Hi. Long time, no post. I want to post some new works and what I did with AI past weeks. The mentioned new HiDream AI wasn't that good (for me) so I did not very much with it. Then I discovered Chroma which is a further development (training) of FLUX. Its character is that this AI creates colorful images and by that offers a certain style I like. It also likes detailed prompts, so you can create more complex images. Here are a few, so you can have a look what style. https://imgur.com/a/Vgff4Sh OK, then I moved again to video creation. There is a new AI called FusionX which is based on WAN 2.1. With that I played a lot past weeks. Especially with using controlnet. Controlnet is a bunch of AI that is specialized to grab certain details of original images. For instance movement and positions of human characters. DWPose. The next popular ones are Depth and Canny. Those AI then pass the info to the actual FusionX AI. Without going into tech detail, some examples. This tech is used by the movie makers in Hollywood already years ago. (CGI) If you want that an AI character does a special move you actually can film yourself and then let pass your movement to the AI character. I took a video where a real woman danced. Then passed the movement to the Ai character. By that the video is that much real and human.... one can hardly distinguish from real anymore. My AI creation I And to another (only the moves are the same, they are from the same controlnet) My AI creation II And finally I controlled 2 women dancing in a night club (different controlnet). This time I post the original one as well. Original video from where DWPose extracted the moves, real women dancing: Then my AI creation with FusionX AI This is really fun....consider you make some fun and record yourself....then pass your crazy moves over to your AI scenery... OK, bye for now, see you soon again.
Hi FusionX AI plus VACE (controlnet) is still much fun. I want to show what you can control. It's not only people. I wanted to animate my painting. Since it is a vortex with a god like figure I simply took a video of a real whirlpool and passed it to the controlnet, this time DepthAnything: Then I wanted to make fireworks a bit more interesting. I took a rather boring and colorless real firework for control.... Similarly I took a video of rather boring and colorless fountains. Then I took a real tornado, created a surreal scenery and let it destroy a wooden house. I tried to control animals. Here I took a video of a real cat roaming around. I created a surreal scenery as well. And a cute kitten. The real cat was completely different, the room and the ball also. Then I upscaled the video and also did other kitten. Here an example. Finally I took a video of somebody who does saber practice and applied an AI character and fire. And I took a video of a drone like perspective moving through an 'ordinary' forest. I created fluorescent mushrooms and a fantasy like scenery. Bye, see you soon again with latest AI creations. Yen
Which kind of AI creations would scare you? I am interested in that topic, because I have objections sometimes as well. I just have installed a new model and created a new workflow and ran the workflow. This kind of AI requires a certain and proper moral to use, IMHO. I actually wanted to test it more before I post here some results. But to your post I reply with my first result. (Turn on your speakers)... What you need: You need a few sentences of audio of a certain person as reference voice. Then you need one image. That's all. The video length doesn't matter.....if you are patient to get it fully rendered. This is the result: IMHO such creations should be done with respect to the person....
The one starting to invade our everyday lives without being asked for? The one companies like MS try to shove into your face constantly? I do not want nor need AI in my life, except for very specific purposes, and this should be respected.
hm. real=real; with all due respect, yen; i did not like your choice of subject one bit. there are enough real girlies around, fortunately. no need to construct fake ones with ai.. it got me scratching my head where nsfw ends and where moderation ought to start.. you can call me a dinosaur any time you like..
Yes, I totally agree. MS recall and the like..... I respect your opinion Anyway I don't agree. My choice of subjects has a great variety. But when you want to find out how good AI can create humans, you have 2 choices. Create man or woman. So being hetero I rather enjoy to go for the latter. To create something that is human or unique to human has a special quality. Since we are human we very quickly recognize differences there. It's a good criterion to categorize the quality of an AI. Is there one litte 'error' you recognize it immediately. The way of arguing is like saying to a photographer why do you make photos of women? There are enough real women out there. This kind of AI btw. will destroy a lot of jobs of photographers. Nobody actually needs real (expensive) photo-models anymore. Here I'd see a negative argument. Well, NSFW.....the definition is vague. But should not differentiate from real images. I guess I did not exceed the limits. ....somebody who is working is usually adult as well
that photographer is creating a picture of an individual within a context, and in a moment in time, but this software is depicting a construction. that`s why i said real=real.
Hmm.. . There is still photography as well. (depicting a construction, too) This would be another subject for discussion. One strictly could say. No, it is the same. A photo is as (un)real as an AI image output. Finally we get an image at both. Technically it is absolutely the same, pixel, dimension, even the file format. image= imāgo (imagination, likeness, "Abbild, Vorstellung"). Either way. I like to make photos since I am 12 years old. And I like to play with diffusion AI since last Christmas. Both combine tech. with art. That's why I like both. Nothing compares to real persons. The primary motivation to post a few AI creations from time to time here (which is actually the photography thread) is to demonstrate what is possible.
it is not same; a photo has a context, and a story to tell. but you can depict any old thing with AI, very sophisticated, and hi-tech, but there is no story. no context, nohing at all... it all depends on the conditions you impose on it..
Well, I guess there is a general, much greater difference. The one who has created it / made the photo vs. somebody who 'simply' has a look at it. I'd say this difference is much greater than AI image vs. digital photo. There is for sure much more context / story to the creator than simply to a viewer. When I did a particular image, there is for sure context and story, but a different one. It is comparable to still photography. You have a vision, idea...then you arrange the scenery, take a view shots, change things, change light.....until you are satisfied with the result. This is very similar to AI. You make the setup, you write your prompts, you choose different tech items like scheduler, you have different runs..... The big story that is tied to photography is the moment the photographer took the photo., yes. Especially out at nature.....but for an unknown viewer the story soon 'degrades' to attributes of what is actually a good photo. And those also apply to AI images.
Civitai is the most popular site where people can publish their creations made with opensource AI. The site actually has 2 versions, a green and a 'full' version. If you login to the full site and disable all NSFW filters you can see what's in peoples mind and was created. Since AI has huge flexibility you can not only mix sexes, but also genera. You can mix animals, create chimera, mythical aspects....etc. You can distort proportions, can create absurd sexual organs, mix styles etc..... TBH some creations are really unsettling at first sight....some are really ugly. When I create female characters I always try to let their human essence shine in a way. But that is just my idea. Civitai is a really good platform with social media functions. The community is helpful and nice. People train their own stuff (LORAS) and publish them there for a try. Others even train entire checkpoints and release them there. For instance FusionX has been trained and merged by a girl named vrgamedevgirl84. She is actually game developer and very helpful at civitai. I also have an account there since January. It's great to exchange knowledge and share results
Impressive. Out of words... I love Natalie Portman btw, but the MDL greating was warming. (Yen, i didn't know you was so close to here... )
Lol, I wish I would. No honestly. I love her, too. She is one of my favorite actresses, especially at V for Vendetta. Star Wars of course. ATM I am doing more tech stuff, creating new workflows. As soon as they work as expected I create new stuff. I noticed that you have replied and I just wanted too. Here a cute little dog I did with my favorite workflow. Vimeo always makes my videos a bit of being choppy. The original is actually much smoother...dunno why. cya soon again.