So what do we conclude from this

So what do we conclude from this?
84 Replies
Furkan Gözükara SECourses
unet becomes your character any noise will be denoised as your character dont prompt anything and generate you will see yourself
HackerPrime
HackerPrime2y ago
I cant wait to experiment with that
Furkan Gözükara SECourses
if enough trained :d
HackerPrime
HackerPrime2y ago
Ok, so 200 steps were used
thirdman
thirdmanOP2y ago
No description
thirdman
thirdmanOP2y ago
wow wowowowow
thirdman
thirdmanOP2y ago
No description
thirdman
thirdmanOP2y ago
200 epoch
HackerPrime
HackerPrime2y ago
wow what did you use for prompt?
thirdman
thirdmanOP2y ago
photo of man art by tomer hanuka
HackerPrime
HackerPrime2y ago
😮
thirdman
thirdmanOP2y ago
No description
thirdman
thirdmanOP2y ago
photo of man art by craig paton
thirdman
thirdmanOP2y ago
interesting
Furkan Gözükara SECourses
try photo of castle
thirdman
thirdmanOP2y ago
No description
Furkan Gözükara SECourses
u didnt train any tench? this is not expected
thirdman
thirdmanOP2y ago
No description
thirdman
thirdmanOP2y ago
it sort of is, its affecting photos
thirdman
thirdmanOP2y ago
No description
thirdman
thirdmanOP2y ago
I think its impacted by weight look
Furkan Gözükara SECourses
ye unet will become yourself i tink you didnt that much overtrianed that is why
IVG
IVG2y ago
did you use blink to do the tags for the dataset?
thirdman
thirdmanOP2y ago
yeah its not overtrained
Furkan Gözükara SECourses
tags are not important he didnt do tags training
thirdman
thirdmanOP2y ago
I did this
No description
Furkan Gözükara SECourses
when text encoder training 0 if code is working as expected tags are not used at all captions only unet trained
thirdman
thirdmanOP2y ago
so text files are irrelevant?
No description
Furkan Gözükara SECourses
yep because they are not used in unet training
thirdman
thirdmanOP2y ago
yup
Furkan Gözükara SECourses
unet trains only based on image and noise
thirdman
thirdmanOP2y ago
ok so then this worked
thirdman
thirdmanOP2y ago
No description
Furkan Gözükara SECourses
nice so it appears code is not working as expected i will ask in dreambooth channel now
thirdman
thirdmanOP2y ago
I will be on standby
thirdman
thirdmanOP2y ago
Epoch 100 vs 200 same seed
No description
No description
thirdman
thirdmanOP2y ago
photo of Cyborg man cyberpunk
Furkan Gözükara SECourses
when you dont use man what happens
thirdman
thirdmanOP2y ago
this is vanilla
No description
thirdman
thirdmanOP2y ago
one second will test Prompt: photo of Cyborg cyberpunk
thirdman
thirdmanOP2y ago
No description
HackerPrime
HackerPrime2y ago
what happens if you use woman? photo of woman cygborg
thirdman
thirdmanOP2y ago
No description
HackerPrime
HackerPrime2y ago
Does that look like zack?
thirdman
thirdmanOP2y ago
No description
Furkan Gözükara SECourses
ok i found the reason even though text encoder is not trained it is used how unet is trained in dreambooth
HackerPrime
HackerPrime2y ago
so it will recognize the face as a man's face?
Furkan Gözükara SECourses
it will be still assigned to the used prompts so in this case since he uses man in file descriptions yes man and also other used words
thirdman
thirdmanOP2y ago
yes
Furkan Gözükara SECourses
will be like subject
thirdman
thirdmanOP2y ago
right
Furkan Gözükara SECourses
if you use castle castle will become like your subjec
thirdman
thirdmanOP2y ago
yup
HackerPrime
HackerPrime2y ago
ok
thirdman
thirdmanOP2y ago
blurry photo
Furkan Gözükara SECourses
so this will only make it subpar if you dont do tenc training if you dont provide any tokens i think entire unet would become you if works haha :d
thirdman
thirdmanOP2y ago
let me run one then
thirdman
thirdmanOP2y ago
Subpar, So still better to run at 0.7 when using filewords?
No description
Furkan Gözükara SECourses
i think yes i think 1 could be better as well they said they did come up with 0.7 based on experiment when you dont train unet, only text encoder will be trained
thirdman
thirdmanOP2y ago
But for next experiment Step ratio 0, only PNG files
thirdman
thirdmanOP2y ago
No description
HackerPrime
HackerPrime2y ago
no filewords either right¿
thirdman
thirdmanOP2y ago
No description
thirdman
thirdmanOP2y ago
filewords yes, but no filewords in the folder
thirdman
thirdmanOP2y ago
No description
HackerPrime
HackerPrime2y ago
ok, in the last experiment you used filewords with .txt file next to png, right?
thirdman
thirdmanOP2y ago
yes
HackerPrime
HackerPrime2y ago
k awaiting result
thirdman
thirdmanOP2y ago
@Furkan Gözükara SECourses
No description
thirdman
thirdmanOP2y ago
holy cow
thirdman
thirdmanOP2y ago
No description
Furkan Gözükara SECourses
as expected this is because text encoder still there
thirdman
thirdmanOP2y ago
No description
thirdman
thirdmanOP2y ago
No description
thirdman
thirdmanOP2y ago
have your opinion on best practice changed?
Furkan Gözükara SECourses
not yet i think both unet training and text encoder training with dreambooth still best but i cant do experiments so i cant say
thirdman
thirdmanOP2y ago
so follow your long stable diffusion video
Furkan Gözükara SECourses
this is best my gpu always full :d
thirdman
thirdmanOP2y ago
I need to buy a motherboard to plug in the other GPU
Furkan Gözükara SECourses
wow you going more power :d
thirdman
thirdmanOP2y ago
I have 2100 watt i think
Want results from more Discord servers?
Add your server