New AI Image Tools: Is Flux Pro the Midjourney 6.1 Killer?

[@gavinpurcell]

I'm telling y'all -- spend some time with Flux Pro it's way too good and we're gonna get some remarkable stuff going forward.

[@bbreimm]

Gavin, brotha, I hate to call you out, but for the Tokyo eye piece, you were suppose to evaluate realism. So while MidJourney respected the prompt more, Flux actually did give you somewhat realistic reflections. Midjourney on the other hand just went for something impossible and very artsy to bend to your will. I'd give that one to Flux!

[@AIForHumansShow]

you know what you're absolutely right -- the Flux eye was a lot more realistic and I kinda screwed up asking for the weirder prompts. MY BAD. (still Flux Pro is pretty incredible)

[@LikelyKevin]

@@AIForHumansShow Gavin. YOU FAILED. This is why I'm sticking to AI co-hosts exclusively. #BooHissGavinMISS

[@edwardferry8247]

MidJourney is basically one guy with his peculiar obsessions determining what people can and can’t do in his sandboxed world. All MidJourney has is the super convenient front end rather than having the models and loras that Civit is full of. You can’t do truly consistent professional work on MidJourney you can create great images but it’s amateur in any real world sense that could replace higher end consistent image creations.

[@AIForHumansShow]

I really, really need to get comfortably in Comfy UI -- I was pretty good at the other SD ux but have never taken the dive. This might mean I go for it.

[@HitnRunTony]

This was a super fun test

[@AIForHumansShow]

hey thank you for this! these videos are harder to make but super fun to see that people like them. i think this week I'm gonna try one that goes through lipsync tools

[@docdanoderm]

Great video and the best comparison I’ve seen. We’ve integrated Flux with our apps and found all of your comments totally in line with ours

[@AIForHumansShow]

hey that's awesome -- thanks so much. def spending a ton of time with it. have another video coming out tomorrow where i used the Flux Realism LoRa with Hedra and my mind got kind of blown

[@archael18]

You should read the enhanced prompts after they're enhanced. You said the hummingbird's feathers weren't okay when the enhanced prompt exactly said "ruby-throated".

[@AIForHumansShow]

Honestly, you're absolutely right. I screwed up this AND also didn't notice that the Flux model didn't say "Cloud Not Found" -- this was kind of one of my first ones of these and I screwed up a few things. That said, hopefully I was still able to get across the visually different styles.

[@FluteTraverse]

9:32 The generated image of a piano has an incorrect arrangement of the pattern of black and white keys on the keyboard: 2 black keys per octave, then a gap, then 3 black keys. Many text-to-image generators display these objects incorrectly: * the correct arrangement of the keys of a flute or other musical instruments * the labeling of a computer keyboard * the face of a clock * a cane for the blind * a wheelchair or rollator from a specific manufacturer * devices that are not shown in the training data, e.g. a car polishing machine

[@AIForHumansShow]

yeah interesting -- it's almost like a weird dreamy version of what reality is but i wonder why they get these specific things wrong

[@FluteTraverse]

@@AIForHumansShow I forgot to thank you for your well-structured and informative video. (That was impolite). The text-to-image generators also have problems here: * the number of legs of an insect (6) or a spider (8) * a sheet of paper with printed music scores * how to hold and play a musical instrument correctly Feel free to test Flux with these challenges.

[@AIForHumansShow]

@@FluteTraverse hey thank you for the kind words -- really much appreciated

[@dwang718]

flux pro nailed the humming bird. look at a Ruby-throated hummingbird.

[@gavinpurcell]

you're right -- def have a few notes for this sort of video for next time. spending a LIL MORE TIME LOOKING AT THE IMAGES.

[@mymotherismoremetal]

Would have loved to see a prompt of a cartoon character ketchup bottle at the apex of overgrown terra scape that meets a clearing. Reflecting on the bottles eyes is Hot Dog City with a heavenly glow. His facial expression in awe…

[@AIForHumansShow]

gonna see what we can do with this

[@mymotherismoremetal]

Another character relish bottle emerging from the sewers in Teenage Mutant Ninja Turtle attire. With a vengeful grin and the reflection of Hot Dog City in moonlight in his eye and perched atop a storm water pipe outlet

[@mymotherismoremetal]

A mustard bottle sky diving. View is exited from plane made from shopping cart. In his goggles is Hot Dog City skyscrapers on the edge of a sabret onion ravine. His face stretches from falling G force with saliva exiting the corner of his mouth.

[@gavinpurcell]

@@mymotherismoremetal give me some public social and I'll post this for you

[@krakenunbound]

I like that Flux is not censored if you run it locally. Get a nice video card and go to town. Blowing me away right now. However, what I enjoy doing is looking at some of the Midjourney 6.1 stuff that people post and trying to replicate it. Sometimes I get close, sometimes not. IF Midjourney was free and uncensored, I would still use it over flux.

