I've been using this in the last couple of days and I'm starting to think this can be quite useful to fix hands. Sometimes with some images the amount of work you have to do in Canvas in order to fix a hand is absolutely insane, but if you just take the image into this feature and generate a couple of new ones using it as base, you will usually end up with at least one image that is almost identical to the original and has the hands fixed. Loving your content btw, I'm still watching the video and paused to make this comment. Will keep watching now :)
So as always when you drop these vids on Leonardo AI I like to follow along inside Leonardo AI, and of course I come to see they've dropped a few more gifts. I really get mind-blowned everytime with this site. They just need to get the consistent charaters down, and I think they could own the crown of top AI image generator.
Thank you, and agreed. I think they're working on that in the background. They actually came out with another update yesterday called Lightening. So many new features being constantly added, love it.
@@metricsmule Yeah that's the one I was talking about. Lightning, plus a few more element features and Anime XL. They have to know with Midjourney getting ahead in the consistent image department they'd have to capitalize on it on their end.
when i do it using a premade leonardo ia woman portrait it works, it mainteins the face. but if i upload external source image as a reference i dont get the same face results, why?
Leonardo AI does mention that it currently works a little better with AI images, however, there has been drastic improvements since Character Reference came out. Have you tried using that yet?
It can do a lot, but it's kind of expensive. Using your specific settings, it's costing 17 tokens per generation. On the $30 dollar plan, that's 1470 generations or 49 generations per day. Which makes me wonder about the options and speed when using "relaxed image generations" after fast generations run out.
Out of all of the generators the most frustrating is to see how well they can do consistent faces but there seems to be no easy way to create a full body and continue to do so even using same model, seed, settings. It gets done somehow by creators & I’m aware of using images, adding additional features as in more models to create the pose in settings but nothing seems to be consistent. Unless I’m missing something. Tensor art is pretty good too but full body consistency seems to be a lot to ask from any generator. Your samples & instruction is among the best I’ve seen & I’ve seen multiple. The only other thing I think we’re all looking for is image to video that is more than seconds in duration. I would just like to be able to make full body characters consistent even with some editing. I’m sure Sora can do it but who knows when that will be released.
Very well said. I'm sure you're aware, but I've gotten a little improvement when using two images for image guidance; one for the character, the other for the pose (post to image). But I have also been experiencing exactly what you're referring to. Have you tried using the --cref command in Midjourney yet?
@@metricsmule not yet. I started w fooocus & then discovered a lot of generators were offering additional tools but simpler to use like adding in Lora & controlnet for the pose and more features w higher efficiency. I did see where there’s a site you can purchase full body model packages and I guess you’d do a face swap. From what I saw the models have full range of motions & muscle flex etc. I work in IT so it’s not about understanding software, it’s I want to get it down, use consistent characters to create & I work all week. I did get one face that is exactly what I wanted but I need to look into consistent hair also. Female characters get away with it easier since real females can use hair extensions, etc. I just watched your vid on rendernet. Sea art has a train model section but 1st attempts were terrible. I saw you had full bodies on rendernet so I guess I’ll look into that also. As for midjourney I’ll look through your vids & check into your suggestion. Thank you.
I've been trying a few things since many others have similar comments. I think I found a solution (almost) in Leonardo AI. But using these settings (see link) in image guidance has produced very good results. But definitely adjust the strength to a lower number instead of using my defaults in this link. I just used the default strength (1.00) and it did it 50% of the time accurately. When I adjusted the strengths between both the image to image photo and the pose to image photo, the results varied much more and I got exactly what I needed. See screenshot: www.canva.com/design/DAGA8OEVc0Y/MTusy_aVBPyYIuXe5MYxFg/view?DAGA8OEVc0Y&
@@metricsmule ok. I’ll try it. I did try 1st attempt w Leonardo. The image produced was different from original but it came out better than anything else. I did select HDR & 55. Wow. Stunningly realistic. However, unlike yours, when I attempted to recreate the same face/ body using that produced model/ seed it wasn’t consistent, different looking face, etc. I’m like, really? Do I have to do face swap? Then I didn’t find face swap on Leonardo. I have used face swap on tensor art & it worked perfectly, fyi. I’ll try the link when I get a chance. Thank you.
