Stable Diffusion Model (SD 3.5)
Stable Diffusion is Stability AI's family of open-weight text-to-image models. The current release, Stable Diffusion 3.5, launched on October 22, 2024 and now comes in three sizes you can download and run on your own GPU. Here is what each version is for, what the license allows, and what to use in LazyKiwi if you would rather not install anything.
Using it
How to use Stable Diffusion
LazyKiwi does not run Stable Diffusion. These are the routes Stability AI lists.
Run it locally
Download the SD 3.5 weights from Hugging Face and run them with Stability AI's inference code or a community UI. Medium fits most consumer GPUs; Large wants more VRAM.
Use it hosted
Stability AI offers the models through its own API, through cloud partners, and in Stable Assistant, its web app. Hosted use bills per image; check the provider for current pricing.
Want images now, no setup?
LazyKiwi's image generator runs open-weight Z-Image and Qwen Image, plus Nano Banana 2 and GPT Image 2, in the browser with free credits on sign-up.
What it is
What the Stable Diffusion model is built for
Stable Diffusion models generate images by starting from noise and removing it step by step under the guidance of your prompt. Their selling point has always been the open weights: you can run them locally, fine-tune them and build on them.
SD 3.5 Large
The most capable Stable Diffusion
8.1 billion parameters. Stability AI calls it the most powerful model in the family, with the best quality and prompt adherence, aimed at professional work at 1 megapixel.
- 8.1B parameters
- 1 megapixel output
- Released October 22, 2024
SD 3.5 Large Turbo
Large, distilled to four steps
A distilled version of 3.5 Large that generates in 4 steps, so it is considerably faster while keeping strong prompt adherence.
- 4 inference steps
- Distilled from 3.5 Large
- Released October 22, 2024
SD 3.5 Medium
Made for consumer GPUs
2.5 billion parameters on the MMDiT-X architecture, released October 29, 2024. It generates from 0.25 to 2 megapixels and, per Stability AI, needs 9.9 GB of VRAM excluding text encoders.
- 2.5B parameters
- 0.25 to 2 megapixels
- About 9.9 GB VRAM (without text encoders)
License
Free below $1M a year
SD 3.5 ships under the Stability AI Community License: free for research, non-commercial and commercial use for individuals and organisations with under US$1M in annual revenue. Above that you need an Enterprise license.
- Community License, free under $1M revenue
- Enterprise license above $1M
- Fine-tunes and derivatives covered
Specs
Stable Diffusion 3.5 specs at a glance
From Stability AI's launch post, model page and license page, checked October 18, 2026.
- Developer
- Stability AI
- Current release
- Stable Diffusion 3.5 (October 22, 2024; Medium October 29, 2024)
- Variants
- 3.5 Large, 3.5 Large Turbo, 3.5 Medium
- Parameters
- Large 8.1B; Medium 2.5B
- Inputs
- Text prompt
- Outputs
- Images; Large at 1 megapixel, Medium 0.25 to 2 megapixels
- Weights
- Open, on Hugging Face; inference code on GitHub
- License
- Stability AI Community License (free under US$1M annual revenue); Enterprise above
- Where to use
- Self-hosted, Stability AI API, cloud partners, Stable Assistant
- In LazyKiwi
- Not available; closest open-weight picks are Z-Image and Qwen Image
Versions
Stable Diffusion 3.5 Large vs Large Turbo vs Medium
How Stability AI positions each model, plus the older SDXL line it still lists.
| Model | Size | Best for | Speed |
|---|---|---|---|
| SD 3.5 Large | 8.1B | Highest quality and prompt adherence, professional use | Standard |
| SD 3.5 Large Turbo | Distilled from Large | Fast drafts with strong prompt adherence | 4 steps |
| SD 3.5 Medium | 2.5B | Consumer GPUs, customization and fine-tuning | Standard |
| SDXL | 3.5B | Older base with a large ecosystem of community fine-tunes | Standard |
| SDXL Turbo | Distilled SDXL | Real-time previews | As few as 1 step |
FAQ
Stable Diffusion FAQ
What is the latest Stable Diffusion model?
Stable Diffusion 3.5, released October 22, 2024 (Large and Large Turbo) with Medium following on October 29, 2024. As of October 18, 2026, Stability AI's image page still lists 3.5 Large as the most powerful model in the family.
Is Stable Diffusion free?
The weights are free to download. Under the Stability AI Community License, use is free for research, non-commercial and commercial work if you or your organisation make under US$1M a year; larger businesses need an Enterprise license.
Can I use Stable Diffusion images commercially?
Yes, under the Community License as long as your annual revenue is below US$1M. Above that, Stability AI requires an Enterprise license.
Which Stable Diffusion model should I use?
3.5 Large for the best quality, 3.5 Large Turbo when speed matters, and 3.5 Medium if you have a consumer GPU or want to fine-tune. SDXL remains popular because of its large library of community fine-tunes.
What GPU do I need for Stable Diffusion 3.5?
Stability AI says 3.5 Medium needs 9.9 GB of VRAM excluding text encoders, so most recent consumer cards can run it. 3.5 Large, at 8.1 billion parameters, needs considerably more.
Is Stable Diffusion open source?
The weights are openly downloadable and the inference code is on GitHub, but the license is Stability AI's Community License, not a standard open-source license, and it has a revenue threshold.
What does "diffusion" mean in Stable Diffusion?
The model learns to remove noise from images. To generate, it starts from random noise and denoises it step by step, guided by your text prompt, until a clear image appears.
Can I use Stable Diffusion in LazyKiwi?
No. LazyKiwi's image generator runs Z-Image and Qwen Image (both open-weight), Nano Banana 2, Seedream 5.0 and GPT Image 2 instead, in the browser with free credits on sign-up.
Open-weight image models, without the GPU
Stable Diffusion needs a download and a graphics card. LazyKiwi runs Z-Image and Qwen Image in the browser, free to start.
More models
Image models you can run in LazyKiwi today.