Stable Diffusion 3.5

Summary

Stable Diffusion 3.5 is Stability AI's flagship open-weight image generation family, released October 2024 — Large (8B), Large Turbo (8B/4-step), and Medium (2.5B) variants under the Stability AI Community License. Remains the foundational reference for the open-weight ecosystem of fine-tunes and tooling.

Overview

Stable Diffusion 3.5 is Stability AI's most recent major image generation model, released in October 2024 in three variants — Large (8B parameters), Large Turbo (8B parameters, optimized for 4-step generation), and Medium (2.5B parameters). The Medium variant has emerged as the deployment-preferred option for many commercial integrations through 2025–2026, balancing image quality against the GPU memory and inference cost realities of production deployment. As of May 2026, Stability AI has not released Stable Diffusion 4 — SD 3.5 remains the company's flagship image generation line despite intense competitive pressure from FLUX (Black Forest Labs), Midjourney V8.x, and frontier-lab integrated image generation models (Nano Banana 2, GPT-image).

While SD 3.5 has lost benchmark and aesthetic leadership to FLUX and Midjourney through 2025–2026, the model's role as the foundational reference for the open-weight image generation ecosystem remains largely undimmed. The vast majority of downstream open-weight image generation projects — fine-tunes, ControlNets, LoRAs, custom UIs (Automatic1111, ComfyUI), and creative workflow tools — assume SD as the substrate, preserving Stability AI's gravitational role even as the company's competitive position has eroded.

Specifications

  • Developer: Stability AI
  • Release Date: October 2024
  • Type: Open-weight text-to-image generation model family
  • Variants:
    • SD 3.5 Large — 8B parameters; highest quality
    • SD 3.5 Large Turbo — 8B parameters; optimized for 4-step generation (faster inference)
    • SD 3.5 Medium — 2.5B parameters; deployment-preferred for many commercial integrations
  • License: Stability AI Community License (open-weight with commercial-use terms; review for specific deployments)
  • Distribution: Hugging Face open-weight downloads + Stability AI Developer Platform + commercial license tiers

Capabilities

Open-Weight Distribution: SD 3.5 Large, Large Turbo, and Medium are all available as open weights — supporting self-hosting, fine-tuning, and the broader open-weight image generation ecosystem.

Three Variants for Different Tradeoffs:

  • Large (8B): Highest quality; for production-grade image generation where compute is available.
  • Large Turbo (8B, 4-step): Optimized for fast inference; useful where latency matters more than maximum image quality.
  • Medium (2.5B): Deployment-preferred for many commercial integrations; best quality-per-cost in the SD 3.5 family.

Ecosystem Centrality: Foundational reference model for the entire open-weight image generation ecosystem. The vast majority of LoRAs, ControlNets, fine-tunes, and custom UIs assume SD as the substrate.

Tooling Maturity: Substantial tooling ecosystem — Automatic1111, ComfyUI, InvokeAI, and many commercial UIs all support SD 3.5 natively. Substantially more mature tooling than newer open-weight image models.

Limitations

No SD 4 Yet (as of May 2026): Stability AI has not released SD 4 despite intense competitive pressure from FLUX 1.2 Pro Ultra, Midjourney V8.x, and frontier-lab integrated image generation. SD 3.5 remains the current flagship by default rather than by competitive choice.

Behind FLUX on Quality: FLUX 1.2 Pro Ultra (and the open-weight FLUX.1 [dev] / [schnell] variants) generally outperform SD 3.5 on raw image quality benchmarks. FLUX has effectively replaced SD as the new default reference model for high-quality image generation in many production workflows.

Behind Midjourney on Aesthetic Frontier: Midjourney V8.x retains a meaningful edge on the consumer aesthetic frontier — especially for editorial and creative use cases where Midjourney's distinctive visual styles drive adoption.

License Complexity: Stability AI Community License includes commercial-use terms beyond simple Apache 2.0 — enterprises should review carefully before redistribution or service-based deployment.

Frontier Lab Integrated Pressure: Frontier-lab image generation models (Nano Banana 2, GPT-image) are increasingly integrated into chat product surfaces with strong quality and cheap distribution — making standalone open-weight image generation a less obviously-needed deployment surface for many consumer use cases.

Recent Developments

  • No SD 4 Release (May 2026): Despite competitive pressure, Stability AI has not shipped SD 4. The company has expanded into video (SV4D 2.0), 3D, and real-time enterprise rendering instead of pursuing a pure image-quality next-generation.
  • Medium Variant Adoption: SD 3.5 Medium has emerged as the deployment-preferred variant for many commercial integrations through 2025–2026.
  • Ecosystem Position: SD remains the foundational reference for the open-weight image generation ecosystem, even as FLUX takes leadership in raw quality.
  • Competitive Pressure: FLUX 1.2 Pro Ultra (February 2026) and Midjourney V8.1 (April 30, 2026) define the current competitive frontier; SD 3.5 has held its position primarily through ecosystem centrality rather than benchmark leadership.

Last Updated

May 7, 2026

→ Back to Models