flux Laura is pushing the boundaries of lifelike AI Photography in this video I researched and tested everything you need to know about flux Laura so you don't have to the video consists of four main sections each with a deep dive I added time stamps for your convenience one of the first places you can use flux Laura is replicate replicate has this model called flux their realism and you can find the link in the description below this model allows us to run flux Laura and we can create extremely realistic looking images in order to do that I'm adding my prompt Here Am matur photography of group of women friends casual noise high contrast viid colors slide motion blur jpeg artifacts on flicker in 2007 2005 block 2007 block I saw this prompt on Reddit and I think it's a really cool one because it really creates realistic looking everyday casual super amateur looking and super realistic looking images I'm setting aspect ratio to 16 to9 keeping the other values as default and I hit Run for the results are here and it looks incredibly realistic there is a slight problem with the te here but besides that the image looks like it's taking with a real camera updated my prompt little bit and edit amateur iPhone photography I kept the rest same except I got rid of the weid colors I'm trying to push even harder for increased realism and I hit run so I'm really like amazed and surprised by the amount of realism here it just looks fantastic the amount of realism this model gives is simply mindblowing I want to try another one say iPhone self Snapchat photo 2018 sa casual outfit let's do this 9 to 16 hit run so in the output there is a a little bit of a slight new but it's nothing drastic or nothing crazy the next place you can use flux Laura model is called free pick Picasso so it's a model called Picasso under free pick unfortunately this option seems paid when you want to switch the flux realism it asks you to get a plan so it's not possible to use it without paying for it I'm still sharing just so you know uh of course you don't have to uh pay for this because there are other options where you can use flux realism for free so plans look like this you can of course use flux fast which is not that great next option to try flux Laura is file. a it's also a paid service but since I was training my own model here I had already some credits I will just demonstrate you since it's a paid option it may be not the most popular option out there for you it's pretty much the same model it's just a different UI let's try a challenging prompt for generative AI models photo of a hand holding a teapot and filling tea to a pink cup I will set guidance into 2. 5 so the result is here everything looks great it's a proper teapot a little small but does the job I didn't actually specify the teapot should be pink but it made everything pink it's fine the grap seems okayish I think it's it's totally fine so it does the job next option to try flux Laura is hugging face I'll try this prompt photo of a man playing a guitar this prompt sometimes gets difficult for models because you need to do the position of man and hand R the guitar the the chords they need to look good so I want to try this one to challenge the model a little bit so I just entered the hugging phase I already had an account but if you don't have account you can actually create one and I I came to the model page I wrote my prompt here and I hit compute and I didn't get charged I didn't ask money and model was direct directly running and it looks fantastic look at the details here look at the chords look at the guitar the the hands the position of man everything looks fantastic it looks it looks amazing so if I'm not mistaken you can use hugging face option for free uh there is probably a limitation and I will add details of the limitation to the video when it comes to pricing but it's uh it's pretty much free to to use it here the other option is called sweet AI see when you visit the sweet AI homepage there are quite lot of models you can choose models and from right corner on the top you can filter model types for example I filtered Laura and flux 1D and I had this model ex flux realism Laura here I click on xlabs flux realism Laura model and when I sign up to the Sweet AI they gave me 100 bus when I click on Create and it shows me the promp box to use the model I see the generating one image actually costs 163 bus like a buzz is like the credit system therefore to generate with this model on sweet AI it seems that it requires topping up my bus credits which therefore is paid so I was able to use flux Laura on replicate and on hugging face for free and full disclaimer for replicate I use the coupon code mentioned in met Wolf's this video here and in the description you will find the $10 credit link so I used that and those credits were added to my replicate account so thanks to this I was able to train my own models and generate with flux Laura model there and I didn't pay for anything so I would probably recommend you to check that video out and thanks again to Matt wolf for providing this credit next we will talk about how to train your own custom flux Laura models to generate AI images of yourself your pet or create Ai influencers and we will train our flux Laura on replicate in order to do this we are visiting ostr flux Dev Laura trainer and don't worry I will provide the link in the description down below we are to the first input field and choosing create a new model then it will ask us to give a new name to our model I will call this cyber jungle flux Laura in the next step we need to provide a zip file containing the images that we want model to be trained on they are recommending at least 10 images for training your model they need to be jpeg or PNG format so if you want to use the images from your for example iPhone you probably need to convert them here I prepared 17 images of me and for some of my subscribers probably this is the first time that you are seeing my face the naming of the images are really important and you need to be careful with this as you can see I did the naming in a way that it includes the Cyber jungle in it this is important because we will use this keyword simply you can write a