by

Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
  • Aug 02 06:32
    kolessios closed #182
  • Aug 02 06:32
    kolessios commented #182
  • Aug 02 05:17

    kolessios on v1.5.5

    (compare)

  • Aug 02 05:16

    kolessios on canary

    - Minor improvements to the des… (compare)

  • Aug 02 04:15

    kolessios on canary

    - Minor improvements to the des… (compare)

  • Aug 02 03:37

    kolessios on canary

    - Minor improvements to the des… (compare)

  • Aug 01 20:16
    asjdf labeled #182
  • Aug 01 20:16
    asjdf opened #182
  • Jul 31 14:10
    @kolessios banned @deeptester1
  • Jul 31 13:08
    @kolessios banned @xrebel1
  • Jul 30 09:23

    kolessios on v1.5.4

    (compare)

  • Jul 30 08:50

    kolessios on canary

    - Fixed: Some photos were not u… (compare)

  • Jul 30 06:58

    kolessios on canary

    - Fixed: Permissions are now se… (compare)

  • Jul 30 06:50

    kolessios on v1.2.11

    (compare)

  • Jul 30 06:49

    kolessios on canary

    - Fixed: Permissions are now se… (compare)

  • Jul 30 05:01

    kolessios on canary

    Possible fix: Mac OS. (compare)

  • Jul 30 04:50

    kolessios on canary

    Possible fix: Mac OS. (compare)

  • Jul 30 04:41

    kolessios on canary

    Possible fix: Mac OS. (compare)

  • Jul 30 04:25

    kolessios on canary

    Possible fix: Mac OS. (compare)

  • Jul 30 04:06

    kolessios on canary

    Possible fix: Mac OS. (compare)

Iván Bravo Bravo
@kolessios
On the Internet I see that it could be related to the lack of an Intel library when compiling DreamPower. It still seems weird to me since as I tell you, this error occurs only on some systems. I will compile a test version so you can tell me if that solves the problem. Do you have a Github account?
Dukiverse
@Dukiverse_twitter
@kolessios I found the fix. I don't know if it is the best fix but I found someone had the same error when importing torch, and removing (or simply renaming) C:\Windows\System32\libiomp5md.dll fixes it because its a conflicting dll
Iván Bravo Bravo
@kolessios
Yeah, I just saw that too. So does that fix your problem?
Dukiverse
@Dukiverse_twitter
It seems like it did work! I was able to move on and download the checkpoints! Thank you for your help.
Iván Bravo Bravo
@kolessios
Thanks to you! Now there is more information on how to fix that problem.
Iván Bravo Bravo
@kolessios
@svnty From what I could understand: It seems that Maskdet is generated correctly but Maskfin only generates the green layer of the clothes, correct? If so, it means that the algorithm is having a hard time using your Maskdet to generate the Maskfin. What I recommend is to keep changing the Maskdet until the Maskfin is generated correctly or try another photo. Keep in mind that the generation of masks still has part of the limitations of the algorithm, creating a mask for the ass or where the body is at strange angles will continue to cause problems.
Jake S. Walklate
@svnty
Okay, cheers
knightrider
@knightrider1314_twitter
hi does anyone has a guide how to correctly install and run cuda 10.2 because i searched the internet and it is a bit mess
Iván Bravo Bravo
@kolessios
@knightrider1314_twitter At least in Windows the process is very simple, go to this page, download the installer, install and you're done. https://developer.nvidia.com/cuda-10.2-download-archive
knightrider
@knightrider1314_twitter
i did it but the program says that i need to install it but i have it already installed. also the nudification has changed for me because in the photos it does not bring the same results like before
Iván Bravo Bravo
@kolessios
@knightrider1314_twitter Have you restarted your computer? Sometimes after installing CUDA it is necessary to restart for it to take effect. The different results can be caused because you have set the Preferences Mode option to Advanced (Custom masks), as its name indicates it is an advanced mode only recommended for users experienced with creating custom masks.
jim-s5
@jim-s5
Hi, I would like to ask about the ImageToNearest step in the latest versions of Dreampower. The --ignore-size argument which I'm using in CLI should not have any resolution limit but every photo is resized with the ImageToNearest in certain resolutions (496x496, 512x512 etc). Why is that?
deeppppp
@deeppppp
The --ignore-size was implemented as a first step towards accepting any size image for all stages (or steps) of the algorithm.
but not all stages can handle that yet (work in progress).
Iván Bravo Bravo
@kolessios
@jim-s5 I think a more specific answer would be that... if I'm not mistaken, pix2pix requires that the dimensions of the photos can be divided by 16, I don't know what the correct term would be, but ImageToNearest takes care of that.
jim-s5
@jim-s5
Yes, I saw that because the MaskrefToMaskdet step is always in different resolution from the Mask itself
So for example if I run a mask in 500x500 with -s 3:3 it produces a 496x496 Maskdet
Iván Bravo Bravo
@kolessios
Yes, basically what that step does is to resize the photo to the closest number that can be divisible by 16. I can't tell you why, since I don't know yet, I haven't investigated more about it but it was necessary for it to ignore-size fulfilled its function.
Located in Config because DreamPower right now is a code spaghetti project.
jim-s5
@jim-s5
Surely I will have a look on that
Iván Bravo Bravo
@kolessios
Maybe this may interest you too, it was one of the parts that I had to modify to make ignore-size work: https://github.com/dreamnettech/dreampower/blob/canary/src/transform/opencv/mask.py#L36
deeppppp
@deeppppp
@kolessios is working on the implementation of ignore-size for all steps.
in general the input should be 512x512 for the pix2pixHD AI steps, because that's what the checkpoints (models) are trained on.
and up or down sampling should occur before and after that with opencv
any resolution difference from 512x512 given to the AI will yield lesser results.
Iván Bravo Bravo
@kolessios
That's right, it is already implemented but any modification to the code I doubt that will help improve the result, that depends on the checkpoints.
knightrider
@knightrider1314_twitter
@kolessios i have tried everything but the most troubled thing that affects me and i don't know the reason is about the nudification results.For example i have tried the same photo in the new version and the old version earlier and the result from the older one was what i expected but the new was not even close is something the problem or what ??
Iván Bravo Bravo
@kolessios
@knightrider1314_twitter That may depend on several factors, if you want you can send me a private message with the photo that has the problem and the preferences you used in the old and the new version to compare from my side.
Iván Bravo Bravo
@kolessios
@/all 🎉 DreamTime v1.5.5 has been released! Enjoy a simpler navigation thanks to the "Forward" and "Backwards" buttons, the updater now handles errors much better, the window dimensions are now saved and several other bugfixes. https://github.com/dreamnettech/dreamtime/releases/tag/v1.5.5
Mick Crown
@MickCrown_twitter
Hi guys, I'm new to DreamTime, but I wondered how do I get a better nudified picture?
Are iterations worth? with 100 iterations I don't see any significant improvement, am I doing something wrong?
Iván Bravo Bravo
@kolessios
@MickCrown_twitter If with iterations you mean to "continue training the algorithm", DreamTime does not work like this. Each photo you nudify with DreamTime/DreamPower goes through the DeepNude checkpoints, these can no longer be improved so the only way to improve the result is with manual modifications to the photo using a different scale method or the "Custom masks" mode. However, this will only bring small improvements within the limits of the pre-trained checkpoints. In other words... the photos you nudify do not improve the quality of future photos since it is not possible to train the algorithm with new photos/iterations.
By the way, welcome to the community! This is the guide in case you are interested in trying the "Custom masks" mode: https://dreamtime.tech/docs/guide/custom-masks
Mick Crown
@MickCrown_twitter
Sorry, so "Number of times the photo will be nudified."
has no meaning? WHy there's this option?
Iván Bravo Bravo
@kolessios
The "runs" are used to process the same photo through different preferences (Progressive or Random) and see which one is the best.
This means that the runs are useless if you don't activate the Random/Progressive option below. Sorry for the confusion if that was the case, it can be considered a UI bug.
Iván Bravo Bravo
@kolessios

