SD3.5 NIM deployments and partnerships

last edited

SD3.5 remains a workhorse open model for local generation, fine-tuning, and commercial deployment. NVIDIA NIM packaging and corporate partnerships broaden access, while SDXL checkpoints sustain an active image ecosystem. Related activity includes workflow creators, diffusion transformer research, and adoption across Southeast Asia.

Stable Diffusion 3.5

Stable Diffusion 3.5 uses its developer’s MMDiT rectified-flow transformer architecture. The largest model has 8.1B parameters. The family also includes an accelerated variant offering 4-step generation and a mid-sized variant. T5-XXL serves as the standard text encoder this year models.

The open-weight family has received 9/10 reviews. Its developer remains private after closing a fresh funding round in August 2026. Partnerships include WPP for creative transformation, EA for 3D assets, and a music-industry partner for generated music. The company also joined an industry coalition combating child exploitation.

NVIDIA NIM and cloud deployments

The 2026 NVIDIA NIM microservice packages SD3.5 with an inference toolkit. FP8 optimizations provide faster generation and reduced VRAM use on RTX GPUs. The largest model is also available through a cloud model catalog, supporting generation at resolutions up to 1 megapixel.

Deployment choices include flagship and accelerated variants, while the mid-sized release is described as forthcoming. The lineup addresses image quality, generation speed, hardware requirements, and production integration.

SDXL ecosystem

SDXL remains widely used through SDXL 1.0, SDXL Turbo, an accelerated release, and a refinement model. Community fine-tunes include photorealistic v9 and v10 releases alongside other realism-focused models. A model-sharing platform hosts SDXL adapters, checkpoints, and embeddings.

Coverage includes Stable Diffusion 3.5 Reception and Commercial Use Cases (Stable Diffusion - 2026). It also covers open-source and specialized models. This page forms part of AI Art Models — Organized Reference.

Sources

The Runpod overview covers SD3.5 deployment through NVIDIA NIM and a cloud model catalog. Magai’s model summary describes the 8B-parameter flagship, 1-megapixel output, and family variants.

CEO Views discusses the developer’s private status and partnerships this year.