Hierarchical text-conditioned image generation now supports CLIP latent conditioning

AI Impact Summary

This change enables hierarchical image generation conditioned on CLIP latents, allowing multi-level control where global semantics from text prompts combine with CLIP-derived latent guidance. Technical teams will need to handle latent inputs, integration points with CLIP encoders, and the potential increase in compute and memory from additional encoding steps. Applications that previously relied on text-only prompts can achieve more precise outputs, but pipelines must adapt to generate or ingest CLIP latents to fully leverage the feature.