mark 90% of the time on the first one or two rolls. Thanks Tim!!
Thank you, Tim Just a suggestion - perhaps we can compare styles other than photorealism (stock or otherwise) and 'standard' digital art, and more on adherence as well. I'd love to see what it can do with AI-assisted traditional art styles, mixed media collages, and crazy illusion/surreal photorealism, for example (off the top of my head). That crossed my mind at 8:50. The photorealism is insane, especially with upscalers, but that right there is underwhelming as far as textures are concerned, granted that could have a lot to do with the prompt. Thanks as always, Tim. 👍Also, I believe all generations of a man in a blue business suit will get exponentially better on _any_ platform now! 😆
Nice vid again Tim! I've been really liking flux's versatility. It's got ideogram level of text adherence, you can make logos, skylines with specific billboards, you can create hyper-realistic people, you can create more artistic illustrations. And prompting is optimized for natural language instead of you trying to trick it with hot words, weights etc.
Flux still has a lot of problem, but its base quality reminds me a lot on professional canon slr pics, and it is nice to see the base of a scenery from a prompt made it into the picture.
Just a note. You mention the cost of images being .004 cents a couple of times, when you meant 4 cents (i.e. 0.04 dollars). Big difference if you generate a bunch!
Hi! Does anyone knows how to have LOGARTIHMIC-TONE videos with KlingAI 1.5? I have tried lot of options (low contrast, ARRI log, soft light...) but it always creates very high contrast videos, and almost always the sun is at a side-front position, creating backlight and flares. Thanks! (I have no problems with KlingAI 1.0, only with 1.5)
Flux 1 dev is a good model, but on my computer at Comfy it generates the same thing constantly, even with a randomized seed. Having to modify the promt very much. About the Flux 1.1, I guess we have to wait for the next model. It's like with 1.5,then 2.0,then XL,then SD 3 and now Flux. That is, through one generation of models, the guys are creating really something incredibly good. Thanks, Detective.
@@TheoreticallyMedia Amen! Also, loved your Batman & Robin joke lol. I have a feeling that these black boxes aren't capable of perfectly generating exact replicas of a specific training images. It's just not in the guts and token roulette is a good way to put it. It's our beautiful uncanny, evergreen valley / Black Mirror that will never look right. I think "they" (Lora & Unsaf Etensor, for example) never get unfomfortable symmetry differences or 'creepy eyes' in the same way that Patrick Bateman will never grasp the feeling of empathy. Absolutely nuts, either way! "Game changer" and "what a time to be alive" are pretty overused, but you get it lol. ¯\_(ツ)_/¯ You could probably use the camera models as a variable in the prompt, in batch, with Py/Comfy and make a big fleshy imagemap to find trends around location / occupation / age / being a hidden creeper on the roof / government bulk sales etc... and see that data ya know? One could do a massive export of prompts with all of the Japanese and different lens/camera combos and hone in on patterns. Maybe get Llava to look at each of the images in its Comfy little Node Workshop, then keep those tags fed via API to LLMStudio, OpenAI or Teddy Ruxpin. Keep the data automated on Sheets/etc. The whole cerebral->built-digital-stuff timeline is shrinking like a teenage Moranis up in this mofo. AI's ultimate use case, if you ask me (and you shouldn't) is to generate the brochure for aliens when they arrive. AI absolutely packages everything up perfectly, but it doesn't always understand the actual content. I don't mind this awkward state, but it wont be around long enough to really get annoying/endearing/reassuring. Right now, it misses the stuff I assume a robot would miss, but images and VOICE is hard to fathom. TTS, specifically Bark, is a true blackbox, and it was the first time I felt that "oh, ok wow" feeling from AI. I think my brain responds to audio with more emotion than visual, but we all have our 𝐡𝐨𝐥𝐲 𝐬𝐡𝐢𝐭 𝐢 𝐧𝐞𝐞𝐝 𝐚 𝐬𝐞𝐜 moments! 😅 It will be spooky when those gaps are suddently gone and surprises are less common, in general. Random thoughts about the uncanny valley that sparked from that bizarre camera model image... For me, the whole concept, and seeing/hearing how we're summarized from the angle of the 'hive' / AI, feels extremely uncomfortable because I have to assume we may not always be relevant. The bell's rung! 🔔#no-unrings - humans usually 𝙟𝙪𝙨𝙩 𝙙𝙤 𝙞𝙩. after learning that it exists especially if 💰🚀!!! But yea, that scenario threatens something primal and people have been going wild arguing about AI , which is not my scene. Great video, man. Spot on with the upscalers. That transition/wipe really shows it. I love Topaz Labs' models, personally! I worked there and built the CX setup in Austin. So much fun working there, they are all on another level - super humbling to be around. Anyways! I'll always throw their models out as suggestions for upscaling, but... they're not as Comfy as they could be. I mean, wearing a suit isn't always Comfy! Can't make cash wearing opensource sweatpants, AMIRITE? 😂 So many new ways to profit, which is cool, but there are also so many emotions wrapped up into it and I didn't expect the polarization of AI art. I'm an artist. Art will be fine. Being able to try all of this COOL SHIT right as it hits github/humanity/HF is exhiliarting because this is totally my scene, and I truly have no clue where these models will take us, but we kinda do because it's our idea in the first place! As the kids say, "yolo....(v8-face ↑)" Until I see AI doing the bad stuff, I'm in the honeymoon phase avoiding the slop.
