A text-to-image diffusion model built from scratch in PyTorch — trained on the Stanford Cars dataset with BLIP captions. Includes a custom BPE text tokenizer/encoder, a conditional U-Net denoiser, a cosine noise schedule, and a separate super-resolution upscaler that takes the 64×64 diffusion output up to 256×256.
Terminator1321/TTI-Diffusion: A text-to-image diffusion model built from scratch in PyTorch — trained on the Stanford Cars dataset with BLIP captions. Includes a custom BPE text tokenizer/encoder, a conditional U-Net denoiser, a cosine noise schedule, and a separate super-resolution upscaler that takes the 64×64 diffusion output up to 256×256.
