The larger models truncate to 2048 dims and don't improve results
over the native 1.7B that PixelGen was trained with.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Pixel-space diffusion text-to-image generation with REPA LoRA training.
6 nodes: Model Loader, Text Encoder Loader, Generate, Training Config, Train, LoRA Loader.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>