General question. I keep not seeing textures or inherent detail on many of these models. When looking at "immersive" experiences, and the point of Unreal's Nanite was about inherent detail added to games systems. So when do we see that, or is there a model you'd recommend with that?
For this one, I just used the built in one that Freepik has. But I use them all: Magnfic, Leonardo's, Krea's-- They all kind of do different stuff. They're all unicorns!
This is the first time I write 3 comments in a video… (annoying I know) sorry for that! but that pika comment at the end was so you! love it! Love your sarcasm! have a great night and thanks for another great video!
Tim. I work a lot in MJ and use it for work professionally. Do you think flux is better than MJ? From the looks here, it doesn't seem to eclipse it but I haven't used it much... just a few rolls here and there.
I used both MJ and FLUX and I found FLUX to be better. MJ does not follow promts that good. When I ask it to make photo it often does not. MJ is more creative though. It all depends on what you need more. If you can use FLUX on your own PC then go with it as you can do inpainting, image2image, styles, workflows and so much more. I also train faces with it though it takes 6-7 hours.
I’d say MJ (to my taste) tends to be more eye pleasing out of the gate, whereas with Flux, you tend to need to work at it a bit with some post processes. On the photographic side, I’m actually starting to lean toward Flux in terms of realism though. Although, you do (currently) need that extra creative upscale to get there!
I finally signed up to test Flux. Where I found it really lacks, is the exact same as Dall-E. Creating realistic food photos. I advertise for a lot of restaurants and Midjourney is exceptional at food pictures, especially Indian food. Flux just fell flat.
Oh, MJ kills it at food! I once gave it the ingredients to a dish as a prompt, and although I know it was just keywords as tokens, I was still amazed that it “made” the dish! I’ve seen some good hamburgers with flux, but I can see Indian dishes kind of failing. …man, now I want a Samosa!
Easiest? Likely Pinokio. I don't go over Flux in particular in this video, but I do go over Pinokio installation here: ua-cam.com/video/3pp7qw19nuA/v-deo.htmlsi=tmOuJhXY15jqhIqw
Tried Flux Pro 1.1 on my provider Freepik and am most disappointed regarding realism. Same prompt of a photo that looks photorealistic on both Flux Dev and Mystic ( Freepiks Checkpoin on Flux basis) turned out as a digital art portrait on Flux pro 1.1.
There's an old Simpsons joke that always comes to mind-- it's a Homer line: "Oh, people can come up with statistics to prove anything, Kent. 14% of people know that". That joke is what, 25 years old? Still holds up!
Hah- somehow it got lost in the edit, but when we got to the viking woman, I had the line: "Now THAT'S a look that says you're going into a trap!" I blame the editor...who, ironically, is also me!
Just follow the links in the video description. For a more general approach there are a lot more videos on his channel that go into detail on how to set things up.
:-) absolutely. - Although Flux does often have problems with the sizes of objects the coffee cup FLUX created is about exactly the typical size you would get in a coffee shop for a normal coffee in Europe.