[@AIForHumansShow]

how big a GFX card do you have? not sure my 4080 16g is gonna cut it for local usage and totally agree on the MJ thing, once they build their front end made a huge difference on discoverability

[@krakenunbound]

@@AIForHumansShow I have a 3090 with 24gb of memory. I wish I had the 4090 instead but it is what I can afford right now. You need the most gpu memory possible. I use comfyui

[@krakenunbound]

@@AIForHumansShow Your 4080 16g will work *fine* just not as fast as the online website for flux. But for normal SDXL stuff you are more than fine. If for no other reason than making more youtube content, you should set aside a few hours to go down the ComfyUI rabbit hole and try some things out.

[@gavinpurcell]

@@krakenunbound yeah I think I might try to do it and see if maybe I can make a real BEGINNER video on ComfyUI

[@ragemax8852]

I'm using FLUX Pro and this is killing Midjourney and SD. FLUX is so fun to use, at first when I was using, I was having problems, but now I'm really loving it. Stuff is coming out so realistic from it and it finally fixed the hands issue as well. LOL!

[@AIForHumansShow]

it's SUPER fun and kind of got me excited about AI image gen again -- this plus Gen-3 is like a super power

[@swordartdesign]

Finally I can leave Midjourney and save 30$/mo... and btw I can use political things, what a good time! Thank Black Forest Labs

[@AIForHumansShow]

Yeah honestly I've been thinking the same but I imagine that MJ has video prob waiting to drop sometime soon as well and I really, really hope it's good

[@gavinpurcell]

after a few more days, I'm VERY close to cancelling my MJ sub

[@PromptBattleRoyale]

Why didn't you use creative vs subtle upscale in midjourney?

[@AIForHumansShow]

well, i didn't want to have to do additional steps -- it was kind of a chance to see what it would look like first shot, obviously theres a ton of different ways to tweak the initial outputs etc

[@JMFromthaStreetz]

I know you guys needed a little vacation, but there is no excuse to have not had AI versions of you both doing a podcast during your absence. Can't wait for next episode!

[@AIForHumansShow]

hahaha well unfortunately, that would be more work than a regular episode probably but you know what we SHOULD try it -- might be a fun experiment

[@gavinpurcell]

tysm we got ourselves a lil break

[@jacobshank9758]

@@AIForHumansShow That would be cool!

[@Brrainz]

Should test with “chessboard” prompts

[@AIForHumansShow]

oooh good idea will try it

[@Brrainz]

In the quick model, I got one good closeup image (with some quite big pieces) and 8x8 but when I forced the perspective from atop it produced a 6x7 board. It’s still hard for AI to do a chessboard correct and with equally sized 8x8 fields and proper pawns.

[@hoangcan91]

great video thank you!

[@AIForHumansShow]

glad you liked it! and come back for other stuff 😀

[@iminumst7827]

As a stable diffusion user who really needs specific and trainable models, Flux is definitely far more exciting to me. The fact that a model even on par with Midjourney is going to be open-source and uncensored is mindblowing, but I think specifically with text generation, anatomy, and realistic styles Flux currently beats out Midjourney.

[@AIForHumansShow]

Yeah, I've been consistently impressed every single time I use it vs MJ. I think I might actually have to just switch to using it full time.

[@JohnsCarnivorousPlants]

I missed you guys last week!

[@AIForHumansShow]

Thank you! We're climbing back out of it all.

[@gavinpurcell]

thanks John -- we gave ourselves a lil break

[@BrianMosleyUK]

Great heads-up, thanks!

[@AIForHumansShow]

you're welcome!

[@creepybeat]

thank you so much for the comparison.

[@AIForHumansShow]

hey it was really, really fun thank you for watching

[@senju2024]

Hands are better BUT still not perfect. I hope both hands and text will be at a perfect state within another year's time. I am more interested in text in a picture but not only English text. We need proper Japanese Kanji writing Neon signs

[@AIForHumansShow]

OH that's interesting -- I hadn't thought about that but you're absolutely right in testing it

[@gavinpurcell]

part of me wonders if the hands will *ever* be perfect

[@Ale-sh2fs]

Hi, first of all congratulations for the video. I just signed up to the channel. I wanted to know if in the Flux Pro version it was: 1) possible to enlarge the images and make some changes. 2) if there is a gallery where the images are collected, to be able to see what has been created. 3) if it is possible to make the images created private and not leave them public. thanks in advance, keep it up the good work!

