sukebenet / dreamtime

Use artificial intelligence to create fake nudes.

DreamTime

Available for Windows, Linux and Mac.

⚠️ Project discontinued.

Thank you for your interest in DreamTime. Unfortunately we have discontinued the project and will focus our efforts on other projects, if you are interested in a better nudification algorithm that can be used from any device visit our new project:

https://www.sukebezone.com/

We have taken the decision to archive the repository since no PR has been received for more than 1 year.

Download

Download the latest version of DreamTime here!

You can also visit the Releases page.

Development

Visit the Getting Started page to start hacking DreamTime. 👩‍💻

Community

• Discussions
• Twitter

Sponsors

🥰 Support this project by becoming a sponsor. Your logo will show up here and in the about section of our apps, with a link to your website.

[Become a sponsor]

dreamtime's Issues

App thinks we are offline

Describe the bug
The app thinks we are offline.
Probable cause of the problem : the .env is present but not loads on time or not at all by the app -> the app have no Nucleus id -> it fails to checks the remote nucleus api -> the app thinks we are offline.

To Reproduce
Starting the app go to the about page but it happen randomly.

Environment :

• Version 1.1.0
• OS: Ubuntu 18/16, ArchLinux, MacOs
• Report URL: Down

Applying the CropperJS function reduces image quality at GUI

The input image becomes blurry when cropped.

Ideas/Suggestions

Hi everyone,

Here are some Ideas I had for the next release. Although I'm not fluent enough in Python or C++ to be of any real assistance, I'm pretty sure that the scale of these proposals is very small. They shouldn't require any modifications to the neural network, just the software.

• Increase zoom size for finished photos in order to better view the result

• When increasing the execution number, automatically offset each cropped photo a random amount before feeding it into the Net, and then correcting it afterwards, in order to increase the variety in each execution

• Add an option to transform photos within a specified size range (for example: I could specify a range for boob size from 0.9-1.5, and the program would automatically generate 7 photos, one for each size

• ~~Add a button to compare different finished photos (it could even work in tandem with the previous suggestion, giving it a slider to shift between each photo in the range you specified) ~~

• Add GIF support to the GUI version

• Add a debug mode where we can dump each step of a transformation, in order to view which stage it went wrong

• Allow batch jobs (multiple photos in a queue)

These are just a few ideas I brainstormed up while playing around with your software, I would love to hear your feedback on the feasibility of some of these additions!

v0.1.2 The transformation has been interrupted by an CLI error.

Error: The transformation has been interrupted by an CLI error.
This can be caused by:
-A corrupt installation (commonly: The checkpoints folder was not found in cli/)
-Incompatible system
-If you are using GPU: The NVIDIA graphics card could not be found -If you are using CPU: Insufficient RAM.Buy more RAM!
The CLI has reported the following error, take a screenshot to get more information:
[18476] Failed to execute script main Traceback (most recent call last):
File "main.py", line 101, in
File "main.py", line 74, in main File "run.py", line 167, in process
File nopencv_transform\dress_to_correct.py", line 12, in create_correct File "opencv_transform\dress_to_correct.py"f line 19, in correct_color AttributeError: 'NoneType' object has no attribute 'shape'

QOL: 2nd cropping step

from jesscold:

feature request: in the cropper tool, give the ability to have another 'cropping' step where you draw a box around the "object" you are interested in. The tool would then crop that out and pass to CLI (padding it to make 512x512) at the end, it would crop out the same coords from the result and overlay on the original. this way we can easily work with images containing multiple people

actually a lasso tool would be better

GPU option by name not shown on linux at [Process Device] page while nvidia card is present.

GPU option by name not shown on linux at [Process Device] page while nvidia card is present.
When using the Custom GPU ID, the CLI will use nvidia card correctly.

Operating System: Debian GNU/Linux bullseye/sid
Kernel: Linux 4.19.0-5-amd64Architecture: x86-64
i7 3930k @ 4.6Ghz

Is the GPU name property not available on linux?

Manual User Interface

At the basic level of implementation:

A "draw in feature" what I mean by that is, perhaps the user could help mark the targets for the ai to better understand the area you want to target, this could help improve results until the algo itself is improved, so for example you could have "auto" mode which is what the program has now which is that it tries to render to the best of its ability what it thinks are the targets, or you could either a) manually draw in or b) using a selection tool set to pick the areas you want to transform.

