Ionic Spark
alright so if we talking deepfacelab.... how much time consuming is it? i have an rtx 4090 & intel 13th i-9 processor. can u give me a rough estimate how much time will a well made single df vid take? i will hv to decide on that basis coz i even hv work & priorities. I would hv looked into AI video faceswap but again i hv no idea how to make these. so if i dont hv a guide i wont understand a thing & would quit. so i guess i really dont hv much of a choice here.
Social Media Girls
-
Going forward please follow these rules when posting:
-The content MUST be from TikTok, do not post slips from other platforms.
-You MUST include profile names / link to the profile, or make every attempt to include it.
-Include timestamps if you post a long video or a full live video.
-Make sure ALL girls posted are 18+
Ionic Spark commented
okay, if I set sections on making dfs on DFL I'll start with:
*Legends*
" + " - means you'll required to do every dfs.
" - " - means you only need to do once.
" > " - optional.
(-) 1. Compile vids of your target idol, select best angles, convert into jpg/png, enhance, and patch it to be readable for DFL. (This is now your src faceset)
(+) 2. Select your target porn, convert to png, extract aligned faces, remove unnecessary images, re-extract aligned images to fix debug images (This is the dst images)
(>) 3. Apply Xseg Generic whole face mask for src and dst, apply xseg labels for dst for accurate masking.
(+) 4. Training should be dynamic, some settings you need to enable/disable then adjust them according to your model needs.
In 2022 you have an option for using models in DFL but now they remove the downloadable pre-trained models so making new one is the only option, but if you're able to create a good one you can reuse it now for every df you'll make.
(-) 5. If you already have a model you just need to train for a bit to learn the new dst video.
(+) 6. Merging will show how your model perfoms, there are multiple settings to choose which best blends even for small details every single frame or you could also pick that one setting and apply it to every frame.
(>) 7. Enhance the merged images for finer blend.
(+) 8. Merge to mp4
"This option is for a well made customed DFs"
Pros:
1. Lots of options for customization
2. Reusable model
3. Covers some very intricate face angles
4. Produces consistency when it comes to facial expressions, eye/mouth movements and easy to hard oclussions
Cons:
1. Requires manual work
2. Training takes time, but starting a new one is prone to model collapse if settings are not properly set.
3. face generation may be blur if the src are not vast enough to cover hard angles
For AI Video Faceswap:
(-) 1. Set your best settings
(+) 2. Edit your target video with less head movements and select or drag it to the app
(+) 3. Select the best frame that matches your single image (up to 3)
(+) 4. Export
"This way I think would make the best output"
Pros:
1. Fast and less manual work
2. Easier to use and generate
3. Can produce multiple dfs with just a single image in minutes
Cons:
1. Face generation and resolution is solid but sometimes it looks weird (ex. eye movement)
2. Some random face warp when sudden movement
Considering your availability, AI Video Faceswap works solid for your time and your pc.
*Legends*
" + " - means you'll required to do every dfs.
" - " - means you only need to do once.
" > " - optional.
(-) 1. Compile vids of your target idol, select best angles, convert into jpg/png, enhance, and patch it to be readable for DFL. (This is now your src faceset)
(+) 2. Select your target porn, convert to png, extract aligned faces, remove unnecessary images, re-extract aligned images to fix debug images (This is the dst images)
(>) 3. Apply Xseg Generic whole face mask for src and dst, apply xseg labels for dst for accurate masking.
(+) 4. Training should be dynamic, some settings you need to enable/disable then adjust them according to your model needs.
In 2022 you have an option for using models in DFL but now they remove the downloadable pre-trained models so making new one is the only option, but if you're able to create a good one you can reuse it now for every df you'll make.
(-) 5. If you already have a model you just need to train for a bit to learn the new dst video.
(+) 6. Merging will show how your model perfoms, there are multiple settings to choose which best blends even for small details every single frame or you could also pick that one setting and apply it to every frame.
(>) 7. Enhance the merged images for finer blend.
(+) 8. Merge to mp4
"This option is for a well made customed DFs"
Pros:
1. Lots of options for customization
2. Reusable model
3. Covers some very intricate face angles
4. Produces consistency when it comes to facial expressions, eye/mouth movements and easy to hard oclussions
Cons:
1. Requires manual work
2. Training takes time, but starting a new one is prone to model collapse if settings are not properly set.
3. face generation may be blur if the src are not vast enough to cover hard angles
For AI Video Faceswap:
(-) 1. Set your best settings
(+) 2. Edit your target video with less head movements and select or drag it to the app
(+) 3. Select the best frame that matches your single image (up to 3)
(+) 4. Export
"This way I think would make the best output"
Pros:
1. Fast and less manual work
2. Easier to use and generate
3. Can produce multiple dfs with just a single image in minutes
Cons:
1. Face generation and resolution is solid but sometimes it looks weird (ex. eye movement)
2. Some random face warp when sudden movement
Considering your availability, AI Video Faceswap works solid for your time and your pc.