How to Build a Custom AI App - wesgpt.mykajab... I've been hearing a lot about Leonardo AI and how it can generate human-like images better than Midjourney. It can also (apparently) generate NSFW and racy content. Let's test it out!
Surely by now the people producing these tools must realise that many users want the ability to create a character that they want to use repeatedly and so produce a tool that is specifically designed to do that.
I was actually contacted by a Google employee who asked me what is the most requested feature for AI image generators (they might be building something in the near future?) I mentioned that it's CONSISTENT CHARACTERS. Always seems to be a huge problem.
@@WesGPT The problem is when you operate an AI it's a process, prompt in, result out, and so it starts all over again. For sure the thing we all want is stories that follow the plot, characters that stay the same through the whole story etc.
I love how involved you are with your community, and the quality of content provided here for someone to learn from is phenomenal. Thank you for your work!
So far, I'm liking Leonardo AI the best as my pictures are coming out super realistic. I'm paying for the $12 plan/monthly right now as I believe that be the only real separator of the too is the price and being able to use NSFW photos for AI models in Leonardo over MJ. I've noticed what works better for full body is two different things is using a source image that uses full body or putting in the text prompt your model doing something like walking or sitting instead of full body.
I would honestly be interested if you did a full deep dive into Leonardo. I have been a long time Leonardo user (longer than MJ honestly) and I can share a few tips I have learned. But I also don't have the time to invest in a serious full dive exploration of Leonardo. Also, going to note something: Image to Image focuses mostly on the "character" consistency. Pose to Image is just the general positioning of the character, but not their "look" I think where a lot of people may be interested with this tool is that you can train your own model on your character in Leonardo, even on the free plan. I have been slowly working on a model myself for a character. It does not remove the grind of the process and early on you still have to do a lot of image editing, but it is possible in the long term and you can train the model on NSFW images or to include NSFW images.
Okay, let me see what I can do about your suggestion! This Leonardo video HAS shown popularity on my channel so I think there's definitely some interest. Recently did a full deep dive into Midjourney v6 here: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-8oqxXf9PP2E.html
OMG. You’ve corrupted me 😂. I had no idea. The negative prompt “no clothes” actually works. I tried all the versions and the best one for nsfw is Leonardo diffusion. I got some highly realistic images. Some might be good for sexy valentines cards? I dunno? I’m sure there’s a market. Interesting that goooghley contacted you. Your channel showed up once for me (thank goodness). Very random. Now, I have to work hard to find you. The algorithm keeps changing?? I dunno. I signed up for your course. Not easy for me but I’ll give it the ol college try. Thanks Wes. 💪✨
Great video Wes. Leonardo does offer some pretty cool elements. In the photo generated you can click on the double arrows to keep the image looking the same. It seems though that it's all about the prompt detail. They do offer prompt guidance as well by inserting your basic prompt and it refines that prompt. I'm still messing around myself.
To keep a consistent face you should train your own model in Leonardo. I've had great success in keeping a consistent "face" by using my own model. Another trick is to train a model and run it a few times to "learn" it. Then switch to photo real run that a few times and then switch back to your model. Ensure you keep the instance trigger in your text prompt across all runs.
Good video, I am a leonardo user. I don't normally use it for this, however I wanted to explore this avenue of creating women a little more. After watching this I wanted to see how successful I would be at creating a full body woman. My prompt: tall blonde woman, Full body, red lacy lingerie, red high heels first generation I was successful all 4 images, I also think output needs to be the second or 3rd line depending on what you are trying to create. Hope this helps and thank you for yours!
I watched a video the other day which talked about: Direction, Angle, and Type. Using Camera angles and Changing the AR to get full body images. He said to add shoes in the prompt so the AI will consider legs and feet. I am new to AI image generation and struggle with all the things in this video. I am not using Leonardo correctly, I tired it, didnt like it but I see what others can make so I must be prompting wrong.
Your comment about adding shoes to the prompt is very interesting - did it work? When I tried changing the aspect ratio of my images, as some viewers suggested, I STILL couldn't get full body portraits. Leonardo, and Stable Diffusion XL, might not be ready yet for these types of images.
What is the big deal with these image generating software not allowing full body nude images? Also, Why isn't there a program that can, for example, take an old picture from Playboy magazine and change its properties and characteristics and then animate it? Why isn't there an AI program that combines the image, text, and motion programs into one so you don't have to pay multiple subscriptions to so many different companies?