As an example you could have a tool to mark the nipples, the outline of the breast, or the lower regions front or back. This might help clear some hurdles as far as how close the distance is, if there are any obstructions, or help mark the check points like where the belly button is since I read (I believe) that is one of the areas it uses to detect removal of clothing. With a manual adjustment system. You could have different markers with different labels and drag and drop them and resize them to the relative area or draw them manually. These types of tools are used a lot when you use custom faces for games and things.

I just thought it might be a way to improve the quality of the tool until it is at a point where it is better understood and can do more diverse angles, shapes, and sizes by itself. Even if the auto mode was substantially improved manually highlighting things and sectioning off things as described would still come in handy for specific situations regardless. Plus the human eye can see things that sometimes an AI can't and manually marking them will help in that regard.

At a more advanced level of integration:

This would be on the more advanced scale after the initial tools were created but since each area could be manually selected theoretically you could choose relative skin tones with a dropper, nipple size, lighting, and other things to help the results blend with the original picture. Just some thoughts on what the benefits could be further down the line if you build this type of system from the early onset.

Apologies if this explanation is at any point disjointed, I copy, pasted, and edited from the discord these ideas I had and at request posted here to keep it visible.

As an addendum:

This is just some more spit balling from the discord that might help illustrate it in terms of the process that would occur theoretically:

"you can imagine running phase 1, show the image with the mask in GUI and allow user to adjust.
have the adjusted as input to phase II show the image with masks in GUI again and allow user to adjust.
have the adjusted input as phase III

We can take the adjustments and use to train the algo on.. so it does it's job better at phase I and II...

it would solve issues like cyclops boobs and stuff bouncing around ... phase I & II are responsible for those issues"

deeppppp

In GUI Result Image saves without proper png extension

In the save prompt window .png has to be typed in manually after file name saving in windows 10 or it saves as a file without an extension, when you go to save the file type is under "all files" and not png.

This isn't a major issue but a mild annoyance, not sure if it's the same using cli as it is in gui or if this is because I don't have extensions on but either way saving as a png automatically would be ideal since that is the proper file extension for the image.

QOL: setup.ps1 / requirements-[OS].txt has pyTorch wheel for Python 3.6

setup.ps1 / requirements-[OS].txt has pyTorch wheel for Python 3.6

"https://download.pytorch.org/whl/cu100/torch-1.1.0-cp36-cp36m-win_amd64.whl"
causing a failure when the user has a different Python version than v3.6 - like current stable build 3.7

can we change this out so that the pyTorch for the users Python version is grabbed and installed instead?

QoL - Allow Photos to be queued/ Allow multiple photos to be imported at once

A simple queuing system would go a long way for the UX of the platform! Keep up the good work!

Cancel In Progress Render

Repeating an idea someone said on Discord: A button that can cancel an in-progress render.

RuntimeError: generic_type: cannot initialize type "_CudaDeviceProperties": an object with that name is already defined

seems related to:
pytorch/pytorch#16559

solutions:
"Pre-initializing CUDA in the main thread mitigates the issue"

avoid issue:
"if you try to create a CUDA tensor in a thread it produces this error"

Error output:

After launching the app, it writes out this error with instructions to post the url here. so here it is:
https://rollbar.com/occurrence/uuid/?uuid=3f71458f-62de-4829-bf40-c81f4079e379

Thanks for any support.

QOL: Detect if WIndows N version and display custom error message

N versions of Windows 10 require installing the Media Feature Pack.

Can we do a version check and customize the error message do point users to the Media Feature Pack instead of the generic message they get now?

Reading the instructions that mention the Windows N issue is hard :)
and users may not even know their version of Windows.

Transformation #1 failed - python

python: can't open file 'main.py': [Errno 2] No such file or directory

How to run DreamTime on Debian as root with CPU rendering

To render a photo as root using CPU on Ubuntu/Debian (personally I use Kali Linux 2018):

1. First off, prerare your app. Rename DreamTime file in DreamTime/gui folder to DreamTime.sh
2. Open a console in the folder with DreamTime and make these procedures:
   cd DreamTime/cli
   chmod +x cli
   cd
   cd DreamTime/gui
   chmod +x DreamTime.sh
3. Run in terminal:
   ./DreamTime.sh --no-sandbox

Setup of my computer:
Intel Celeron N3350 (2 cores, 2.3 GHz), 4Gb LDDR3 RAM.
Rendering took 174 seconds.

