Sorry, I know nothing about AI but I want to research it so that I can be more informed on it, and I didn’t really get what AnythingV5 meant
Anything V5 is the (latest/upgraded?) version of Anything V3. It’s a stable diffusion anime model both versions are made by the same author. Not to be confused with Anything V4.5 which is not the same author.
Sorry for late reply. Here’s how I did it and a small guide.
I used Stable Diffusion Automatic1111 Here’s how to install it locally if you have strong Nvidia GPU,
Here’s how you can do it online for free.. Same software used for creating images of famous people in weird situations.
Then download the model AnythingV5, it allows to create decent anime-like images. But there are also different models, with slightly different styles etc.
Put “.safetensors” file in “stable-diffusion-webui\models\Stable-diffusion”. After starting the app you can then select model on top (Stable Diffusion Checkpoint)
Paremeters are optional, but most important is to use same seed for images of the same person. Play with CFG scale (how much AI can get “creative”), and put Samplings Steps to ~40 and Sampling Method to “DPM++ SDE Karras” or “DPM++ 2M Karras”.
Last is prompts, they are just trail and error… like a lot of trail and error, small prompt change can make or break whole image. Here’s basic one I used:
- ((masterpiece, best quality)), 1girl, (blurry background), [skinny, obese:0.5] girl, red hair, standing, pretty face, orange eyes, full body
- negative prompts: (low quality, worst quality:1.4), (bad anatomy), (inaccurate limb:1.2),bad composition, inaccurate eyes, extra digit,fewer digits,(extra arms:1.2), (disfigured hands), (disfigured fingers)
Hands require another technique called “inpainting”. AI can be good at hands but only if it focuses specifically on them.
Upscale images for better resolution. (I was lazy and just used Waifu2x :P)
I’m still looking for good way how to make characters into ssbbw category (but there are specific VAE and LORA enhancer files trained for it). You can also train your own neural network on series of images.
Rest just play around, Google, and if something’s unclear just ask.
4 Likes
I’m gonna be honest, I’m really struggling with this colab thing lol its giving me an error every 3 steps
ok my computer cant physically handle the desktop version, so i’m trying to suffer through the web version lol. Got ONE image to generate, but then a minor connection error meant now everything below requirements is riddled with errors permanently??
I dont understand, its perfectly working and then suddenly stops, and now it doesn’t recognize the code it was JUST using
That’s problem with Colab - lately it started to close sessions more aggressively.
You just need to on Colab go to runtime → run all (you can’t just run last block).
Sometimes servers are at capacity and you’re just out of luck at that time.
But if it doesn’t finish execution even then, one solution is to copy source notebook once more, and delete the old one. Afterwards go to Google Drive, and back up pictures from sd/stable-diffusion-webui/output, and drag .safetensors files from sd/stable-diffusion-webui/models/Stable-diffusion so you don’t have to upload them again. Finally and remove the /sd folder and clean the bin. On colab enter your previous settings, and click first 3 play arrows. After running “Requirements” Put the .safetensors back. You should run rest of the blocks fine then, but if not click “run all”.
2 Likes