I've seen users having trouble or even giving up on DreamTime just because they did this without knowing what it does. I guess I will have to remove this option in a new version. 😕

Piergiox
@Piergiox
@kolessios hi, I’m trying the new version of DreamTime and sometimes when I edit and upload a custom maskdet, the maskfin has only the classic green to mask clothes without the other colors. And if I upload a custom maskfin, the result isn’t good
Iván Bravo Bravo
@kolessios
@Piergiox This can happen if the Maskdet is outside the limits of the algorithm, the Custom masks are very useful for making small corrections that are still within the limits of a good photo, if you try to create masks for positions or angles for which the algorithm was not trained then the Maskfin will come out as if it were the Maskref (only cloth)
Michael Landinger
@Lndngr
Hi, first thanks for your great work. Is there an opportunity to edit the "Maskdet " mask? And if so, is it possible to provide a short HowTo like for the "Maskref"?
Iván Bravo Bravo
@kolessios
@Lndngr Hi and welcome! If you mean a tutorial to use the Custom masks, this guide has information about it: https://dreamtime.tech/docs/guide/custom-masks
Piergiox
@Piergiox
Ok @kolessios thank you
nickzaiz10
@nickzaiz10
Hello
Can anyone explain waifu2x please?
Iván Bravo Bravo
@kolessios
@nickzaiz10 Hi and welcome! Need help with Waifu2X? Tell me what your problem or question is and I will try to help you.
nickzaiz10
@nickzaiz10
@kolessios just need do know what functions it does, no clue about waifu2x does
Iván Bravo Bravo
@kolessios
@nickzaiz10 It is an algorithm to increase the resolution of the photo and try to decrease the noise generated by the low quality. In DreamTime if you do not use the Overlay scale method, the resolution of the exported photo will be 512x512, when passing it through Waifu2X you get a photo of 1024x1024 or higher.
IlyaPavlovich
@IlyaPavlovich
hi! will there be a feature to nudify girls from a behind view? :)
Iván Bravo Bravo
@kolessios
@IlyaPavlovich Hi! It is a plan for the future, however there is no planned release date. Meanwhile you may be interested in xxxPaint.
Nemesis
@Nemesis-9
Anyone got the discord ?
Iván Bravo Bravo
@kolessios
@Nemesis-9 We don't have Discord. We tried 2 times but those times the server got banned.
mateogsm12
@mateogsm121_twitter
hey i cant download the program, dreamtime because google says "the connection is not private" "Attackers may be trying to steal your information from gateway.ipfs.io" any solution? great work btw :)!
ThrillKill187
@ThrillKill187
is there any guide on how to crop? i mean is it so, that the middle fixation point (cross) must be right between the breasts or is ist just so that the AI can read it no matter what ? also is there an explanation what i could do to manually enhanche the picture before ? (lower contrast, higher contrast etc) also when is the best time to use crop, when to overly and how is padding actually working - just telling me to have the "colors visible" on both sides is a bit to less to understand how it works.
THX :) pls keep up the great work
John Summers
@Aximili55_twitter
installed cuda on 1070 and it still says cuda is not installed