makegirlsmoe / makegirlsmoe_web

Create Anime Characters with MakeGirlsMoe

MakeGirlsMoe-Web

This repo contains the front-end part of make.girls.moe.

Running on local machine

You need git-lfs to clone this repo.

1. Modify src/Config.js and set var debug=true;
2. npm install
3. npm start

License

MakeGirlsMoe is freely available only for non-commercial use. Please, see the license for further details. For commercial queries, contact Yanghua Jin. The code of web interface is under the GPL v3.0 license and can be redistributed. All the compiled model files are privately owned by Yanghua Jin and are not allowed for unauthorized commercial use.

makegirlsmoe_web's Issues

噗哈哈哈哈

https://twitter.com/kaKyoKsis/status/898986015997558784/photo/1?ref_src=twsrc%5Etfw&ref_url=http%3A%2F%2Fmake.girls.moe%2F

https://twitter.com/kamixadze/status/898987011318480896/photo/1

---

PS: I don't mind if you change the issue title.

The Training dataset

Is it possible to share the 31255 dataset?
31255 128p images will be not that big. It is very easy to upload it to Internet.
In fact we have some unique GAN training tricks and would like to help to improve the predicted image quality.
It will not be offensive to some illustration websites' lisense, if we only share the clipped 128p face images.In fact many other datasets is already shared without any lisense problems, including nico-opendata.
The dataset in your paper is described as prepared very carefully, so many reseachers are interested in it. And we do not want to bother to collect these and clip these and vectorlize these again and again.
Could you please do these reseachers a favor :)
"We hope our work would stimulate more studies on generative modeling of anime-style images."
And the dataset can helps a lot.

Generate anime girl based on real image

I'm not an expert in GAN but I heard about encoder/decoder and the fact that picture can be generated by feeding an other picture to the encoder then decoding the resulting vector.
So is it possible to add a feature to generate an image based on an image provided by the user ?

SVG and Config export

It is a really nice project.
Is it possible to export the current noise and settings and can be tweaked later on?

Whats your thought on adding SVG support as well.

Only half of the picture is generated

Console shows:
Failed to initialize webgpu backend: Error: WebGPU backend is not supported in this browser.
But half of the image is still generated.

I have an error while using it locally. Network Error

I have an error while using it locally. Network Error
Unable to load the model, the node I am using is v8.10.0

store data from generated images to future model training

For example:

• Is it a boy or a girl?
  
• Is it a teenager, a girl or an adult?
  
  
• face position.
  and if possible share it

😭 What's this?

How to increase the resolution from 128*128 to 256*256？

Thanks for this project.

In the original paper, it was mentioned that 'scaling all training images to a resolution of 128*128 pixels'. But the current website can produce 256*256 pictures.

How did you achieve this improvement? Better data set, image super-resolution or other methods?

Looking forward to your answers. Thanks in advance.

Error: WebGPU is not supported on this browser

My browser is Chrome 60.0.3112.78.
I tested the browser, including Firefox and Edge.
It shows:

Releasing the Chainer model

Hello, do you plan to release the trained models in chainer? Technically, we can access models in web_dnn format, but having the chainer model would be more convenient.

Internal Server Error when vote

Request on http://make.girls.moe/api/stat/rate reponse with 500 error

{"message":"could not find host matching read preference { mode: "primary", tags: [ {} ] } for set data1","error":{"code":133,"index":0,"errmsg":"could not find host matching read preference { mode: "primary", tags: [ {} ] } for set data1"

It's 2017, give us ethnicity options

Can we get the ability to create black girls?

Clarification about the generator archetecture

Hi,

Thanks for your contribution. I am just curious about the architecture. In the original SRResNet, after the residual blocks, there was CONV, BN, ELEMENTWISE-SUM.

However, in your paper you seems to use CONV, BN, ReLU, ELEMENTWISE-SUM. Is there any reason why ReLU is inserted?

BTW, https://arxiv.org/pdf/1708.05509.pdf shows BN, ReLU, ELEMENTWISE-SUM, which seems to be a mistake.

decide color of clothes

No glasses

Looking for cooperation.

Several days ago I found an interesting phenomenon that images generated by makeGirlsMoe are always good style maps for style2paints. After several experiments I am confident that it is not my personal subjective feeling but objective fact. Maybe it is caused from your latent distribution from illustration2vec.

Defected background when generating with WebGPU 使用WebGPU生成图片背景异常

While generating a image with WebGPU, there are large chunks of black space in the background. Provided here are the images generated on Chrome (normal) vs. generated with Safari Teh Preview WebGPU (defective), as well as the noise.
I am using MacBook Pro (Retina, 15-inch, Mid 2014) with NVIDIA GeForce GT 750M.

用WebGPU生成图片时背景会有大块的黑色。上面的图片是chrome生成的（正常）和用WebGPU生成的（不正常）。噪音也已经给出。
我是用的是MacBook Pro (Retina, 15-inch, Mid 2014)。显卡是 NVIDIA GeForce GT 750M。

Multiple-Language Support

The new version should come with multiple-language since we need to add an EULA for new version.

Can I train the model myself?

Can I train the model myself? If yes, what should I do? I have almost no experience.

population centre

<--dispatch by community-->

Show the real deal

how its exactly done any description or guide will be good

High resolution image

Please be able to export images with high resolution.

