User explores Ideogram V4 LoRA training: fast, 13GB VRAM

A user reports training LoRAs on Ideogram V4 is fast using ~13GB VRAM. The model supports bboxes and .json prompting, making generation a joy.