That seemed to be their thing with the increased diversity. I mean, to be fair, MJ v4 had a LOT of that "same" face issue as well. Flux is only at 1.1-- so, give them time to spread their wings.
well, it is-- just not the Pro models. Black Forest Lab are made up of Stability employees, so-- I think they're trying to head off the mistakes that led to Stability's fall (that's a funny phrase to type)-- Not to say that Stability is done for-- I'm still rooting for a comeback.
My problem with AI and I would hope I please interview an artist that knows about anatomy The problem with AI is it's not trained on anatomy it's just trained on the final result of an image not the steps to actually create anatomy this is the major issue with them and why you get these hands that are too small joints and all that because it's only trained on the final result And if you could interview an artist that knows about anatomy and shape theory it would expose the flaw of AI it has no anatomy and shape theory coherence
Haha, it can go overboard from time to time. The best bet for that is to focus on the promoting with the upscaler and give it some weights like, beautiful::1, realistic skin, ((wrinkles)) Try that out. Although, you might have to play with it to not get a plastic character!
Yeah, I'm going to pass for now and wait a bit longer. Stylized super models are nice, but a bit overdone. Then there are the small immediately eye catching flaws. SD 1.5 forever!! lol
Totally get you! SD 1.5 really was a golden era! On the plus side: Flux is still only at 1.1, it has some room to grow! And containing Flux is ex-SD team members, I have NO doubts it’s going to grow to become an amazing model!
5:00 Looks like their nipple censoring is still causing nipple jewellery... sigh. My guess is their filtering added the little balloons to nipple bumps not just bare nipples.
OK. I spent $1 and made a nice, excellent image. And what now? What am I going to do with it? Save it in a cloud. And what then? Spend another $1 for another nice image? To get addicted to make nice images? It's called milking cows. Thank, but not me. .
To each their own. As with any act of creativity, value is hard to quantify. Tangentially related, I think that's a byproduct of digital media. My camera roll (on my phone) and cloud backups are FILLED with thousands of moments from my life. But-- do I ever really go back and look at them? It's funny, obviously I'm old-- so, the era of my childhood is really summed up in a handful of photo albums. My kids, on the other hand? I think every moment has been documented...
Really? Better? I thought they were "perfect"? I thought they had revolutionized the industry. I thought they had "turned the corner" after SD3. Now... upgrade? What? Shocking.
Everything’s upgrading, all the time, everyday. I get clickbait words might be annoying and even angering, but it’s probably best to just let it go. When news comes out it will always be sensationalized, and always has
@@RealityRogue I'm a curmudgeon. There is nothing offensive or criminal that I let slide without comment. What happens after is up to you. These people are feeding you ground meat while calling it "steak". I think you deserve better.
@@dirtydevotee Hey fair enough. Some people tend to let their emotions control them on the internet and get angered without even a thought. But you obviously are aware, and that’s enough for me. There’s always a time and place for people with tough opinions I’d say
Eh, I get it. Look, I think the content of the video is fairly balanced. I think it's a great model for a 1.1 with room to grow-- but, I also understand the frustration w/ thumbnails and titles. The thing is: That's UA-cam in a nutshell. Attention economy and all that-- you gotta play the game to get the message out.
Flux is probably one of the most useless pieces of crap ever It says I have 10 daily credits and when I try to do a prompts then it says I don’t have enough credits
i know nothing about this channel and i really don't like the name theoretical media....you could put two random words and it would be better .. cheeseClick doubleCake steveTake
In case you missed it, here's yesterday's runthrough of Pika 1.5, Kling's new lip-sync, a new player on that front, and Minimax is getting image to video! ua-cam.com/video/qW7uaxBz_-Y/v-deo.html
Yo!
Hi please drop a video about deep live cam
how to create consistent characters in flux pro?
Flux is CRUSHING it. The text2video is going to be mind blowing if their image model is any indication.
I cannot WAIT to try it!
is txt2vid confirmed?
@@VisionOfFury Yes. It's on their website under "What's next?"
@@VisionOfFury Yes it's confirmed but I don't think you can just assume from how good the text to image is... video is substantially harder and I wouldn't be surprised to see them behind other players.
