FLUX Pro vs. Dev vs. Schnell: Which Image Model Is Right for You?

If you've been following AI, you’ve probably heard about Flux, a set of AI image models trained by Black Forest Labs (the minds behind Stable Diffusion).

Flux comes in three flavors: Pro, Dev, and Schnell, each with its own strengths in quality, speed, and use cases.

But which one should you use?

In this comparison, I’ll break down the key differences between each Flux model, which ones you should use, and where you can use them for free.

Image Quality

The models are ordered in descending terms based on their model size, meaning their quality is also descending from Pro to Dev to Schnell

Pro has the highest quality of the three models and is also one of the best models in general. After Flux launched, Flux Pro ranks #2 in the AI image leaderboard on Artificial Analysis.

Dev and Schnell aren't too far behind however, ranking 37 points and 79 behind Pro, respectively.

And arguably, the different isn't major. Below is the same prompt used in Schnell, Dev, and Pro.

I'd say Dev and Pro are noticeably better than Schnell, which has a more 3D, animated type quality, whereas Dev and Pro have a more realistic quality.

However, between Dev and Pro, I'd say the main difference is Pro has slightly more detail and better lighting, but not enough to consistently notice.

Here's how I'd summarize it:

• Schnell prioritizes speed over details and is solid but may lack depth and precision
• Dev is nearly as good as Pro, but with slight compromises in fine details and lighting
• Pro is the highest-quality and is detailed and realistic with good lighting, texture, and sharpness

What about for anime/cartoon images?

I think this is a good comparison because you can similarly see the jumps between Schnell and Dev in terms of quality, but notice the minor differences between Dev and Pro. For example, there are more lanterns in the Pro image, and they have a more luminescent glow. The windows and stone road also display more luminescence. The jump between Dev and Pro however is still more subtle.

The main choice between these models, therefore, should likely be oriented around their speed, cost, and controllability.

Speed

Below are benchmarks from Artificial Analysis. Three of the four red bars are Flux endpoints on Replicate, which I chose to keep the comparison more apples-to-apples. Replicate optimizes the speed of these endpoints, so I would focus on considering the relative differences between the three Flux models.

• Schnell is significantly the fastest
• • Dev is 3.6x slower than Schnell
  • Pro is 8.1x slower than Schnell
• Dev is somewhat faster than Pro
• • Pro is 2.28× slower than Dev

Note, the differences aren't crazy on absolute terms, but that's because Replicate uses high-end GPUs. If you use slower GPUs, like a T4, these seconds really matter and you can multiply them by over 100x in some cases.

The way I would think about these three options is:

1. Schnell is for when you really need speed. For example, prototyping, casual content creation, and storyboarding.
2. Dev is for when you want a jump up in quality. For example, when you're creating something user-facing or a personal project youw ant to be proud of.
3. Pro is for when you are working on a commercial project that requires the best quality. For example, marketing materials, product visuals, and concept art, where you might need that extra 10% of quality.

There are more considerations however, beyond speed and quality. For example, how accessible are these models, and what about the ecosystems around them? E.g. is there enough support, open source development, loras, etc.?

Accessibility

• Schnell is fully open-source (Apache 2.0), free for commercial use, and can be run locally with minimal setup (provided you have a GPU) or accessed online instantly on nearly any image generation API platform, including Replicate and FAL.
• Dev is open-weight and free for non-commercial use. Can be run locally but requires a solid GPU (ideally 3090 for consumers, or A100 on cloud providers)
• Pro is available through commercial platforms, requiring licenses or API access. Note: this makes Pro much harder to develop around, meaning Pro usage is both low and doesn't have a community or many resources.

For these reasons, you really should for practical purposes just need to use Schnell and Dev. Pro would only be in consideration if you're a marketing agency, corporation, or shipping a high-end project for high-value customers.

The Easiest Way to Try Flux Schnell

It's much easier to get started just by creating your own image.

Magic Hour offers a fast, free version of Flux Schnell online, with no sign up required. I recommend generating 20 images (your daily free limit) to get a sense of how it looks for the types of images you want to create.

Recommended Test Prompts

To get a sense of the range of image generators, try generating these types of images.

• Text: Flux works quite well with text in images. Just add single quotes around the text you want to add, like: "An astronaut holding a sign that says 'Dream Big' near the space station."
• Illustration: "Cartoon-style forest with colorful trees and a small wooden cabin."
• Realistic Portrait: "A candid photo of a young woman smiling on a sunny beach."

Final Thoughts: Which Flux Model Should You Choose?

Choose Schnell if you're getting started with Flux, and use it to get a sense of how Flux models work. Use Schnell to prototype, test, and generate images locally.

Choose Dev if you want higher quality and are willing to wait longer, have a high-end GPU, or are willing to pay for API providers.

Both Schnell and Dev have an open source community building on top of them, meaning you can get controllability by using things like loras, which push the style in certain directions, e.g. an anime lora. Dev is also integrated with Flux tools, which allow you to further push things like controllability.

The easiest way to use Flux is to use free tools online. The easiest way to use Flux in production is to use API providers. The easiest way to go deep into Flux tools is to run everything locally, but you will need a powerful GPU for that, which is a big upfront cost.

Overall though, you can't go wrong with using Flux models right now. They've stood the test of time in AI, which is no easy feat, and they'll likely be around for a while.