Stable Diffusion is one of the few AI image generators you can actually run on your own computer for free. No subscription, no per-image fees, no cloud dependency. You download the model weights, install the software, and generate as many images as your GPU can handle.
That said, the ecosystem has gotten messier over the past year. There are multiple model versions (SDXL, SD3, SD3.5), different licensing terms for each, and a cloud API if you don't want to deal with local setup. Figuring out what's actually free, what costs money, and what you're legally allowed to do with the output isn't straightforward anymore.
Which Version of Stable Diffusion Should You Use?
As of early 2026, there are three main versions floating around. Each has different quality, hardware requirements, and license terms:
| Version | Release | Quality | Min VRAM | License | Best For |
|---|---|---|---|---|---|
| SDXL 1.0 | 2023 | Good | 8GB | Open (commercial OK) | Most users, huge community |
| SD3 Medium | 2024 | Better | 10GB | Free under $1M revenue | Quality-focused work |
| SD3.5 Large | 2025 | Best | 12GB+ | Research + limited commercial | Cutting-edge quality |
For most people, SDXL is still the best choice. It has the largest community, the most LoRA models and fine-tunes available, and a fully permissive license that lets you use the output commercially without restrictions. SD3 looks better out of the box, but the licensing headaches make it complicated for anything beyond personal projects.
Hardware Reality Check
SDXL generates images at 1024x1024 by default. SD3.5 can go higher, but needs significantly more VRAM. If you're running a consumer GPU (RTX 3060/4060), SDXL is the practical choice.
What Does Stable Diffusion Cost?
This is where it gets confusing because there are two completely different ways to use it:
Option 1: Run it locally (free)
If you have a decent GPU (8GB+ VRAM), you can run Stable Diffusion locally through tools like Automatic1111, ComfyUI, or Forge. The model weights are free to download from Hugging Face. Your only cost is electricity and the hardware you already own.
This is genuinely free. No hidden fees, no credit system, no usage limits. You can generate thousands of images per day if your hardware keeps up. The catch is that setup takes some technical knowledge, and results depend heavily on your GPU. A RTX 4090 renders in seconds, a 3060 takes noticeably longer.
Option 2: Use the Stability AI API (paid)
If you don't want to deal with local setup or need to integrate image generation into an app, Stability AI offers a cloud API. Pricing is per-image based on complexity:
| Model | Cost Per Image | Resolution | Notes |
|---|---|---|---|
| SDXL via API | $0.002-0.006 | Up to 1024x1024 | Cheapest option |
| SD3 via API | ~$0.035 | Up to 1024x1024 | Higher quality, higher cost |
| SD3.5 via API | ~$0.04-0.065 | Up to 1536x1536 | Best quality |
New accounts get 25-200 free credits to start with, which is enough for 100-200 basic images. After that, credits cost $10 per 1,000. For most casual users, the free credits are enough to decide if you want to run things locally instead.
The License Situation (It's Complicated)
This is the part that causes the most confusion and, frankly, the most frustration in the community.
SDXL: fully open
SDXL and all earlier versions use a permissive open-source license. You can use the outputs commercially, sell AI-generated art, build products on top of it, modify the model, share fine-tunes. No revenue caps, no restrictions. This is why SDXL remains the backbone of most Stable Diffusion workflows.
SD3: free but with limits
SD3 changed the game with a more restrictive license. Stability AI allows free use for research, non-commercial projects, and businesses making under $1 million per year in revenue. If your company makes more than that, you need a commercial license from Stability AI. The community wasn't happy about this, and it's a big reason many people stick with SDXL.
SD3.5: even more restricted
SD3.5 follows the same pattern as SD3 but with tighter terms around commercial use and redistribution. It's the best model quality-wise, but the licensing overhead makes it impractical for many commercial applications. If you're building a product, talk to Stability AI's sales team before using SD3.5 in production.
Commercial Use Warning
If you're selling AI-generated images or running a service that uses Stable Diffusion, stick with SDXL unless you've confirmed your SD3/SD3.5 usage is covered by the license. The revenue threshold is based on total company revenue, not just AI-related income.
Alternatives Worth Considering
Stable Diffusion isn't the only option anymore. The AI image generation space has gotten crowded, and some alternatives are worth a look depending on what you need:
| Tool | Price | Open Source? | Best For |
|---|---|---|---|
| FLUX.1 (Black Forest Labs) | Free locally, API available | Yes (some versions) | High quality, good text rendering |
| Midjourney | $10-60/month | No | Artistic quality, easy to use |
| DALL-E 3 (OpenAI) | Via ChatGPT Plus ($20/mo) | No | Integration with ChatGPT |
| Google Imagen 3 | Via Gemini | No | Photorealism |
| Leonardo AI | Free tier + paid | No | Web UI, easy workflow |
FLUX has been gaining serious ground since late 2024. It handles text in images much better than Stable Diffusion and is open-source (for some versions). If you're starting fresh and don't have years of SDXL workflows built up, FLUX is worth trying first.
Midjourney is still the quality leader for artistic and creative images, but it's not open-source and has no local option. You're locked into their Discord/web interface and their pricing.
Getting Started for Free
The fastest way to try Stable Diffusion without installing anything:
- Google Colab notebooks: free GPU access, no setup (search for "SDXL Colab notebook" on GitHub)
- Hugging Face Spaces: browser-based demos of various SD models
- Stability AI free credits: sign up at platform.stability.ai for 25-200 starter credits
For local setup, the most popular tools are ComfyUI (node-based, powerful but steep learning curve) and Automatic1111 WebUI (simpler interface, huge community). Both are free and open-source.
For AI video (not just images), see our AI video generator comparison.
Don't want to set up local GPU infrastructure? Merlio's text-to-image tool runs Stable Diffusion and other image models in the browser with no install needed.
Sources
- Stability AI Platform Pricing - API credit costs and plans
- Stability AI License Page - current licensing terms for all models
- SDXL 1.0 on Hugging Face - model weights and license
Frequently Asked Questions
Generate Images, Chat with AI, Create Videos.
No credit card • Cancel anytime