@@MiraPloyIt’s not harder necessarily, it’s more complex but they have a great dataset. You can literally have chatGPT build you a styleGAN 3 with VAE and then add more functionality to achieve great results
The main issue for normies is finding, structuring, and processing insane amounts of data but the technology isn’t that hard to understand (not saying I can code like a pro) but like I said, I tested it with chatGPT.
Faster and more efficient Flux is all I needed to hear. My RTX 3080 with only 10gb of VRAM loves efficient models.
It's not open source :(
You cannot use it on your own. And also I love charts that doesn't start from zero - you know the charts where you could think that the results are much better than they realy are 🤫
Why don’t you just use colab?
Gay
@@antoninsvrdlik7526 you can for schnell and dev version on your own pc
I was running both schnell and dev (from full version through fp8, GGUF and NF4) on literally 1060 3GB VRAM. It just works!
WOW- -that's impressive! Hmmm, maybe it's time to dust off my old Dell Potato!
A1111?
Where did you get the models?
How fast for 1024 by 1024 on that?
@@jac001 Forge UI😊Also Comfy.
I love your content Tim. Very in depth, easy to follow and adding cheesy humor is a winning format.
The coffee cup was actually a very good size. Barista here.
Haha, I literally just got back from Europe this week-- and I'll say, after a week of drinking tiny espressos-- I get it. It packs a punch! First time I've had that sweating from coffee overload feeling in awhile!
I can't wait for the next versions of Flux Dev and Schnell.
Especially since the community got quite a bit of experience with V1.
likely won't be any. that was just to boost popularity and get their name on the map
3:29 There is a car driving in the wrong direction in the lower right corner. Brace for impact.
7:44 "That's not a mug. THIS is a mug!"
Holy s#!+ Tim, that is definitely a mug! Kudos!
excited for the video gen model from black forest labs team. can't wait to test it
REALLY excited about that one as well!
Tim, thank you so much for these videos. Its very difficult to keep up with everything thats going on. As for FLUX, I still love my MJ, but I do tend to gravitate to FLUX when I need a certain image fast! 90% of the time, when I prompt in FLUX, I get the image I want one the first or second roll. Its unreal!!
Prompt coherence is kind of crazy good with Flux! Still love MJ, and generate there at least one session a day. I was going to put a bit in here about the “Midjourney killer” aspect, but I didn’t want to add fuel to the fire.
That said; I know MJ is working on some really interesting stuff!
@@TheoreticallyMedia I agree. Love MJ and I will never kill my Sub. FLUX is simply stunning and hits the mark 90% of the time on the first one or two rolls. Thanks Tim!!
Thank you, Tim Just a suggestion - perhaps we can compare styles other than photorealism (stock or otherwise) and 'standard' digital art, and more on adherence as well. I'd love to see what it can do with AI-assisted traditional art styles, mixed media collages, and crazy illusion/surreal photorealism, for example (off the top of my head). That crossed my mind at 8:50. The photorealism is insane, especially with upscalers, but that right there is underwhelming as far as textures are concerned, granted that could have a lot to do with the prompt. Thanks as always, Tim. 👍Also, I believe all generations of a man in a blue business suit will get exponentially better on _any_ platform now! 😆
Nice vid again Tim! I've been really liking flux's versatility. It's got ideogram level of text adherence, you can make logos, skylines with specific billboards, you can create hyper-realistic people, you can create more artistic illustrations. And prompting is optimized for natural language instead of you trying to trick it with hot words, weights etc.
Great I waited for someone to cover Flux 1.1! :)
Haha, as soon as I saw it this morning, I was like: Well, I PLANNED to take today off...sigh
Flux still has a lot of problem, but its base quality reminds me a lot on professional canon slr pics, and it is nice to see the base of a scenery from a prompt made it into the picture.
Just a note. You mention the cost of images being .004 cents a couple of times, when you meant 4 cents (i.e. 0.04 dollars). Big difference if you generate a bunch!
Ahhhh- thank you! The math was a little screw on .04 for 25 images. Good call. Still, super cheap!
@@TheoreticallyMediaInsert office space jokes here..
The thumbnail master strikes again! Absolute ninja!
