Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
stable diffusion training
| Date | Stars |
|---|---|
| 2026-07-31 | 295 |
| 2026-08-02 | 295 |
| 2026-08-06 | 295 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# GLID-3-XL-stable GLID-3-xl-stable is [stable diffusion](https://github.com/CompVis/stable-diffusion) back-ported to the OpenAI guided diffusion codebase, for easier development and training. features: [inpainting/outpainting](https://github.com/Jack000/glid-3-xl-stable/wiki/Custom-inpainting-model) [classifier guided stable diffusion](https://github.com/Jack000/glid-3-xl-stable/wiki/Classifier-guided-stable-diffusion) [super-resolution](https://github.com/Jack000/glid-3-xl-stable/wiki/Double-diffusion-for-more-detailed-upscaling) # Install First install [latent diffusion](https://github.com/CompVis/latent-diffusion) ``` # then git clone https://github.com/Jack000/glid-3-xl-stable cd glid-3-xl-stable pip install -e . # install mpi and mpi4py for training sudo apt install libopenmpi-dev pip install mpi4py ``` # Get model files from stable diffusion ``` # split checkpoint python split.py sd-v1-4.ckpt # you should now have diffusion.pt and kl.pt # alternatively wget -O diffusion.pt https://huggingface.co/Jack000/glid-3-xl-stable/resolve/main/default/diffusion-1.4.pt wget -O kl.pt https://huggingface.co/Jack000/glid-3-xl-stable/resolve/main/default/kl-1.4.pt ``` # Generating images note: best results at 512x512 image size ``` python sample.py --model_path diffusion.pt --batch_size 3 --num_batches 3 --text "a cyberpunk girl with a scifi neuralink device on her head" # sample with an init image python sample.py --init_image picture.jpg --skip_timesteps 20 --model_path diffusion.pt --batch_size 3 --num_batches 3 --text "a cyberpunk girl with a scifi neuralink device on her head" # generated images saved to ./output/ # generated image embeddings saved to ./output_npy/ as npy files ``` # Inpainting A custom inpainting/outpainting model trained for an additional 100k steps ``` # install PyQt5 if you want to use a gui, otherwise supply a mask file pip install PyQt5 # download inpaint model wget -O inpaint.pt https://huggingface.co/Jack000/glid-3-xl-stable/resolve/main/inpaint/ema_0.9999_100000.pt # inpaint with gui python sample.py --model_path inpaint.pt --edit your-image.png --text "your prompt here" # the previously painted mask is saved as mask.png python sample.py --model_path inpaint.pt --edit your-image.png --text "your prompt here" --mask mask.png # partial inpaint by skipping timesteps python sample.py --model_path inpaint.pt --edit your-image.png --text "your prompt here" --skip_timesteps 20 # outpaint extends the canvas # --outpaint options: expand, wider, taller, left, top, right, bottom python sample.py --model_path inpaint.pt --edit your-image.png --text "your prompt here" --outpaint wider ``` # Generate with classifier guidance note: best results with --ddim --steps 100 ``` # download photo classifier wget -O class-photo.pt https://huggingface.co/Jack000/glid-3-xl-stable/resolve/main/classifier_photo/model060000.pt # download art classifier wget -O class-art.pt https://huggingface.co/Jack000/glid-3-xl-stable/resolve/main/classifier_art/model110000.pt # download anime classifier wget -O class-anime.pt https://huggingface.co/Jack000/glid-3-xl-stable/resolve/main/classifier_anime/model070000.pt # generate python sample.py --ddim --steps 100 --classifier_scale 100 --classifier class-anime.pt --model_path diffusion.pt --text "anime Elon Musk" ``` # Upscaling note: best results at 512x512 input and 1024x1024 output (default settings) 11gb vram required ``` # download model wget -O upscale.pt https://huggingface.co/Jack000/glid-3-xl-stable/resolve/main/super_lg/ema_0.999_040000.pt python super_large.py --image img.png --skip_timesteps 1 ``` # Training/Fine tuning Train with same flags as guided diffusion. Data directory should contain image and text files with the same name (image1.png image1.txt) ``` # minimum 48gb vram to train # batch sizes are per-gpu, not total MODEL_FLAGS="--actual_image_size 512 --lr_warmup_steps 10000 --ema_rate 0.9999 --attention_resolutions 64,32,16 --class_cond Fal
Excerpt of 6,733 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:78daf3f5682ed727, desc:stable diffusion