🖍️ AlignDRAW

One of the first text-to-image models (Mansimov et al., 2015) — resurrected. A recurrent-attention VAE that draws your caption onto a 32×32 canvas over 32 steps. The original weights were lost; this was re-trained from scratch on the original COCO data. Output is deliberately blurry — that's authentic 2015. Weights: mediasynthesismuseum/aligndraw.