I am really struggling to generate multiple scenes with two characters. For example, I have an old man and a young traveler as characters separated. I want to create scenes like walking on a path & sitting on a bench. The problem I struggle with is to keep the same characters. You would really help me and many other people by explaining how to do this
my model face is not the same bro, and im looking to change a lot of things but keeping my model face, can you tell me why i do the same of you with differents results
And yet , like me, when you attempt a full body shot as with your yoga pants, it tends to cut off the head. It's frustrating. I have been working on it for a bit and still it's frustrating
Agreed. Sometimes I'll get the perfect image, but other times it cuts either the head or the legs off, lol. But there's actually a solution and I'll have to make another video to help others who have this same problem. This is what you do: Upload two images for your image guidance. First, upload the main character that you want other images to look similar to. Then in the other image guidance upload box, add an image that you want Leonardo AI to use to create the same post. On that one, make sure you choose one of these two options: depth to image or post to image. Using this method usually works. However, the one caveat is that you can't use two reference images with XL models or PhotoReal. So I've just been using Absolute Reality v.16 for my model. Try that and let me know if that makes it any better.
See another comment I replied to with a similar question. Try the settings in this screnshot (but make your strength lower) Screenshot: www.canva.com/design/DAGA8OEVc0Y/MTusy_aVBPyYIuXe5MYxFg/view?DAGA8OEVc0Y&
Great feedback. It seems a lot of people in the comments are experiencing the same issue with creating full body images. Using two images for image guidance has been working pretty well for me but it's still not perfect.
I'm trying to do crowd shots, it's like pushing a wet noodle. I find I can get any kind of crowd I want, so long as it's a bunch of vaguely Asian-looking guys wearing work jumpsuits and heavy coats standing in a dimly lit room or landscape, glaring at the viewer. This includes prompts that have terms like "drunken partygoers" and "smiling" and "pointing." Image to image solves that problem, but then you get into copyright issues.
I'm not sure if you have my Leonardo AI prompts database, but if you do, I added a new section where you can provide details and images for any prompt/help you need and I can usually create the type of prompt you're looking for. I wish images could be shared here because it'd be so much easier. metricsmule.com/leonardo-ai-prompts-database/
Thanks for this Pal. But damn, I've never experienced such profound frustration and annoyance, where I can't even generate the simplest of images. I'm a former professional writer, editor, English teacher, and at times have even been called a wordsmith by my ex-clients. Yet this darn Leo AI can't even come up with basic silhouettes of two horses next to each other in plain black, on a white background after countless prompts and formulas?! If I can't even get an incredibly simple and basic image, how the heck can I get something with more detail as I've seen in its models? I'm stymied and mystified with my constant road blocks and such contradictory and bizarre results. Moreover, my current business model depends on it. Grrr....
I understand & share your frustration. My project is creating a red, black, grey and white silhouette logo of two humans, not horses. But I can never get Leonardo AI to output anything close to usable, after countless hours and days, despite using all combinations of chatGPT prompts and Leonardo's prompt generator or improved prompts and negative prompts. He suggested trying image to image and style or content guidance or reference. I have better success using the free Bing AI image generator.
@@PositivityPlusPlus Thanks for your reply and understanding. yes you're right...image to image, content guidance and reference generates MUCH better results indeed. Best of luck to you and thanks again for you reply.
"Same exact face, same exact person."...uh, no, not exactly. She looks slightly different in each image. Kinda like telling twins apart, they look alike, but there are subtle differences.
Thank you for making all these videos. I’ve learned a lot from it. Bought your triple bundle to support you. (And to play with it of course!)
Thanks very much!
I've been using this in the last couple of days and I'm starting to think this can be quite useful to fix hands. Sometimes with some images the amount of work you have to do in Canvas in order to fix a hand is absolutely insane, but if you just take the image into this feature and generate a couple of new ones using it as base, you will usually end up with at least one image that is almost identical to the original and has the hands fixed.
Loving your content btw, I'm still watching the video and paused to make this comment. Will keep watching now :)
Thanks so much, and what a great piece of advice!
You... Are... A.... Legend !
🙏
Learned a lot. Perfect for beginners.Thank you
Glad it was helpful!
So as always when you drop these vids on Leonardo AI I like to follow along inside Leonardo AI, and of course I come to see they've dropped a few more gifts. I really get mind-blowned everytime with this site. They just need to get the consistent charaters down, and I think they could own the crown of top AI image generator.