Execute on ubuntu

Hello I have ubuntu 16.04 LTS

I give permision of execution at cli file, then when execute trhows this messages:

GPU IDs: [0]
Executing phase: dress_to_correct
Traceback (most recent call last):
  File "main.py", line 101, in <module>
  File "main.py", line 74, in main
  File "run.py", line 167, in process
  File "opencv_transform/dress_to_correct.py", line 12, in create_correct
  File "opencv_transform/dress_to_correct.py", line 19, in correct_color
AttributeError: 'NoneType' object has no attribute 'shape'
[13483] Failed to execute script main

I made something wrong?

Thanks!

Torch Dependency on Macbook

For Mac users
torch-1.1.0-cp36-cp36m-linux_x86_64.whl
Should be replaced on ubuntu-requirements.txt for this one:
torch-1.0.0-cp37-none-macosx_10_7_x86_64.whl

Transformation #1 failed

Transformation #1 failed
DreamPower has reported that the photo has been transformed but the file does not exist! This may be due to a problem saving the photo, verify that DreamPower has write permissions and that your Antivirus is not detecting false positives. It is also possible that DreamPower is ending abruptly due to a major problem.

For more information please report the following URL on Github or here: https://rollbar.com/occurrence/uuid/?uuid=8b1428ca-83c3-4073-d167-196c5e33f5ac

Transformation #1 Failed

Trying to run in windows 10 but without success.

Updated windows and all drivers, and tried using both cpu and gpu and reinstalling.

The error I get is

Transformation #1 failed
The process has been interrupted by an error. This can be caused by:

- A corrupt installation
- Insufficient RAM. Buy more RAM! (8 GB recommended)
- If you are using GPU: The graphics card was not found, it is not compatible (minimum: 3.5 CUDA compute capability) or the drivers are outdated.
- Check the console for more information.


For help, please report the following to one of our developers:

https://rollbar.com/occurrence/uuid/?uuid=cfe7433d-6aca-4685-df94-111ed6e203c3

v1.0.0-Convert picture error display save link

The converted image is displayed as a blank image and cannot be saved locally. Even if you save it locally, it's a link shortcut.

Setting to Change Output Dir

This was in the last issue tracker so I guess I'll put it here too. Changing the output directory for files will be a useful feature for some users.

GUI input image size is restricted to 5Mb

QOL:
Minor issue but still a little annoying if you want to transform from a large file and have to use an external software to resize or compress to avoid that limit.

can we have CLI verify the input image file exists?

currently it will throw this error if the input file doesn't exists:

cli -i test.png -o out.png
GPU IDs: [0]
Executing phase: dress_to_correct
Traceback (most recent call last):
File "main.py", line 101, in
File "main.py", line 74, in main
File "run.py", line 167, in process
File "opencv_transform/dress_to_correct.py", line 12, in create_correct
File "opencv_transform/dress_to_correct.py", line 19, in correct_color
AttributeError: 'NoneType' object has no attribute 'shape'

and that is not very descriptive...

GUI result image is white rectangle only

Hello.
When I tested your app, it produced only empty white image. After save attempt is created file only with size 1kb, which is containing word "undefined". Version 0.1.2.
OS Windows 10 64b, used GeForce GTX 1050 Ti and also CPU for rendering.

QoL - Allow dragging and dropping of images from browsers

So I can drag and drop from a browser, rather than having to save the file somewhere first.

Long input file names causes failure to write intermediate output files

If the input file name exceeds a certain amount of characters (haven't determined the exact amount), no intermediate output files are written.

app-builder-bin-2.7.1.tgz: unexpected end of file

Hello!
I've got an error on the second stage of "yarn install"

error An unexpected error occurred: "https://dreamnet.tech:4873/app-builder-bin/-/app-builder-bin-2.7.1.tgz: unexpected end of file

Copy Result by Right Clicking

Copied from the last issue tracker. Basically, since electron lacks chrome's context menus, users can't right click to copy/save images. I worked on fixing that using electron's nativeImage api, but i stopped because I couldn't make a way to satisfactorily tell the user that an image was copied.

