I’m pretty sure, but no expert on the matter, that correct text rendering was solved by feeding in bitmaps of rasterized fonts as supplemental context to the im

I’m pretty sure, but no expert on the matter, that correct text rendering was solved by feeding in bitmaps of rasterized fonts as supplemental context to the image generation models.