Hi! Does anyone knows how to have LOGARTIHMIC-TONE videos with KlingAI 1.5? I have tried lot of options (low contrast, ARRI log, soft light...) but it always creates very high contrast videos, and almost always the sun is at a side-front position, creating backlight and flares. Thanks!
(I have no problems with KlingAI 1.0, only with 1.5)
Flux 1 dev is a good model, but on my computer at Comfy it generates the same thing constantly, even with a randomized seed. Having to modify the promt very much.
About the Flux 1.1, I guess we have to wait for the next model. It's like with 1.5,then 2.0,then XL,then SD 3 and now Flux. That is, through one generation of models, the guys are creating really something incredibly good.
Thanks, Detective.
So glad we have a good uncensored model that is (currently) open source unlike Stable Diffusion which sold out
I realy like that white/blue painting at 8:49
Yeah! totally random-- to be honest, that was a lot of fun-- just seeing what popped out!
this once again proves my theory that Germans love David Hasselhoff.
It's been far too long since I've heard that line! Rest in Peace, Norm!!
@@TheoreticallyMedia Amen! Also, loved your Batman & Robin joke lol. I have a feeling that these black boxes aren't capable of perfectly generating exact replicas of a specific training images. It's just not in the guts and token roulette is a good way to put it. It's our beautiful uncanny, evergreen valley / Black Mirror that will never look right. I think "they" (Lora & Unsaf Etensor, for example) never get unfomfortable symmetry differences or 'creepy eyes' in the same way that Patrick Bateman will never grasp the feeling of empathy. Absolutely nuts, either way! "Game changer" and "what a time to be alive" are pretty overused, but you get it lol. ¯\_(ツ)_/¯
You could probably use the camera models as a variable in the prompt, in batch, with Py/Comfy and make a big fleshy imagemap to find trends around location / occupation / age / being a hidden creeper on the roof / government bulk sales etc... and see that data ya know? One could do a massive export of prompts with all of the Japanese and different lens/camera combos and hone in on patterns. Maybe get Llava to look at each of the images in its Comfy little Node Workshop, then keep those tags fed via API to LLMStudio, OpenAI or Teddy Ruxpin. Keep the data automated on Sheets/etc. The whole cerebral->built-digital-stuff timeline is shrinking like a teenage Moranis up in this mofo.
AI's ultimate use case, if you ask me (and you shouldn't) is to generate the brochure for aliens when they arrive. AI absolutely packages everything up perfectly, but it doesn't always understand the actual content. I don't mind this awkward state, but it wont be around long enough to really get annoying/endearing/reassuring. Right now, it misses the stuff I assume a robot would miss, but images and VOICE is hard to fathom. TTS, specifically Bark, is a true blackbox, and it was the first time I felt that "oh, ok wow" feeling from AI. I think my brain responds to audio with more emotion than visual, but we all have our 𝐡𝐨𝐥𝐲 𝐬𝐡𝐢𝐭 𝐢 𝐧𝐞𝐞𝐝 𝐚 𝐬𝐞𝐜 moments! 😅 It will be spooky when those gaps are suddently gone and surprises are less common, in general.
Random thoughts about the uncanny valley that sparked from that bizarre camera model image... For me, the whole concept, and seeing/hearing how we're summarized from the angle of the 'hive' / AI, feels extremely uncomfortable because I have to assume we may not always be relevant. The bell's rung! 🔔#no-unrings - humans usually 𝙟𝙪𝙨𝙩 𝙙𝙤 𝙞𝙩. after learning that it exists especially if 💰🚀!!! But yea, that scenario threatens something primal and people have been going wild arguing about AI , which is not my scene.
Great video, man. Spot on with the upscalers. That transition/wipe really shows it. I love Topaz Labs' models, personally! I worked there and built the CX setup in Austin. So much fun working there, they are all on another level - super humbling to be around. Anyways! I'll always throw their models out as suggestions for upscaling, but... they're not as Comfy as they could be. I mean, wearing a suit isn't always Comfy! Can't make cash wearing opensource sweatpants, AMIRITE? 😂
So many new ways to profit, which is cool, but there are also so many emotions wrapped up into it and I didn't expect the polarization of AI art. I'm an artist. Art will be fine. Being able to try all of this COOL SHIT right as it hits github/humanity/HF is exhiliarting because this is totally my scene, and I truly have no clue where these models will take us, but we kinda do because it's our idea in the first place! As the kids say, "yolo....(v8-face ↑)" Until I see AI doing the bad stuff, I'm in the honeymoon phase avoiding the slop.
you didn't translate what is pro in german… 😢
profit!