Semi-automatic mode: editing the mask

Is it possible to add a feature of manual editing the mask after the app does it? Just to get better results if neural network did it badly.

Python 3.6.2+ required on Mac builds

pytorch/pytorch#17237

can we increase the Python version requirement in the README to 3.6.2+ ?

Discord server disappeared

NOt working with GPU anymore

I have a 965M TI and it was working good before the update now keeps saying fail

Add GitHub Release Links to ReadMe

If a user tries to get the checkpoints and the Cli/Gui from Mega, they will go over the limit for regular users and be forced to wait. You should put the Github Releases on the Readme so people have a way to download everything at once.

GUI/CLI - Set multiple runs and auto-output for multiple iterations

I'd like to request that a feature be added that runs the same image multiple times, so you can set-and-forget an image (cropped) to generate and save, so that after a user-defined number of runs, the user can pick the best result.

This is because of the sheer amount of RNG in the application, and that it never creates the same result twice. This way, instead of having to manually re-run to get a good result, you can set it to run multiple times overnight, to get the perfect result.

Selecting custom GPU with value 0 in GUI results in CLI param error

CLI error:
--gpu invalid int value: 'undefined'

when custom GPU is set via GUI.

can I train the model by myself?

Easy of use - enhance check if checkpoints is there - allow for download and install.

GUI: check if the cli\checkpoints folder with content is present and if not show an option to download it and automatically place the checkpoints folder and contents in correct location.

Equal for CLI; check if the cli\checkpoints folder with content is present and if not show an option to download it and automatically place the checkpoints folder and contents in correct location.

Transformation #1 failed

The process has been interrupted by an error. This can be caused by:

• A corrupt installation
• Insufficient RAM. Buy more RAM! (8 GB recommended)
• If you are using GPU: The graphics card was not found, it is not compatible (minimum: 3.5 CUDA compute capability) or the drivers are outdated.
• Check the console for more information.

For help, please report the following to one of our developers:

https://rollbar.com/occurrence/uuid/?uuid=479bb768-0145-4cc8-a6d2-f2644f83e016

console ：
[2608] Failed to execute script main
Traceback (most recent call last):
File "main.py", line 348, in
File "main.py", line 243, in main
File "run.py", line 146, in process
File "gan.py", line 85, in initialize
File "gan.py", line 172, in __define_G
File "site-packages\torch\nn\modules\module.py", line 265, in cuda
File "site-packages\torch\nn\modules\module.py", line 193, in _apply
File "site-packages\torch\nn\modules\module.py", line 193, in _apply
File "site-packages\torch\nn\modules\module.py", line 199, in apply
File "site-packages\torch\nn\modules\module.py", line 265, in
File "site-packages\torch\cuda_init.py", line 162, in lazy_init
File "site-packages\torch\cuda_init.py", line 91, in _check_driver
AssertionError:
The NVIDIA driver on your system is too old (found version 9020).
Please update your GPU driver by downloading and installing a new
version from the URL: http://www.nvidia.com/Download/index.aspx
Alternatively, go to: https://pytorch.org to install
a PyTorch version that has been compiled with your version
of the CUDA driver.
Executing phase: correct_to_mask
Executing phase: dress_to_correct
GPU IDs: [0]
[8532] Failed to execute script main
Traceback (most recent call last):
File "main.py", line 348, in
File "main.py", line 243, in main
File "run.py", line 146, in process
File "gan.py", line 85, in initialize
File "gan.py", line 172, in __define_G
File "site-packages\torch\nn\modules\module.py", line 265, in cuda
File "site-packages\torch\nn\modules\module.py", line 193, in _apply
File "site-packages\torch\nn\modules\module.py", line 193, in _apply
File "site-packages\torch\nn\modules\module.py", line 199, in apply
File "site-packages\torch\nn\modules\module.py", line 265, in
File "site-packages\torch\cuda_init.py", line 162, in lazy_init
File "site-packages\torch\cuda_init.py", line 91, in _check_driver
AssertionError:
The NVIDIA driver on your system is too old (found version 9020).
Please update your GPU driver by downloading and installing a new
version from the URL: http://www.nvidia.com/Download/index.aspx
Alternatively, go to: https://pytorch.org to install
a PyTorch version that has been compiled with your version
of the CUDA driver.
Executing phase: correct_to_mask
Executing phase: dress_to_correct
GPU IDs: [0]
[19692] Failed to execute script main
Traceback (most recent call last):
File "main.py", line 348, in
File "main.py", line 243, in main
File "run.py", line 146, in process
File "gan.py", line 85, in initialize
File "gan.py", line 172, in __define_G
File "site-packages\torch\nn\modules\module.py", line 265, in cuda
File "site-packages\torch\nn\modules\module.py", line 193, in _apply
File "site-packages\torch\nn\modules\module.py", line 193, in _apply
File "site-packages\torch\nn\modules\module.py", line 199, in apply
File "site-packages\torch\nn\modules\module.py", line 265, in
File "site-packages\torch\cuda_init.py", line 162, in lazy_init
File "site-packages\torch\cuda_init.py", line 91, in _check_driver
AssertionError:
The NVIDIA driver on your system is too old (found version 9020).
Please update your GPU driver by downloading and installing a new
version from the URL: http://www.nvidia.com/Download/index.aspx
Alternatively, go to: https://pytorch.org to install
a PyTorch version that has been compiled with your version
of the CUDA driver.
Executing phase: correct_to_mask
Executing phase: dress_to_correct
GPU IDs: [0]

