Stable Diffusion Review: Is It Free? SD 3.5, License and Price

Stable Diffusion is a family of AI image models from Stability AI, not an app, and the current generation is Stable Diffusion 3.5, in three sizes: Large, Large Turbo and Medium. Is Stable Diffusion free? The models are: you can download them and use them, commercial work included, for free as long as you or your company make under $1M a year. Running it online is a different story: the official demos on Hugging Face give you a small daily quota, Brand Studio (the app that replaced DreamStudio) gives new accounts 25 free credits, and after that the API charges by the image, from a few cents each.

What Stable Diffusion is today

If this page brings back memories of Stable Diffusion 2 and endless prompt fiddling, a lot has changed. Stability AI now calls Stable Diffusion 3.5 "our most powerful image model yet" and sells it in three flavors. Large is the most powerful, aimed at professional work at about 1 megapixel. Large Turbo is designed to make good images in just four steps, and Stability says it runs faster than Large. Medium "strikes the balance between quality and customization" and, according to Stability, is designed to run on consumer hardware. The older SDXL and SDXL Turbo are still listed as additional models.

What makes Stable Diffusion different from most of the tools on this site is that you can get the model itself. The weights are on Hugging Face, where you sign in and accept the license before downloading, and from there you run them on your own computer with ComfyUI or the diffusers library. That's why tinkerers love it: you can fine-tune it, add LoRAs (small add-on models that teach it a style or a subject) and run it offline.

There are three other ways in. There's the Stability AI API for developers, which also covers editing (inpaint, outpaint, erase, remove background, search and replace, relight) and upscaling. There's Brand Studio, the web app. And there are cloud partners if your company already lives in one.

How I tested it

I started where anyone curious would: the official Stable Diffusion 3.5 Large demo that Stability AI runs on Hugging Face, without an account. My prompt was a little American road trip, with text in it on purpose, because text is where image models usually trip:

"Photo of a roadside motel in the Arizona desert at dusk, a red neon sign that reads "VACANCY - OPEN 24 HRS", a turquoise 1950s car parked in front, warm sky, no people"

My first run started and I cut it short by mistake. The second one never started: "You have exceeded your ZeroGPU quota," and a window telling me I'd hit my daily limit, with an offer to go PRO for $9 a month or create a free account. The Large demo asks for a big slice of GPU time per image, so without an account you get very little.

The Large Turbo demo, right after, did work. It returned a square 1024 by 1024 image in a few seconds, with 4 steps by default. The scene was spot on: a low motel, desert mountains, a turquoise car with whitewall tires, nobody in sight. The big "VACANCY" sign was perfect. The second sign read "OPEN 2IRS".

Then I signed in to Brand Studio with a Google account. It opened with a yellow banner, "You're currently using a Free Trial of Brand Studio!", and 25 credits in the corner. The model menu starts on Auto, which picks a model for you, but you can choose Stable Diffusion 3.5 Large, Large Turbo or Medium yourself, along with Stability's Stable Image Ultra and Core and models from other companies.

With SD 3.5 Large selected and one image, the counter went from 25 to 18 (the API dashboard showed the exact 18.5). And this time the sign was right on both lines: "VACANCY" and "OPEN 24 HRS", letter for letter. The PNG I downloaded had no visible watermark. I ran the same prompt once more with Medium, for 3.5 credits: "VACANCY" fine, "OPEN 2 HIRS" not so much, plus a second neon sign in squiggly script that doesn't spell anything.

Large, Large Turbo or Medium: same prompt, three results

Here are the three images side by side. Sorry if this gets a bit nerdy, but if you care about text in your images, this is the part that decides which model you pay for.

None of them added people, and all three got the turquoise 1950s car right. It's one prompt, so don't treat it as a lab result. But if your images need readable words, start with Large and drop to Turbo or Medium only for drafts.

The license, in plain English: who pays and who doesn't

This is the question most people come here with, so here's what Stability AI's license page says. The Community License is free "for researchers, developers, small businesses, and creators with less than $1M in annual revenue," and it covers the Stable Diffusion 3.5 models, SDXL Turbo and other Stability models. It allows research, non-commercial and commercial use. If you're making images at home for yourself, Stability's own answer is "No you won't need to pay - create away!"

Two more details from the license FAQ. You can fine-tune the models and train LoRAs, and use the results, but you can't use the models or their outputs to build a competing foundation model. And as between you and Stability, you own what you generate, as long as you follow the law and Stability's acceptable use policy. The license can be revoked if you break its terms.

What Stable Diffusion costs online

If you don't have a graphics card that can run it, you pay Stability by the image. The API page, read in US dollars, sets 1 credit at $0.01, gives new accounts 25 free credits and doesn't say whether prices include tax. Those 25 credits are the same ones Brand Studio shows: after my Large image, the API billing page read 18.5. The buy box was set to $10, which gets you 1,000 credits, or about 150 SD 3.5 Large images.

Beyond the three SD 3.5 models, the API lists SD 3.5 Flash, a distilled version of Medium, at 2.5 credits, Stable Image Core at 3 and Stable Image Ultra, built on SD 3.5 Large, at 8. Editing tools cost 4 to 8 credits, the fast upscaler 2 and the creative one 60. Brand Studio's own credit table matches those prices for the Stability models.

When the 25 credits run out, Brand Studio says you'll need to upgrade to a plan, and its plans page doesn't show prices. Stability describes two plans, Core and Enterprise. Keep in mind who Brand Studio is built for: brand IDs, campaigns and a Producer Mode for marketing teams. If you only want Stable Diffusion images, buying API credits or running the model yourself is the simpler route.

Is Stable Diffusion worth it?

If you have a decent graphics card and some patience, it's a strong option. A free license for commercial work under $1M, models you can run offline, fine-tune and plug into ComfyUI, and SD 3.5 Large got my two-line sign right. Paying by the image through the API is also cheap if you only need a few.

If you want to type a prompt into a nice app and keep a gallery, look elsewhere: Stable Diffusion no longer has a simple consumer app, since DreamStudio now redirects to Brand Studio, which is aimed at brands. And the free demos have tight limits.

Stable Diffusion or something else?

Flux is the closest match: another family of image models, with an open-weight version you can run yourself and a pay-per-image API. If you'd rather not touch any settings, Midjourney is a finished app with plans from $10 a month. And if you want a free plan with daily tokens and many models in one place, try Leonardo AI.

Written by Miguel Ángel, with help from an LLM to verify facts.