FLUX.1
VS
Stable Diffusion 3

A picture is worth a thousand words, but which words will these AI models paint better?

Introduction

AI Image Generation is the latest luxury that is now available at our fingertips. One of the pioneers in the field was Stability AI, creators of Stable Diffusion. Recently some of their top minds went rogue, and then out of nowhere released FLUX.1 sending waves across open-source community. Let's follow the bees, and put Flux.1 against Stable Diffusion 3 Medium in a head-to-head review.

Get API Key

Practical tests

Alright, let's put these bad boys to the test. After we go through commonly known image gen weakspots, we'll see who comes out on top.

Prompt 1. Hands/Fingers:

A 1st person view of a man holding a paintbrush over a canvas.
The canvas has a drawing of an unfinished horse in a garden. 
His other hand over the paint colours.

‍

Stable DIffusion 3 Medium Output

a man drawing a horse. 3 hands are present — SDM3 fingers prompt ❌

FLUX.1 Output

a man drawing a horse — FLUX.1 fingers prompt ✅

Results:
FLUX.1 is a couple of steps ahead, while Stable Diffusion 3 notoriously struggles with fingers, as expected.

If such pressure-testing is not for you and you want to see the aesthetic site of FLUX.1 - check out 10 cool prompts for the model.

Prompt 2. Font Eligibility

Create a classroom of young robots. 
The chalkboard in the classroom has 'AI Is Your Friend' written on it.

Stable DIffusion 3 Medium Output

half-robots sitting in front of a whiteboard with writing "AI is your friend" — Stable Diffusion 3 Medium font eligibility output ❌

FLUX.1 Output

robots sitting in front of a whiteboard with writing "AI is your friend" — FLUX.1 font eligibility output ✅

Results:
FLUX.1 just looks right. Stable Diffusion, on the other hand - even with many generation attempts fails to deliver.

Prompt 3. The Devil's in the details.

Rough thread embroidery on a girl's face, poor quality, 
no photoshop, no retouching --ar 3:2 --stylize 250

Stable DIffusion 3 Medium Output

a girl with embroidery on her face — SDM3 detailing output ✅

FLUX.1 Output

a girl with heart embroidery on her face — FLUX.1 detailing output ✅

Results:
Asking Stable Diffusion to generate humans at this point is just asking for trouble. BUT! It performed rather well, with no significant artifacts. FLUX.1 did great as always.

Prompt 4. Number of details until it gives out.

A misty forest landscape with Snow White and her seven 
dwarves walking along a carpet of lilies at 
sunset with enchanting fireflies drawn in cartoon style. 

A river is flowing in the background, 
and tress have 5 apples on them in total.

Stable DIffusion 3 Medium Output

5 creatures looking like half snow-white, half-gnomes are walking across a forest — SDM3 attention to details prompt ❌

FLUX.1 Output

snow white and 5 gnomes walking across a forest — FLUX.1 attention to details output ✅

Results:
Stable Diffusion generated something powerful, but in all the wrong ways. The unsettling mix of gnomes with snow white was revealed each generation, and usually the apples were missing.
Flux.1 did a great job, only failing to count apples and gnomes.

Prompt 5. Variable Binding.

An Image depicting four cubes with the word FLUX on them, 
each letter of FLUX is on one side of a separate cube

Stable DIffusion 3 Medium Output

three cubes with cryptic symbols — SDM3 variable binding ❌

FLUX.1 Output

cubes with lettering FLUX — FLUX.1 variable binding ✅

Results:
The difference is huge - but we were testing variable binding. FLUX.1 succesfully bound each letter to a separate cube.

Get API Key

Pricing

The Pricing model is given in AI/ML API tokens. We will provide pricing for FLUX.1 pro, although some of the generations where taken from the cheaper FLUX.1 dev.

1024x1024 price	FLUX.1 pro	Stable Diffusion 3 Medium
1024x1024 price	$0.0525	$0.03675

Get API Key

Compare for yourself

You've seen what these models can do - now try them for your use case. Plug the code below into Google Colab or any IDE, use your API Key, and get testing!

%pip install openai
import os
from openai import OpenAI
import requests
url = "https://api.aimlapi.com/images/generations/"

model1="flux-pro"
model2="stable-diffusion-v3-medium"
prompt="""
Create an image of four books on a shelf.
"""
payload1 = {
  "prompt": prompt,
  "model": model1,
}
payload2 = {
  "prompt": prompt,
  "model": model2,
}
headers = {
  "Authorization": "Bearer <YOUR_API_KEY>",
  "content-type": "application/json"
}

print("\n RESPONSES BELOW")
response1 = requests.post(url, json=payload1, headers=headers)
response2 = requests.post(url, json=payload2, headers=headers)
print(f"{model1}: {response.json()}")
print("\n")
print(f"{model2}: {response2.json()}")
print("\n")

‍

Conclusion

It is crystal clear that Flux1. Pro produces the finest imagery out there while abiding with intricate details. The Stable Diffusion 3 model was only superior in the matter of speed where it generated images an average of 5 seconds to produce two images, while Flux1. Pro took an average of 16 seconds.

You can check our model lineup here - try any of them for yourself with our API Key.

Get API Key