Diffusers 0.38.0 adds LLaDA2, NucleusMoE-Image, Ernie-Image pipelines and modular support

AI Impact Summary

Diffusers 0.38.0 introduces multiple new pipelines (LLaDA2, NucleusMoE-Image, Ernie-Image, LongCat-AudioDiT, Ace-Step 1.5) and adds modular pipeline support for LTX-2 and Hunyuan 1.5, signaling a shift toward more capable, sparse MoE and diffusion-based generation. These additions bring new APIs and models into production workflows, with potential performance and quality gains from MoE and diffusion-based synthesis, but also require validation of integration points, model loading, and hardware resource planning. Teams should assess migration steps for pipeline-specific code paths and plan benchmarks for latency, memory, and output fidelity across the new pipelines.