other
2.5B parameters
Commercial OK
Reviewed May 2026

Stable Diffusion 3.5 Medium

2.5B MMDiT-X with improved Querying Key Normalization and dual attention blocks at lower resolutions. Trained for 0.25-2MP output. Positioned as the mid-tier of the SD3.5 family, designed to run on consumer hardware while staying close to SD3.5 Large quality.

License: stabilityai-community·Context: 0 tokens
BLK · VERDICT

Our verdict

OP · Fredoline Eruo|VERIFIED MAY 29, 2026
unrated

The best license-friendly option for teams that don't want to touch FLUX. Quality clearly behind FLUX.1 [dev] but the Community License (with the $1M cap) is the most generous of any frontier-tier image model. Watch the revenue cap closely.

Overview

2.5B MMDiT-X with improved Querying Key Normalization and dual attention blocks at lower resolutions. Trained for 0.25-2MP output. Positioned as the mid-tier of the SD3.5 family, designed to run on consumer hardware while staying close to SD3.5 Large quality.

Strengths

  • Stability Community License: free for commercial use up to $1M annual revenue
  • MMDiT-X with QK-norm — markedly improved training stability and prompt adherence vs SD3 Medium
  • Trained for variable resolutions from 0.25 to 2 megapixels out of the box
  • Runs in ~9.9GB VRAM at FP16 — comfortable on a 12GB consumer card
  • Three text encoders (T5-XXL + 2x CLIP) — strong long-prompt comprehension

Weaknesses

  • $1M revenue cap in the Community License triggers an enterprise license requirement
  • Still trails FLUX.1 on typography, hands, and fine detail
  • T5-XXL text encoder adds memory pressure; offloading needed on 8GB cards
  • Smaller ecosystem than SDXL/FLUX — fewer LoRAs and ControlNets at launch

Quantization variants

Each quantization trades model quality for file size and VRAM. Q4_K_M is the most popular starting point.

QuantizationFile sizeVRAM required
Q4_K_M1.4 GB2 GB

Get the model

HuggingFace

Original weights

huggingface.co/stabilityai/stable-diffusion-3.5-medium

Source repository — direct quantization required.

Hardware that runs this

Cards with enough VRAM for at least one quantization of Stable Diffusion 3.5 Medium.

Compare alternatives

Models worth comparing

Same parameter band, plus what's one tier above and below — so you can decide what actually fits your hardware.

Frequently asked

What's the minimum VRAM to run Stable Diffusion 3.5 Medium?

2GB of VRAM is enough to run Stable Diffusion 3.5 Medium at the Q4_K_M quantization (file size 1.4 GB). Higher-quality quantizations need more.

Can I use Stable Diffusion 3.5 Medium commercially?

Yes — Stable Diffusion 3.5 Medium ships under the stabilityai-community, which permits commercial use. Always read the license text before deployment.

What's the context length of Stable Diffusion 3.5 Medium?

Stable Diffusion 3.5 Medium supports a context window of 0 tokens (about 0K).

Source: huggingface.co/stabilityai/stable-diffusion-3.5-medium

Reviewed by RunLocalAI Editorial. See our editorial policy for how we research and verify model claims.

Related — keep moving

Before you buy

Verify Stable Diffusion 3.5 Medium runs on your specific hardware before committing money.