Thank you, and agreed. I think they're working on that in the background. They actually came out with another update yesterday called Lightening. So many new features being constantly added, love it.
@@metricsmule Yeah that's the one I was talking about. Lightning, plus a few more element features and Anime XL. They have to know with Midjourney getting ahead in the consistent image department they'd have to capitalize on it on their end.
Exactly. I do know that Leonardo AI is also working on bringing generating accurate text/words a future update. That will be huge.
@@metricsmule That'll be huge, especially for merch design.
This is amazing!
Thanks!
Great JOB thanks for sharing :)
Thanks for watching!
hi thanks for the great info. the Triple Bundle price on your website is monthly subscription or one-time?
Hello. Just one time only.
Great info in there, thanks!
Glad it was helpful!
when i do it using a premade leonardo ia woman portrait it works, it mainteins the face. but if i upload external source image as a reference i dont get the same face results, why?
Leonardo AI does mention that it currently works a little better with AI images, however, there has been drastic improvements since Character Reference came out. Have you tried using that yet?
thanxs man, i will
It can do a lot, but it's kind of expensive. Using your specific settings, it's costing 17 tokens per generation. On the $30 dollar plan, that's 1470 generations or 49 generations per day. Which makes me wonder about the options and speed when using "relaxed image generations" after fast generations run out.
Do how do you get it to generate a full body pic?
Good question. Check some of the other comments/replies because a few people have had the same question and some helpful ideas.
Thank you but my images are not turning out well without the Alchemy (paid pro feature on)
Out of all of the generators the most frustrating is to see how well they can do consistent faces but there seems to be no easy way to create a full body and continue to do so even using same model, seed, settings. It gets done somehow by creators & I’m aware of using images, adding additional features as in more models to create the pose in settings but nothing seems to be consistent. Unless I’m missing something. Tensor art is pretty good too but full body consistency seems to be a lot to ask from any generator. Your samples & instruction is among the best I’ve seen & I’ve seen multiple. The only other thing I think we’re all looking for is image to video that is more than seconds in duration. I would just like to be able to make full body characters consistent even with some editing. I’m sure Sora can do it but who knows when that will be released.
Very well said. I'm sure you're aware, but I've gotten a little improvement when using two images for image guidance; one for the character, the other for the pose (post to image). But I have also been experiencing exactly what you're referring to. Have you tried using the --cref command in Midjourney yet?
@@metricsmule not yet. I started w fooocus & then discovered a lot of generators were offering additional tools but simpler to use like adding in Lora & controlnet for the pose and more features w higher efficiency. I did see where there’s a site you can purchase full body model packages and I guess you’d do a face swap. From what I saw the models have full range of motions & muscle flex etc. I work in IT so it’s not about understanding software, it’s I want to get it down, use consistent characters to create & I work all week. I did get one face that is exactly what I wanted but I need to look into consistent hair also. Female characters get away with it easier since real females can use hair extensions, etc. I just watched your vid on rendernet. Sea art has a train model section but 1st attempts were terrible. I saw you had full bodies on rendernet so I guess I’ll look into that also. As for midjourney I’ll look through your vids & check into your suggestion. Thank you.
Oh ya. Automatic1111 & comfyUI. It like never ends. Another reason I can’t seem to get anything done. lol.
I've been trying a few things since many others have similar comments. I think I found a solution (almost) in Leonardo AI. But using these settings (see link) in image guidance has produced very good results. But definitely adjust the strength to a lower number instead of using my defaults in this link. I just used the default strength (1.00) and it did it 50% of the time accurately. When I adjusted the strengths between both the image to image photo and the pose to image photo, the results varied much more and I got exactly what I needed. See screenshot: www.canva.com/design/DAGA8OEVc0Y/MTusy_aVBPyYIuXe5MYxFg/view?DAGA8OEVc0Y&
@@metricsmule ok. I’ll try it. I did try 1st attempt w Leonardo. The image produced was different from original but it came out better than anything else. I did select HDR & 55. Wow. Stunningly realistic. However, unlike yours, when I attempted to recreate the same face/ body using that produced model/ seed it wasn’t consistent, different looking face, etc. I’m like, really? Do I have to do face swap? Then I didn’t find face swap on Leonardo. I have used face swap on tensor art & it worked perfectly, fyi.
