MOSS-Image is up on Moescape AI. It's an open model by OpenMOSS that reads booru tags, plain sentences or layout JSON, and it can draw the text you put in quotes. English and Chinese work best.

What's in the slides: • Text on signs, posters and speech bubbles • Characters from just their name and series tags • Building a prompt in three layers: tags, quality tags, then scene and style • Why artist tags don't work, and what to write instead • Framing tags that keep your character big in the shot • Placing things with layout JSON (experimental) • Using community finetunes and uploading your own MOSS checkpoints and LoRAs

Defaults are 25 steps and CFG 4 for 15 credits. 50 steps gives cleaner hands for 30 credits.