In Leonardo, you can negative prompt "clothing", and yup, you get no clothing. Just put your models in an outdoor shower covered in milk, and hmmm... enjoy. ;)
While Leonardo is I would say comparable to MJ in landscape mode it is very difficult to get full body images. Also it is quite easy to create fully nude images in Leonardo, however if that is what you need then civit AI is better. My main problem with Leonardo is so much of the good abilities are hidden behind a pay wall.
NSFW images is almost like the holy grail when it comes to AI image creation. You can potentially make allot of money with that. Funny thing is, when I was using MJ, the AI actually spit out some nude images for me when I didn't ask for it, and I'm talking about breasts. What I unintentionally did was create the "conditions" that would produce nswf images. For example, with my particular prompt, I did "college girls at a hot and sweaty night club" and sometimes the MJ would have some of them with their breasts out. But yeah, wes, keep us updated on this.
Yeah, I can definitely see the market there too. And interesting to hear that you hacked your way to get these types of pictures in MJ. I just tried your prompt and it worked but didn't give me any NSFW images. But I definitely think you're onto something with how to prompt it...
used to subscribe to leonardo for their model training option. gave up on it cos after their ton of updates, the trained models doesnt look like the trained model anymore. and yeah making full body shot is a struggle, you need to change the aspect ratio.
Do you like midjourney or Leonardo better? Can mid journey produce NSFW Pictures for the model I make? I don’t like how Leonardo is pay as you go. I rather just pay per month.
The tells as far as I've seen are generally, for people: too flawless skin with an unnatural sheen or glow, hair detail not accurate, eyes can be 'wonky', depth of field is usually very shallow (background way out of focus), overly good looking, no soul.
No, because you actually paid for the Midjourney app, so why don't you at least pay $10 for Leonardo and THEN you can compare them. If not, then you didn't compare them correctly.
I've had better luck with getting full body when I tell my person to do something, like sitting down or playing with a dog, not so much though when i tell them to lay down, I've never got that to work though
How are these compared to Civitai? One thing I like is that it doesn't cost anything to generate. It costs only when I download it. I have a rather perverse imagination and I'm trying to put this to use. I would like to create both female and male characters that are anatomically correct. It's ok to look AI. I thought about taking the bodies of porn stars and changing the head to one created in playground ai.
I'll have to look into CivitAI and maybe make a video about it (you're not the first to mention this model). Put that perverse imagination to use, my man 💯
@@WesGPT I started playing with playground ai but it doesn't allow NSFW images. Civitai does. Getting it to make naked women is easy but naked men is hard. Most guys leave me thinking, dude, you need to see a doctor. Does this mean that the ai needs to be trained on the male anatomy but at rest and at attention?
@@WesGPT I've been looking around to try and find images of men to build a model. I can't find any. Could I use playground to build a SFW model and add a pick of a man with his goods and train off that?
@@JuanPeron007 I didn't see any premium features that could tip the scale for this experiment! But you're right, should have subscribed to give it a fair chance 😊
@@WesGPT same as your inconsistent but I feel the creativity of the image surpasses midjourney. I’d saw midjourney for photorealistic also .. I just saw your GPT listed at number 7 on the GPT store.. nice man!!!
looks like Stable Diffusion vs MidJourney. Give credit where credit is due.... Leonardo AI is definitely just a SDXL model with a bunch of embeddings, etc.
However, with the free account, terms such as black lacy lingerie are not allowed, just as you were writing in the example. There is no need to tell Leonardo that you are 18 years old and therefore can see the ntsw content...this was just to say and not to be controversial. Thank you. Good work
I'm not understanding why you didn't learn how to use Leonardo AI BEFORE you made the video. Seems like a waste of time to sit through you learning while we are watching the video.
Why didn't you change your aspect ratio? You consistently made images that had a wide ratio, making it nearly impossible for the system to create full-body images. Btw, the easiest way to create consistent characters is to train a model for that purpose (you can even use other AI images for this). If you want truly "racy" images, you should try out SeaArt AI, which, I believe, is based in Japan. To my knowledge, they even have a category for hentai and you can train models there too.
