Technology A-I image and video editing?

[JohnBoy2000]

Anyone into this?

Some results I've seen are just crazy.

Looks like a tool one must familiarize themselves with and develop proficiency at.

Comfy UI is the program I've read as good for beginners.

Any do A-I image/video editing?

 

[thujone]

yea. i posted my AI PC before, don't need a high-end processor but do need a good GPU

automatic1111 is the beginner solution for txt2img and img2img

ComfyUI is a more advanced interface, not sure i'd recommend that one without some good workflow recipes already figured out

not really sure what you mean by editing as it's purely image generation

you can add images and steer it towards something similar but the results will always be generated

 

[JohnBoy2000]

yea. i posted my AI PC before, don't need a high-end processor but do need a good GPU

automatic1111 is the beginner solution for txt2img and img2img

ComfyUI is a more advanced interface, not sure i'd recommend that one without some good workflow recipes already figured out

not really sure what you mean by editing as it's purely image generation

you can add images and steer it towards something similar but the results will always be generated

If I wanted to put my head (including hair) on the body of say, a prime Arnold Schwarzenegger, that's what I mean.

Or make myself look incredible for some Tinder profile pics, maybe partying on a beach in Ibiza or with some hot chicks etc.

This type of thing I mean.

 

[thujone]

you could, but since the image is completely generated your head would also end up being rendered as opposed to just photoshopped in

 

[JohnBoy2000]

you could, but since the image is completely generated your head would also end up being rendered as opposed to just photoshopped in

But I would use a pic of me, then a pic of someone else's body, and use the A-I tool to splice it in seamlessly.

That's how it works right?

That's how "deepfakes" are made, no?

 

[JohnBoy2000]

yea. i posted my AI PC before, don't need a high-end processor but do need a good GPU

automatic1111 is the beginner solution for txt2img and img2img

ComfyUI is a more advanced interface, not sure i'd recommend that one without some good workflow recipes already figured out

not really sure what you mean by editing as it's purely image generation

you can add images and steer it towards something similar but the results will always be generated

Do you have any examples of images you've done, also?

 

[thujone]

But I would use a pic of me, then a pic of someone else's body, and use the A-I tool to splice it in seamlessly.

That's how it works right?

That's how "deepfakes" are made, no?

no. the best way to add your face is to train a LoRA. for that, you'd need 30+ images of you cropped to your head with good lighting, like a headshot for a passport photo

this has to be done carefully as too many angles or expressions could lead to your facial details being averaged out.

i have yet to see a convincing deepfake, btw. txt2speech is pretty good now but image/video is only convincing if the original source is already low fidelity (e.g. CCTV footage).

the most realistic AI images of people now are on the level of the ones seen here - https://thispersondoesnotexist.com/

Do you have any examples of images you've done, also?

this is a cabin in the woods done with an older SD1.5 model -



ruins of a lost civilization, still SD1.5 but a different model -



this is a cyberpunk bluelighter done with a semi-realistic SDXL model -



this is a dieselpunk bar girl with the same SDXL model -