photo of any trigger word that you want to use and then you can add a number after it and these photos needs to be a zip file that's why now I will select all of my photos right click and choose compress this will create a zip file I uploaded zip file which consists of all of my photos next thing is choosing your trigger word I will use cyber jungle as a trigger word here I keep autoc caption default on I will not change anything here and here and I will keep the steps in 1,000 in the default value same for learning rate and bet size that I will keep it as a default here you can use your hugging face repository ID since I'm leaving this empty for now I also don't need the hugging face token and I'm leaving this part also empty I click on create training and here as you can see training is running at the moment so our training is completed it took 23 minutes you can see the status here as succeeded and now we can run our model to generate images with my own face normally training your model on replicate is paid because you are basically ranting the gpus from them but I use the $10 coupon code mentioned in met Wolf's video let's go ahead and run our trained model you can see the name of my model cyber jungle flux Laura is trained and I'm in my playground first thing first I will write my prompt I wrote this prompt portrait photo of cyber jungle and his m friend in the jungle it's very important that I use the keyword cyber Jungle which is exactly the same trigger word when I used while training my model I'm choosing aspect ratio keeping the number of outputs default lower scale number of inference steps model and guidance scale and everything in default and I hit run so in the first try I wasn't able to get the results that I wanted I'm pictured as a miky here I can't say it's a it's a wrong depiction I think I'm less hairy than this monkey here let's try something little bit more simple portrait photo of cyber jungle and hit run this time I got the female version of me which is quite interesting it's not exactly what I wanted but I feel like I'm coming closer and closer at least this time I'm not a monkey and I actually look quite pretty so let's try another prompt I have a feeling that current guidance level of 3.
5 is already way too much creativity for the model I will try to increase it to six and I will keep the prompt portrait photo of cyber jungle let's run again so we are coming closer and closer to my real photo for the reference I normally look like this so we are not that far yet not quite there but I have a feeling that maybe this guidance level actually worked much better better I got better results after updating my prompt little bit as well as guidance it looks much closer to how I look I specified the gender because model was mixing up and bringing lot of results the results started to get better and better and I came to the conclusion that the problem was probably about my uh training images I had so many close-up shots but maybe not enough full body shots or medium shots I don't exactly know if that was the reason but because of that in certain cases I got like when I mention cyber jungle I got like animal results another option to train your own flux Laura is called f. a here when you come to model Galler on top you can see the option to train a flux Laura I will also add this link to the description down below we click on train a flux Laura and then we are coming into this training page it's a very similar workflow to replicate training your model on f. a is paid it costs approximately around $5 but I will still demonstrate to you in case you want to prefer to use this one over the replicate version this time I will train the model with the images of Loney my doc so you can see I added couple of her photos and I will again zip them so I can actually upload to the fault.
a for training I'm going to press on compress and I'm creating a zip file so back to f. a I will choose my images I'm just selecting the zip file that I created with photos of Loney couple of additional settings I will set up a trigger word again L I will keep the steps at thousands and rest of the settings I will keep as a default besides that there is nothing I need to set it up I simply click on start so after we hit start you can see training is in progress so my training is completed it took 27 minutes and you can see I'm charged by $5 I will click on run inference and this will bring me to the model that I trained I'm coming to the prompt box and let's start with something simple to verify that model training is successful portrait photo of learning I will hit run so images generated and resemblance is impressive as you can see this is one of the images I added to the training and it really looks like lony that resemblance is really high let's try something a little bit more difficult photo of lony as a fighter pilot flying an airplane in World War II here's the result it's quite impressive here's another one photo of lony as a boss sitting in the table with a bunch of money quite impressive results here's another photo from the training set and resemblances is just amazing one thing I want to add to the generation process about guidance setting if if you want to max out the photo realism and skin realism I would highly recommend you to set the guidance number into two because when it comes to higher numbers which is by the way maximum value is 10 it listens to the prompt much more closely but you will start to see more plastic looking or 3D or illustrative looking skins or details and it gets and far away from the photo realism set the guidance setting into two next thing I want to show you how to use image to image prompting using flux by the way I managed to find a flux Laura image to image prompting possible and this is on fault. a so when you go to f.
a there is this model called flux one de with luras and when you visit this model it's possible to change from this drop down here to image to image so when you choose image to image and click on more you will see it's possible to add an image URL or choose an image from your computer and just upload it I will not show details of this since this is a paid model but if you want to experiment with it on f.