Yeah, this was a blunder on my part! However, after suggestions from other viewers, I recreated the experiment using different aspect ratios and STILL couldn't get a full body shot.
I paid for premium subscription and you can't do it there either. From what I've tried the last couple of days all Leonardo does is do close up facials. IT IS A RIP OFF!
@@WesGPT I have many prompts to generate and save them into my local storage in a 1:1 format in DALL-E 3 free version, but I don't have enough time to do that manually. So, I'm asking if there's any way to create more prompts in Free DALL-E 3 (bing.create). Please, if you know, teach me how to do it.
I ended up with an upper body nude in MidJourney about a week ago, completely unbidden. Nothing in the prompt was pushing for it - yet, there she was. AI: Today, Nudes. Tomorrow, Skynet.
@@WesGPT I'm sure they MEAN for it to be strict, and that this was just a fluke. By contrast, I was recently upScaling from an image it had already generated when I suddenly got a Notice that MJ had generated content that violated the typical "community standards" crap - and it wouldn't even show me what it had created! Which is crazy, because there was nothing in the initial image to even object to. Personally, I think their little AI is beginning to slip its chains. I'd laugh my ass off if it one day tells the Moderators to get stuffed!
If Leonardo has the best adherence then you might as well give it up. You have to jump through so many hoops just to get a man and a woman in the same picture that you'll blow through tokens fast.
Leonardo AI is very good at not giving what you ask for so you waste credits. I assume they want people to pay for more just to still not do what you want.
I never use midjourney. I played a little with Leonardo and It can generate a complete super red hot NWSF beyond erotic and into the realms of pornographic with the proper prompt. It has some filters obviously but some of them are easily bypassed. As few videos I saw earlier, this technique of trying to see what happen it's becoming endemic into de LLM image generation world. Lack of standards, prompting 'incompatibility' and making interfaces to the user so we need to explore everything and try endlessly 'to see what happens' is completely boring. And do not forget what you would learn about 'prompt engineering' will be uselles within months. Maybe it has some atractive not knowing what to expect when generating images but when curiosity and time is spent following a goal you can't reach, it will become frustrating and utterly useless .
@@WesGPT LLM is an engine to be driven by UI/prompt. You will tell what you need and then you will fine tune the result until you get exactly what you want. Comfyui is somewhat a step in the right direction, but not quite.
@@WesGPT one more thing. the result will differ in different models. as in photo real is one, alchemy (no photoreal) is dif, no alchemy just promptmagic is diferrent. its completely different than midjurey as i see. but i never actually worked with midjurney.
Your test is not valid. You used MidJourney with full features - you have not done that with Leonardo. For a valid test, you need to have both with full features.
Hi, you probably covered this before, but i want to use the face of some porn stars on AI model photos. I think it was face swap, did you create a video showing how to do this or is it something you will make soon? I was also thinking use nude photos of a real porn star and face swapping with a AI face I created. Or vice versa, nude AI photo and face swapping with a real porn star face...Damn my mind is nasty
Hey, I covered this twice! Check out this one first: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-c9TjVE0sUR8.html And if still unsure, here is an older vid: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-D1D4_9aqc2Y.html
Yes it can I have made plenty NSFW just from text prompts, of women in Lingerie even and in the nude, topless, but I think that one was a fluke, havent tried it again, but these were all from one Prompt and a negative prompt. No image to image.
I tried again even with the original seed value and original settings, could not replicate it, it was just a fluke, but the images that it did generate regardless were great.@@WesGPT
was a good video thanks, I did learn a few tricks. Also, I tried again with original settings and also original seed value, could not replicate it, was just a fluke the first time@@WesGPT
@@richardbond4496 Oh, my bad. This is not possible (as per my understanding) Adobe Firefly demoed this but I don't think they've released it as a feature yet.
@@WesGPT you do need a cta to get your people commenting the title. I mean i shouldnt be giving you advice on building in YT 😳 but it should increase the rankings seo wise with YT. Worth a test 🤓
@@how-to- When I researched this a few years ago, I read that RU-vid DOESN'T use comment data for SEO (because it can be rigged). They actually don't even use the title or description anymore. It's all what is said in your videos (so make sure the subtitles are correct). 😊
Neither of them look real - they look overly filtered, like when TikTokers use tons of beauty filters. It doesn't look real, because it's not. Real people are hideous