Приветсвую !!! Парни подскажите ИИ или софт который хорошо заменяет лицо(дипфейк) и ИИ которая создает реалистичные нейромодели,stablediffusio так и не смог установить.Но и конечно жедательно бесплатные
. Спасибо
. Спасибо-
XSS.stack #1 – первый литературный журнал от юзеров форума
Нужен софт или ИИ для дипфейка - подмена лица
- Автор темы D0n1k
- Дата начала
Легкая установка Stable diffusion локально / нейроарт :: Stable diffusion :: нейросети :: art (арт)
Существует довольно простой вариант установить Stable diffusion (SD) на ваш компьютер. Даже не надо с консолью ковыряться, все гораздо проще и работает с обширным функционалом. Итак, инструкция по установке: 1) Переходим по ссылке: Скачать SD с гитхаба без регистрации и СМС2) Качаем ZIP архив с инте
joyreactor.cc
Torch может и через CPU вроде как обрабатывать, и карты AMD под линуксом: https://pytorch.org/get-started/locally/
Также на хабре есть статья как запускать модели через Pytorch на видеокартах от AMD под Windows: https://habr.com/ru/companies/ruvds/articles/835782/
Модель тут: https://huggingface.co/stabilityai/stable-diffusion-3-medium
поищи тут
/threads/81485/
/threads/81485/
Может но не так эфективно как Nvidia
Для дипфейка хороший варик Face Fusion, можно намутить оффлайн модельку и поставить на компудахтер, в реалтайме тоже можно менять лицо. Чистый stable diffusion для такого не подойдет, разве что Fooocus v2.
DeepFaceLive - как по мне отличный вариант. На гите код лежит, часто использовал, так что могу советовать с увереностью. Правда единственный минус есть, она не оптимизированая.
отвечал на соседнем боарде. решение уже у тебя, стабла которой много-много лет справляется с практически всеми потребностями тех кто создаёт топики в этой ветке.
"
specific references would be helpful, but I haven't worked with these plugins anyway
all you need is stable-diffusion-webui and ReActor plugin , and preferably Code Former , I think "Restore Former" .pth model is good for that.
most often it all depends on the level of solutions research , the mentioned above ones released and last updated a few years ago can still do the task if properly configured and properly researched
use a good and detailed base template, as well as do not allow the value for 'denoising strength' to go above 0.03-0.05, or better yet, turn it off and work with wrinkles/veins/lips.. and other things as you need them - separately via inpaint , just spending ~30 seconds on each of them.
thus you won't have any problems with pose and creativity during generation, everything will be exactly the same as on the basic template but with the face you need.
you should also devote time to searching for generation model, there are a lot of models on the open source for working exactly with faces, focusing on maximum realism.
below I will give an example of the result I got for ~10 seconds of time, it is only 1 generation without prompt. I took the first model for the base from the Internet, as well as any ordinary photo of a girl where the face is normally visible.
parameters 'denoising strenght' = 0 , generating model: realisic-vision .ckpt + its vae ;
and just 1 plugin enabled - ReActor (codeformer chosen, i am using restore-former model for it )
"
"
the first result is the compliance with the parameters for the face of the girl who is in the passport image for the test.
and the second result is the result of following the passport photo as a base.
everything is very simple and flexible, it all depends primarily on your preferences and goals, as well as on the level of research and processing of everything.
there is a lot of info on youtube on this topic, and in general the result I showed above was done in ~10-15 seconds.
If you take the time to do it, the result will definitely be even better. although I think that for passports the references I provided above are perfect, it depends on the type of passport template and the overlay parameters of the photographer.
It's all in your hands
"
переводчиком думаю умеешь пользоваться
"
specific references would be helpful, but I haven't worked with these plugins anyway
all you need is stable-diffusion-webui and ReActor plugin , and preferably Code Former , I think "Restore Former" .pth model is good for that.
most often it all depends on the level of solutions research , the mentioned above ones released and last updated a few years ago can still do the task if properly configured and properly researched
use a good and detailed base template, as well as do not allow the value for 'denoising strength' to go above 0.03-0.05, or better yet, turn it off and work with wrinkles/veins/lips.. and other things as you need them - separately via inpaint , just spending ~30 seconds on each of them.
thus you won't have any problems with pose and creativity during generation, everything will be exactly the same as on the basic template but with the face you need.
you should also devote time to searching for generation model, there are a lot of models on the open source for working exactly with faces, focusing on maximum realism.
below I will give an example of the result I got for ~10 seconds of time, it is only 1 generation without prompt. I took the first model for the base from the Internet, as well as any ordinary photo of a girl where the face is normally visible.
parameters 'denoising strenght' = 0 , generating model: realisic-vision .ckpt + its vae ;
and just 1 plugin enabled - ReActor (codeformer chosen, i am using restore-former model for it )
"
"
the first result is the compliance with the parameters for the face of the girl who is in the passport image for the test.
and the second result is the result of following the passport photo as a base.
everything is very simple and flexible, it all depends primarily on your preferences and goals, as well as on the level of research and processing of everything.
there is a lot of info on youtube on this topic, and in general the result I showed above was done in ~10-15 seconds.
If you take the time to do it, the result will definitely be even better. although I think that for passports the references I provided above are perfect, it depends on the type of passport template and the overlay parameters of the photographer.
It's all in your hands
"
переводчиком думаю умеешь пользоваться