Any indication for the 1.1 [dev] version?
i think it stopped on version 1
General question. I keep not seeing textures or inherent detail on many of these models. When looking at "immersive" experiences, and the point of Unreal's Nanite was about inherent detail added to games systems. So when do we see that, or is there a model you'd recommend with that?
Okay, I'm stealing that inflection of "Bland Photos" after the iPhone format mention. So slick.
Your content are always top tier and the jokes are always on point 🎉 Keep up the amazing work 💪 Love your channel 👍
Nice! BFL video is either going to be great, or disappointing. They've set the expectations pretty high based on the quality of the image generation.
I know-- I've got my fingers (non-ai generated) crossed for them!
What "creative upscaler" are you using on those generated images?
For this one, I just used the built in one that Freepik has. But I use them all: Magnfic, Leonardo's, Krea's-- They all kind of do different stuff. They're all unicorns!
Thanks tim!
This is the first time I write 3 comments in a video… (annoying I know) sorry for that! but that pika comment at the end was so you! love it! Love your sarcasm! have a great night and thanks for another great video!
I'm actually drinking blueberry yogurt drink right now 0:27 :D
Haha-- I missed Breakfast for this video!
Ideagram is my favourite at the moment.
I love them all-- but yeah, that latest update is pretty great. I did a whole video on it here:
ua-cam.com/video/gBnVsOc6Tdc/v-deo.html
what creative upscaler do u use?
For this I just used the one built in for freepik. I bounce around, Magnific, Leonardo, Krea.
They all do slightly different things!
So no downloadable 1.1 dev version?
Nope. Only through API. Which, you can apply for-- but, it isn't really build for home usage. This is more for pretty high end machines.
@@TheoreticallyMedia I have a pretty high end machine 😃
Tim. I work a lot in MJ and use it for work professionally. Do you think flux is better than MJ? From the looks here, it doesn't seem to eclipse it but I haven't used it much... just a few rolls here and there.
I used both MJ and FLUX and I found FLUX to be better. MJ does not follow promts that good. When I ask it to make photo it often does not. MJ is more creative though. It all depends on what you need more. If you can use FLUX on your own PC then go with it as you can do inpainting, image2image, styles, workflows and so much more. I also train faces with it though it takes 6-7 hours.
I’d say MJ (to my taste) tends to be more eye pleasing out of the gate, whereas with Flux, you tend to need to work at it a bit with some post processes.
On the photographic side, I’m actually starting to lean toward Flux in terms of realism though. Although, you do (currently) need that extra creative upscale to get there!
Sweet! I have Fal_ai. Can't wait to play with Pro 1.1 - Green tea, don't fail me now.😁
Just make sure you have it in the big cup! haha
9:50 That's "Man in a blue business suit" by Kandinsky.
Arpad Elo looks like he could be David Lynch's older brother. Jeff Lynne would be proud!
I finally signed up to test Flux. Where I found it really lacks, is the exact same as Dall-E. Creating realistic food photos. I advertise for a lot of restaurants and Midjourney is exceptional at food pictures, especially Indian food. Flux just fell flat.
Oh, MJ kills it at food! I once gave it the ingredients to a dish as a prompt, and although I know it was just keywords as tokens, I was still amazed that it “made” the dish!
I’ve seen some good hamburgers with flux, but I can see Indian dishes kind of failing.
…man, now I want a Samosa!
@@TheoreticallyMedia That is really cool!
What CFG you use? I find flux is best around 2.5, especially for skin texture.
Tx love 😻
♥️!
As always, great info and highly entertaining. You could be talking about binomial equations and it would probably still be entertaining.
What the heck did you use to upscale at 6:17? im so lost
I just used the built in upscaler that Freepik has
We need that dev and schnell upgrade too
would be nice if you compared it side by side with 1.0 ...
