Reupload of the Image to Vroid Studio Asset generator. Image to 3D asset. VRass v1.2, Img2Ass. Original filename is VRass1.2_Img2ass_byAbysz.ckpt. Archived Tutorial:
This model understands more visual patterns, but in return is stubborn as an underworld mule. To obtain incredible results, I leave you a small guide.
Prepare your ASS. Remove anything unnecessary put into, to make the job easier. The cleaner the better.
Put it in IMG2IMG, in any GUI of Stable Diffusion.
Use "vrass" as the initial word, and then the prompt (color, texture, artist, etc). The model hasn't learned styles, so be generous with your description, and feel free to be very aggressive in your use of weights, denoising, and CFG (Seriously). The model will resist using your ASS, but you must force him! Don't worry, it's worth every second.
Scale the ASS x4 and use it at will.
Import and enjoy!
Archived Notes:
Put vrass at the beginning of the prompt.
Use denoising between 0.7-0.8 and play with CFG.
Choose a clear description of the style and texture of clothing you want.
Make sure to adjust the aspect ratio. For example T-Shirt asset is 2:4
This model handles high levels of denoising well without "going too crazy". I would say at 0.8+ it starts to lose the pattern. More than enough.
Obviously, the result loses the transparency of the png, so you'll have to deal with that, but... hey! these are free and unlimited assets.
WARNING: This process, if successful, can lead to addiction and social isolation. Use at your own risk.
Images of Raw Results:
Reddit posts before they were deleted, either by the original uploader or subreddit moderators:
https://www.reddit.com/r/VRoid/comments/z0c9f4/ai_cloth_generator/
https://www.reddit.com/r/VRoid/comments/z2n1h0/vrass_free_ai_cloth_generator_update/
https://www.reddit.com/r/StableDiffusion/comments/z09inh/img2ass_3d_cloth_generator/
Initially uploaded at HuggingFace before the links became unavailable a few months ago. Unfortunately v1.0 and v1.1 are not available for download, although the HuggingFace links are available on the Wayback Machine.
Description
v1.2: Stable Diffusion model patiently trained to understand the concept of "asset", specifically the one used for clothing in VRoid. Through IMG2IMG it can understand and modify the vast majority of clothing assets (UV map based). The imagination is the limit. The potential uses, needless to say, are enormous. The model works with "vrass" token. The more description, the better. It is very stiff, so weights, denoising, and CFG should be used liberally. It resists, but it works... and boy does it. Go wild!
v1.1: Model based on AnythingV3, trained with selected (Free) VRoid clothing assets. Any standard VRoid image asset (or others) can be used in IMG2IMG. This model recognizes the pattern and will replace the textures instead of going crazy. Supports high denoising strength and wide versatility (0.6-0.8 recommended). Basically, a free Img2Ass. Just use "vrass" token, and go crazy!
v1.0: Model based on AnythingV3, trained with many clothing assets. Any standard VRoid image asset (or others) can be used in IMG2IMG. This model recognizes the pattern and will replace the textures instead of going crazy. Supports high denoising strength and wide versatility (0.6-0.8 recommended). Basically, a free Img2Ass. Just use "vrass" token, and go crazy!
FAQ
Comments (18)
Thanks for this but you should release a safetensor version
Unfortunately I have no idea on how to perform that task.
All that I can do for now is to reupload the original cpkt file that was shared on HuggingFace in November 2021 by Ne_Nel on Reddit before the links were deleted.
If you are able to make a new model, use the Standard Vroid Asset files and train the model with updated and existing models such as Anythingv4.
@Abysz It's pretty straightforward with this tool https://github.com/diStyApps/Safe-and-Stable-Ckpt2Safetensors-Conversion-Tool-GUI
You don't need to retrain it or do anything like that.
@Raphanon Okay, I've let it process in the mean time, but I don't know how long it will take because I have a low powered system. I was also not able to test the model on my PC because of the high system and GPU requirements that Stable Diffusion needs to run the CPKT file locally.
I discovered that the developer for the CKPT to Safetenors Converter, diStyApps, had made a SD Pickle Scanner, and the CPKT file that I have on my PC passed their test.
I also scanned the file with Windows Defender and there were no detected viruses or trojans.
For other questions, it's better to ask Ne_Nel on Reddit because they have more experience than me when it comes to Stable Diffusion. They might also have the original files if they were in contact with the original creator of this data model. https://www.reddit.com/user/Ne_Nel
"img2ass" you have my attention!
Can you share your training data and some training info? It would be great to learn from and do more like this!
It's better to ask Ne_Nel on Reddit because they have more experience than me when it comes to Stable Diffusion, because they were active on the Reddit posts. They might also have the original files, training data and training info if they were in contact with the original creator of this data model. https://www.reddit.com/user/Ne_Nel
I was lucky to have downloaded v1.2 of this data model before the Reddit posts and HuggingFace uploads were deleted.
If it's to get the official and default Vroid textures, a user on Booth uploaded the entire collection that is installed with Vroid Studio. (For any other information, I wouldn't know, so check with Ne_Nel on Reddit):
https:// booth.pm/en/items/2907431
VRoid All Female and Male Default Textures.
The page description:
All female and male default textures created by VRoid Studio. Download for easy access to any default texture. Simply download the .zip file, extract it anywhere on your pc, and start using the textures in it.
What is included?
Folder containing all default VRoid Studio female and male textures.
How do I use it?
- Open VRoid Studio and go to the place you want to import, (for example, face > irises) right click on a "layer," select "import," go to this folder and click the image you want, finally click "ok." - Edit in external drawing programs by opening the drawing application, under "file" choose "open," then go to this folder and click the texture you want to edit, finally click "ok."
I have been looking everywhere for this. Thank you.
you're welcome.
Can this be done for the model skin and face textures??
I'm also not sure, because the original uploader said in the Reddit posts that they used standard Vroid Studio clothes and textures when making the data model. There also aren't many skin and face textures that were released by the Vroid team, so that may not be enough information to make a fresh data model.
For other questions, it's better to ask Ne_Nel on Reddit because they have more experience than me when it comes to Stable Diffusion, and they are very active in many AI/ML subreddits: https://www.reddit.com/user/Ne_Nel
@Abysz I've created skins for vroid models, I also have a few downloaded from booth, but I don't know much about training sd models. Do you think 30-50 skin textures would be enough to train a model to a similar effect?
@CHARZULU I've seen that people are now making LoRas in addition to the switch to the safetensors file format, so that could help if you have a sample size like that.
If you used some downloads from Booth, make sure that the artists are well aware that you are using their resources for this project. There is still the discussion of ethics and legal uses on art for AI research and machine learning because artists feel like they aren't in control of their artwork when it is uploaded and displayed online by them for everyone to view.
I am also new to Stable Diffusion and to the whole idea of art as a resource in machine learning so I wouldn't know what you need for training a model. If there are Discord servers that focus on creating data models and the tech behind it, then you can ask for more questions in those servers if Ne_Nel is temporarily unavailable on Reddit.
I already used SD for that on a face texture and it can work but it is often not perfect. I also wish there would be a lora or so for that. Trained for it the results would be much better. But I wish there would be also one for the skin.
THANK YOU!
need update
thanks for the feedback, however Ne_Nel was the original uploader on Reddit before the links were taken down by the Reddit sub moderators. (I am glad to have downloaded the ckpt file). I do not have access to any hardware that is focused on AI/ML systems, so it is better to ask Ne_Nel or other users on Civitai and Huggingface.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.