What's this about?
LLaDA-Image (52 upvotes): strong image generators with fully open training recipes — data, code and process published, in a field where open weights usually come with closed recipes.
What exactly is a training recipe?
Everything beyond the weights: datasets, preprocessing, schedules, tricks. Weights let you use a model; recipes let you reproduce, audit and improve one.
Why is this worth noting?
Image generation has been especially recipe-secret (data provenance is legally fraught). Full openness here is both a research resource and a transparency data point, same week as K2 Horizon's open-pipeline language models.