Imagine flux 1.1 with minimax image to video 🤤
Wish they had corrected the finger/toes disfigured digits issues, its still has that problem and my 1st generation gave 6 fingers😮- creepy
did they remove that bloody cleft chin though?
Haha, I think so-- Yeah, I remember old MJ had that same thing!
I'm still wondering why every street shot has people standing in the middle of the street instead of the sidewalk.
very cool..
Flux is super exciting overall! Black Forest Labs is really on the cusp of something big!
That's a normal cup size in the pic :)
Haha, I'll tell you what, I'll order the extra-extra large pot of coffee and pour some for you!
What is the best way to imstall this - i have 16 core Ryzen / Rtx 3090 system 👀
Easiest? Likely Pinokio. I don't go over Flux in particular in this video, but I do go over Pinokio installation here: ua-cam.com/video/3pp7qw19nuA/v-deo.htmlsi=tmOuJhXY15jqhIqw
Tried Flux Pro 1.1 on my provider Freepik and am most disappointed regarding realism. Same prompt of a photo that looks photorealistic on both Flux Dev and Mystic ( Freepiks Checkpoin on Flux basis) turned out as a digital art portrait on Flux pro 1.1.
I just tried it, and I have to say, Midjourney is still better.
can't wait for their video model. Too bad there is absolutely no information of it other than their one preview video compilation
why are this elo scores starting from 950? Thats a manipulative method to make the difference seem larger.
There's an old Simpsons joke that always comes to mind-- it's a Homer line: "Oh, people can come up with statistics to prove anything, Kent. 14% of people know that".
That joke is what, 25 years old? Still holds up!
I love the prompt: "lets us know that she is leading them into a trap"
Hah- somehow it got lost in the edit, but when we got to the viking woman, I had the line: "Now THAT'S a look that says you're going into a trap!"
I blame the editor...who, ironically, is also me!
4:45
Yup, there's almost no fall-off...... a bit unrealistic
I looks really weird-- like, that era of late 90s green-screen in movies.
The is no "Just" when it comes to Dad Jokes. :)
You have just earned the title of "My favorite comment"
How is it that I've never realized that
Mr. Blue Sky = Mr Blues Guy...
How to use this ?
links below! Easiest is probably Freepik-- since it has 5 straight up free generations to check it out.
@@TheoreticallyMedia thank you !
Just follow the links in the video description.
For a more general approach there are a lot more videos on his channel that go into detail on how to set things up.
If I can't download it and run it myself, it's not interesting to me.. so can I ?
top tier dad jokes!
Haha- always gotta have some levity with all the insane AI stuff!
7:09 They both have six fingers. Where's the thumb?
Tim, what are you talking about, that coffee cup is way too big.
Sincerely, an Italian
:-) absolutely. - Although Flux does often have problems with the sizes of objects the coffee cup FLUX created is about exactly the typical size you would get in a coffee shop for a normal coffee in Europe.
Please make a update video about akool realtime deepfake video.
Flux makes the same female face every time, although this version looks improved in that area
That seemed to be their thing with the increased diversity. I mean, to be fair, MJ v4 had a LOT of that "same" face issue as well. Flux is only at 1.1-- so, give them time to spread their wings.
“Point zero zero four cents” is $0.00004 not $0.04! But I don’t know if you’re trolling us 😂
that car on the right side going on the wrong direction on the first blue suit pic
Drop a video about deep live cam
Try Glif with 300 styles.
Its asterisk, not asterik
haha, yeah, I get that all the time.
....and obelix
Five fingers, no thumb!
6:59 while both people do have 5 fingers, unfortunately they each have 0 thumbs. HAHA!
And I love the modern iPhone in a photo from the 1980s.
Sorry but FLUX is not an opensource model.
well, it is-- just not the Pro models. Black Forest Lab are made up of Stability employees, so-- I think they're trying to head off the mistakes that led to Stability's fall (that's a funny phrase to type)--
Not to say that Stability is done for-- I'm still rooting for a comeback.
Although FLUX is MUCH faster!
