SimpleDiffusion is a basic diffusion model trained on approximately 150,000 image-text pairs from image forums across the internet.
This model was trained on about 30 epochs over a two month period. I don't have access to amazing gpus as I'm just one person, but the model shows promise if trained further.
The model utilizes the SDXL vae for compression and laion/CLIP-ViT-bigG-14-laion2B-39B-b160k for text latent creation.
The model is roughly 831 million parameters, and is trained at 256x256.
If you'd like to continue training the model or test the model, the latest train is on huggingface.