eps696/aphantasia

Parameterizes image generation using FFT, DWT (wavelets), or direct RGB optimization—avoiding GANs entirely—enabling high-resolution outputs (fullHD/4K+) with stable, controllable synthesis. Supports multi-modal queries combining text prompts, image references, style descriptions, and negative prompts with weighted syntax, plus continuous video generation via frame interpolation with optional depth-based 3D effects. Integrates with multiple CLIP vision models (ViT and ResNet variants) and includes experimental aesthetic loss and progressive learning rate strategies for compositional control.

789 stars. No commits in the last 6 months.