danbooru2021/256px guided-diffusion checkpoint
This is an unconditioned 256x256x3 guided-diffusion checkpoint trained with 4.8M images from the danbooru2021 dataset for about 22 epochs.
Sampling
Run image_sample.py from OpenAI's guided-diffusion repo or plug it into Disco Diffusion if you wish to diffuse with CLIP guidance.
python3 scripts/image_sample.py --num_samples 128 --timestep_respacing ddim250 --model_path ema_0.9999_940000.pt --attention_resolutions 32,16,8 --class_cond False --diffusion_steps 1000 --image_size 256 --learn_sigma True --noise_schedule linear --num_channels 256 --num_head_channels 64 --num_res_blocks 2 --resblock_updown True --use_fp16 True --use_scale_shift_norm True --use_ddim True
Hardware
- 8xNVIDIA A100 80GB
Credits
- Sta, AI Novelist Dev (https://ai-novel.com/) @ Bit192, Inc.
License
MIT License