My problem with AI and I would hope I please interview an artist that knows about anatomy The problem with AI is it's not trained on anatomy it's just trained on the final result of an image not the steps to actually create anatomy this is the major issue with them and why you get these hands that are too small joints and all that because it's only trained on the final result
And if you could interview an artist that knows about anatomy and shape theory it would expose the flaw of AI it has no anatomy and shape theory coherence
Flux where MidJourney is Lazy & fearful of input & testy they are not 😂
06:12 I don't like how the creative upscalers constantly make characters older and, sorry to say, uglier (not in terms of detail)
Haha, it can go overboard from time to time. The best bet for that is to focus on the promoting with the upscaler and give it some weights like, beautiful::1, realistic skin, ((wrinkles))
Try that out. Although, you might have to play with it to not get a plastic character!
If you love censored Ai, Flux is your new go to
AI dressed up the black guy in European style clothes 😆
Yeah, I'm going to pass for now and wait a bit longer. Stylized super models are nice, but a bit overdone. Then there are the small immediately eye catching flaws. SD 1.5 forever!! lol
Totally get you! SD 1.5 really was a golden era! On the plus side: Flux is still only at 1.1, it has some room to grow! And containing Flux is ex-SD team members, I have NO doubts it’s going to grow to become an amazing model!
Flux needs negative prompts
1:54 it's pronounced Elo. lol
since ai belive nobody is writing you message in spanish! Que buenos son tus videos! Gracias nen! Lo mejor!
5:00 Looks like their nipple censoring is still causing nipple jewellery... sigh. My guess is their filtering added the little balloons to nipple bumps not just bare nipples.
Oh yay, more diversity..
"This guy would have been burned as a witch" 😅😅😅😅😅😅😅😅😅😅😅😅😅😅😅😅😅😅
you look like guy from free guy
Acknowledges it's pronounced Elo
Still calls it E.L.O.
xD
Haha, did I do that? Needed more coffee today apparently!!
OK. I spent $1 and made a nice, excellent image. And what now? What am I going to do with it? Save it in a cloud. And what then? Spend another $1 for another nice image? To get addicted to make nice images? It's called milking cows. Thank, but not me.
.
To each their own. As with any act of creativity, value is hard to quantify.
Tangentially related, I think that's a byproduct of digital media. My camera roll (on my phone) and cloud backups are FILLED with thousands of moments from my life.
But-- do I ever really go back and look at them?
It's funny, obviously I'm old-- so, the era of my childhood is really summed up in a handful of photo albums. My kids, on the other hand? I think every moment has been documented...
Really? Better? I thought they were "perfect"? I thought they had revolutionized the industry. I thought they had "turned the corner" after SD3. Now... upgrade? What? Shocking.
Everything’s upgrading, all the time, everyday.
I get clickbait words might be annoying and even angering, but it’s probably best to just let it go. When news comes out it will always be sensationalized, and always has
@@RealityRogue I'm a curmudgeon. There is nothing offensive or criminal that I let slide without comment. What happens after is up to you. These people are feeding you ground meat while calling it "steak". I think you deserve better.
@@dirtydevotee Hey fair enough. Some people tend to let their emotions control them on the internet and get angered without even a thought. But you obviously are aware, and that’s enough for me. There’s always a time and place for people with tough opinions I’d say
@@RealityRogue I appreciate the kind words.
Eh, I get it. Look, I think the content of the video is fairly balanced. I think it's a great model for a 1.1 with room to grow-- but, I also understand the frustration w/ thumbnails and titles. The thing is: That's UA-cam in a nutshell. Attention economy and all that-- you gotta play the game to get the message out.
Stop burning witches.
what?
"bokehish"?
you can add -ish to any word in my book!
Flux is probably one of the most useless pieces of crap ever
It says I have 10 daily credits and when I try to do a prompts then it says I don’t have enough credits
bye bye Midtrip.
I don't like Blake. He looks like he's into RICO.
DILF 😉
isn't this the time when you change your channel name to something more real like yourself or your nick name 😕
Is that a thing? I’m clearly bad at marketing! Hah
@@TheoreticallyMedia
your dad jokes and warm and friendly behavior are your marketing point fr
i know nothing about this channel and i really don't like the name theoretical media....you could put two random words and it would be better ..
cheeseClick
doubleCake
steveTake
@@jso19801980
yes , the name is very cold and unfriendly unlike his behavior and connection to the viewers which is very warm and cozy