[@AIForHumansShow]

thanks so much for the sub! overall, with tools like Flux you have to kind of collect them yourself in a file, either locally or specifically find a host that will allow you to directly access the export file. in regards to changes, as of right now, not entirely but there's more to come since it's open source etc etc

[@FriscoFatseas]

Hope everyone who reads this has an amazing week 😊 hoping this month is full of more cool surprises

[@AIForHumansShow]

tytytyty

[@gavinpurcell]

We love this sort of comment -- we too hope YOU have an amazing week.

[@tgo007]

Are either one of them good at using an image of as a baseline? Like could you give them pictures of yourself and tell it to make you look like a bad ass warrior? I've played with old midjourney and it didn't really work. Gave me a Bad ass warrior but didn't really resemble the baseline much.

[@AIForHumansShow]

So honestly yes you can do this with Flux but it's semi-complicated. The best way to do this is prob using FaceSwap or one of those types of tools. Like make the best image of a barbarian you can and then use FaceSwap or other AI tools to swap your face on there.

[@Bullrunnerrr]

its close to MJ, but not quite there yet. It wont be long, and at least its cheap to use on X/Grok

[@kamruzzamanuzzal3764]

for now the only section where flux is losing to midjourney v6.1 is aesthetic style. but don't forget, it's the base model, so when ppl fine tune this model on like laion aesthetics dataset or other such datasets, then midjourney will lose in all aspects to this free open source model

[@AIForHumansShow]

i absolutely cannot wait for Civitai to start filling up with Loras and all the like for this model. When SD 1.5 came out, I went very deep on it and good lord it was so much fun. Also, I've been putting off really digging into Comfy but this is a good reason to do it.

[@kamruzzamanuzzal3764]

@@AIForHumansShow am also waiting for fine tuned models, I tested it and this model is not that great with anime style for now as base model. Need lora or fine tuned anime models

[@frumSrelliK_TV]

Quick question, since alot of these sites seem shady to me, I like paying with a Visa gift card. Do these site accept them? 'cause not all sites do.

[@AIForHumansShow]

always a good question -- replicate is pretty well known and i put a credit card in there

[@Danceverse786]

How can i use can you share a link.

[@AIForHumansShow]

YT doesn't like links but if you google "flux pro" and "replicate" you'll get to the replicate page I was showing off in the video

[@dr.cherhanlau7251]

Midjourney still can't get the number of subjects right. Never.

[@AIForHumansShow]

it's why i think the prompt coherence part of flux is such a big deal -- once some *really* makes a front end client for this that can do all the stuff I want without having to deal with the Comfy silly string, I will prob switch

[@dr.cherhanlau7251]

@@AIForHumansShow I am with you sir

[@ShanieMyrsTear]

Run a couple Flux images of a woman laying on the floor, or on her belly with her feet in the air, and enjoy the strange entanglements that happen.

[@AIForHumansShow]

👀

[@alangabilan]

Can I run flux on sagemaker with jupyter notebook!? I tried it but got some code error. I copied the google colab notebook. any help?

[@AIForHumansShow]

You should be able to but honestly this is why I generally run it locally installing with Pinokio OR just one of the larger hosting sites that pre-install. Saves so much hassle.

[@geolitz2665]

Why can’t both exist? Why do you have to have a high school football player mentality about AI programming?

[@AIForHumansShow]

they both CAN exist -- and good lord i'm at least a college football player mentality seriously, as excited as I am about Flux (especially the open source aspect) I really do love MJ

[@geolitz2665]

@@AIForHumansShow Then stop talking like an algorithm. Stop allowing AI systems to dictate your moral framework and making decisions for you. You should be the one deciding how to use your language to describe reality. Be human, not a slave.

[@AIForHumansShow]

@@geolitz2665 SIR I AM AN ALGORITHM

[@geolitz2665]

@@AIForHumansShow I can tell.

[@AIForHumansShow]

@@geolitz2665 hahaha well you're absolutely wrong but happy to keep chatting about it. this was (unfortunately) not scripted and entirely my brain's thoughts. so... I guess that's all there is to it

[@Trancer006]

How can we use Flux for free?

[@AIForHumansShow]

Check out the link in the video description - Replicate will let you get some free generations

[@Trancer006]

@@AIForHumansShow If i use it within Comfyui, will it be free?

[@Trancer006]

@@AIForHumansShow Is it free if i use the desktop software or connect it thru Python?

[@fittersitter]

Referring to your video title: "Is Flux Pro the Midjourney 6.1 Killer?": Killing Midjourney? Why would some idiot want to do this? It's a complete different tool. I'm not watching videos with stupid titles anymore.

[@AIForHumansShow]

well, they are both cutting edge AI image generation so I think it's def feasible to compare them no? and truthfully, i've personally been using MJ for 16 months or so now and Flux feels like the first time since SD 1.5 I'd have a real possibility of moving to something else. I DO pay MJ monthly for this right so I guess it's *kind of* accurate in the title? I mean not everyone is going to use all the tools right?