[Idea]Guideline photo Generate

hope someday this idea will become real ;)

this is the picture i generate with the tool

in this case i mean the picture was create but will be cool to see the tool with export function that when the photo will be export he will create 2 files one is the photo himself and another file is a guideline photo i think this is cool idea for beginners who want learn about anime draw style like me ;)

Strange result from the noise seed.

Model training scripts code

Hi. I liked MakeGirlsMoe demo very much, thank you for great work. Repo description states that you decided to open-source model training scripts after the release of Chainer 3.0. Looks like it has been released (https://groups.google.com/forum/#!topic/chainer/Raw_ms3jHcs). Do you still have plans to open-source training scripts?
Thanks.

A simple way to save 99% network data flowing and 3x speed up the generating with much more face quality

step 1. generate about 1x10^7 face images with your tesla in about 24 hours on monday every week.
step 2. delete all the generated images on sunday every week.
step 3. do binary search each time your user requires a image and send them the 128p result in jpg version.

Because the AI do not need user to upload their sketch or highly Individualization like paintschainer or styletransfer, you can do as follow.

step 4. You can use discriminator to give "score" to the mentioned 1x10^7 face images when generating them in step 1 and sort this score in step 3 to give your user the best result.

how does this model being trained?

can you share?

Hairstyle does not reflect on generated image

For example I choose ponytail but I get either short hair or long hair most of the time or hard to see that they are ponytail

Please re-compile your model to support iOS 11

Yesterday, we updated WebDNN to support iOS11, new iOS which will be released next week.

iOS11 supports WebGPU, but some hardware-oriented constriction is different from macOS and re-compiling your model is needed.
Could you re-compile your model to support iOS11? it is not needed to modify your JavaScript frontend code.

Provide up-to-date links to the data set

Following this paper, we tried to obtain the data set through a sql query to http://erogamescape.dyndns.org/~ap2/ero/toukei_kaiseki/sql_for_erogamer_form.php. Unfortunately, our requests all seem to time out. Has the link moved? Is there a way to get step-by-step instructions on how to obtain the data set?

The 'hat on' option mostly didn't add hats from what I tried

Maybe training DS has too few images with hats

Keep waifu the same, randomize pose and/or gesture

Once I settle for my perfect waifu, can I somehow tell the generator to leave her design unchanged, and only change the parameters related with pose and/or gesture?

Need more data packs

When I used git-lfs clone [email protected]:makegirlsmoe/makegirlsmoe_web.git I got following error:

batch response: This repository is over its data quota. Purchase more data packs to restore access. 
error: failed to fetch some objects from 'https://github.com/makegirlsmoe/makegirlsmoe_web.git/info/lfs'

So I was wondering should we donate or how could I make the git-lfs clone work? Thank you!

Look at your creation

You are the ones that released this monster to the world...

“Your scientists were so preoccupied with whether they could, they didn’t stop to think if they should.”

WebGPU on iOS 11

Safari on iOS 11 supports WebGPU, but we find that output results stay the same no matter what the input is.

Add precommit hooks using husky to run npm test before each commit

This would make sure that all tests pass before each commit.

Feature Suggestion - Add survey

Let's gather information about the satisfaction of the result. A simple Good / Bad choice for each generated image will provide good information.

Display driver crash when using WebGL

If the WebGL option is enabled, after clicking the "Generate" button, I get the error "Display driver stopped responding and has recovered". If WebGL is disabled, generation is successful.
Depending on the browser, the following happens after the crash:

• Google Crome - freezes;
• Firefox - continues working, the site displays a message asking to disable the WebGL option.

System Configuration:

• Windows 7 Professional, ver 6.1 (build 7601: Service Pack 1), x64
• CPU: Intel Core 2 Quad Q9550
• RAM: 6 GB
• GPU: Nvidia GeForce 9800 GTX+, 512 MB
• Video Driver: NVIDIA version 342.01 (for my video card this is the latest driver)
• Google Chrome 86.0.4240.111, (64 bit)
• Firefox 81.0.2 (64 bit)

Please advise if additional information is required.

Options on the site:

Error Message:

bodys

maybe can be option to add body generate too i can help with money if needed

文字列でノイズをインポート/エクスポートする機能が欲しい

画像より文字列の方が共有が便利な事もあるので

make result image not scroll off page

There's no content below the result image (besides a few buttons), and it would be nice if that was prevented from being scrolled past (along with the page header?). In Expert Mode, it's easy to have enough options open that accessing all the options means you can't even see the result image and have to scroll back to change it.

Feature Request: Tweet a facial image with a noise image

Thank you for your great app!
When I see cute girl images that other users generated and shared on Twitter, I would like to get their noises and reproduce them.
How about adding a noise image to a tweet by default when sharing a facial image on Twitter?

Sanity down results: problem of model or program?

For a quick example:

Using 256x256 ver.171125
White twin tail, purple eyes. All off but hats on. style to the right.

That's just an easy condition for sanity down result. Actually, any random-by-human conditions will lead to such result. Also, setting hats option is really bad. That should be reset by default.

My guess is that the problem is the lack of such type of girls in the training data.
Anyway, though the problem really sucks, the project seems good.

Allow to save/load random noises.

I am trying to generate good images.

As you've mentioned in tips, a good random noise is important for the better generation.
However, a random noise is reset when reloading or closing the browser tab.
There is no way to recall it.

So I would like to save/load random noises.
It would be very useful when reproducing the good images found before.

How about make boys

Do you have the training data of boys? If adding an gender option, I believe more girls will visit your website~
Just a suggesstion, please ignore me if not intersted~