I’ll try the link when I get a chance. Thank you.
I am really struggling to generate multiple scenes with two characters.
For example, I have an old man and a young traveler as characters separated. I want to create scenes like walking on a path & sitting on a bench. The problem I struggle with is to keep the same characters.
You would really help me and many other people by explaining how to do this
Gotta go make me a sparklecore toon ass at the beach now. LOL. Bro skipped riiiiight over that picture.
😂 LOL you caught that. I didn’t want to risk it.
my model face is not the same bro, and im looking to change a lot of things but keeping my model face, can you tell me why i do the same of you with differents results
Try Character referenced and toggle the strength to the highest setting.
And yet , like me, when you attempt a full body shot as with your yoga pants, it tends to cut off the head. It's frustrating. I have been working on it for a bit and still it's frustrating
Agreed. Sometimes I'll get the perfect image, but other times it cuts either the head or the legs off, lol. But there's actually a solution and I'll have to make another video to help others who have this same problem. This is what you do: Upload two images for your image guidance. First, upload the main character that you want other images to look similar to. Then in the other image guidance upload box, add an image that you want Leonardo AI to use to create the same post. On that one, make sure you choose one of these two options: depth to image or post to image. Using this method usually works. However, the one caveat is that you can't use two reference images with XL models or PhotoReal. So I've just been using Absolute Reality v.16 for my model. Try that and let me know if that makes it any better.
See another comment I replied to with a similar question. Try the settings in this screnshot (but make your strength lower) Screenshot: www.canva.com/design/DAGA8OEVc0Y/MTusy_aVBPyYIuXe5MYxFg/view?DAGA8OEVc0Y&
I see you're using a default input dimension for that specific model you're using.
So if you want a full body, just use another input dimension.
Great feedback. It seems a lot of people in the comments are experiencing the same issue with creating full body images. Using two images for image guidance has been working pretty well for me but it's still not perfect.
@metricsmule not really, "standing" works well.
But I prefer a vertical input dimension.
512x1024
I'm trying to do crowd shots, it's like pushing a wet noodle. I find I can get any kind of crowd I want, so long as it's a bunch of vaguely Asian-looking guys wearing work jumpsuits and heavy coats standing in a dimly lit room or landscape, glaring at the viewer. This includes prompts that have terms like "drunken partygoers" and "smiling" and "pointing." Image to image solves that problem, but then you get into copyright issues.
I'm not sure if you have my Leonardo AI prompts database, but if you do, I added a new section where you can provide details and images for any prompt/help you need and I can usually create the type of prompt you're looking for. I wish images could be shared here because it'd be so much easier. metricsmule.com/leonardo-ai-prompts-database/
@@metricsmule Thanks, I'll take a look.
Thanks for this Pal. But damn, I've never experienced such profound frustration and annoyance, where I can't even generate the simplest of images. I'm a former professional writer, editor, English teacher, and at times have even been called a wordsmith by my ex-clients.
Yet this darn Leo AI can't even come up with basic silhouettes of two horses next to each other in plain black, on a white background after countless prompts and formulas?!
If I can't even get an incredibly simple and basic image, how the heck can I get something with more detail as I've seen in its models? I'm stymied and mystified with my constant road blocks and such contradictory and bizarre results. Moreover, my current business model depends on it. Grrr....
I understand & share your frustration. My project is creating a red, black, grey and white silhouette logo of two humans, not horses. But I can never get Leonardo AI to output anything close to usable, after countless hours and days, despite using all combinations of chatGPT prompts and Leonardo's prompt generator or improved prompts and negative prompts. He suggested trying image to image and style or content guidance or reference. I have better success using the free Bing AI image generator.
@@PositivityPlusPlus Thanks for your reply and understanding. yes you're right...image to image, content guidance and reference generates MUCH better results indeed. Best of luck to you and thanks again for you reply.
What if you domt want same faces
Use image to pose or image to depth instead.
brov.. bro (+1 sub)
Thanks!
"Same exact face, same exact person."...uh, no, not exactly. She looks slightly different in each image. Kinda like telling twins apart, they look alike, but there are subtle differences.
True. But their latest update solves all this (Phoenix).