sorry Im Chinese Not very English, please help me. thank you。

Arch Linux additional steps for Ubuntu binaries

at first 'chrome-sandbox' may not be present in gui directory; it will appear after running CLI once.
from main directory

1. cd cli
2. sudo chmod +x ./cli
3. ./cli --help
4. cd ../gui
5. sudo chown -R root:root chrome-sandbox
6. sudo chmod +4755 chrome-sandbox
7. chmod +x DreamTime

now the GUI should work.

Diacritic letters in the filename generate white image

Nudify an image that contains diacritic letters in the filename (in the name of the file, such as ž or ý) it will generate just white image.

personal suggestion

I think you can add a function that can adjust the brightness, such as white or black breasts. Allow users to adjust the effect they want, more in line with the imagination。I hope the author can think about it. thanks

Windows 10 Pro X64 not 'N' version Media Pack detection

Windows 10 Pro X64 not 'N' version Media Pack detection fails, because the
HKLM\SOFTWARE\Microsoft\Windows\CurrentVersion\Setup\WindowsFeatures\WindowsMediaVersion
is not present, had 2 users experience this issue; both on Windows 10 Pro.
a [Find] in the registry editor did not return a location for any key with 'WindowsMediaVersion'
while the Windows Media features are on the OS.

HKLM\SOFTWARE\Microsoft\Windows\CurrentVersion\Setup
was not present, maybe this has to do with the way Windows 10 was installed?

GUI cropper uses black colored padding, that the algorithm takes into account to produce results

Use transparent or white padding on cropping to have no effect on the outcome.

After install Release 1.0.0 nothing happens. (win7)

With Release 1.0.0 nothing happens. (win7) The program use 47k memories but I don't see the apps.

pixeled parts

don´t know if deleted or if I just didn´t post it.

There are parts that are pixeled somehow.
https://imgur.com/a/QNT0ob0

CUDA 9 Support in v0.13

Please add a support of CUDA 9 for older NVIDIA videoboards, cause CPU rendering doesn't work good at all, causing an evil freezes of whole computer.

Ease of Use - Magnet link for checkpoints

In case people want to torrent, instead of mega/mediafire the checkpoints, I have set up a seederbox for the checkpoints.zip that the aforementioned sites provided.

If you like, you can add to the documentation to grab them from here:
magnet:?xt=urn:btih:5TQD3H3JNM2AV3RD74D7WC6EJMO5GS3F&dn=checkpoints.zip&xl=2032833191&tr=udp%3A%2F%2Ftracker.opentrackr.org%3A1337%2Fannounce

this is seeded 24/7 via seederbox, and during the week by my work PC (I work in IT. Who watches the watchmen? No one)

Put Mac Compatibility Status on Readme

When dreamtime compatible with macOS? :(

Load Image from URL

Request from the last issue tracker: The ability to load a file from a web url. This seems pretty trivial. I think we could do something like web req => temp folder => normal process and expose a textbox or something.