SREF (Style Reference) is one of the newest features released by Midjourney which essentially behaves like a style training LoRA. It allows you to consistently generate images in a particular style aesthetic simply by learning from a supplied image.
Its pretty cool because you can get consistent results with just one image.
How to use Style References
- Start with your prompt and use the argument
--sreffollowed by a URL of an image (one or more).
- Midjourney will analyse the image(s) provided and create your image using the style and aesthetics of that image.
- It works with both V6 and Niji V6 models. Does not work with earlier models.
- You can use multiple images like this
--sref urlA urlB urlC
- You can also set weights of styles like this
--sref urlA::2 urlB::3 urlC::5
- You can also set the total strength of the stylization via
--sw 100(100 is default, 0 is off, 1000 is maximum)
- You can also use it when using image prompts, however your images must go before
/imagine cat ninja ImagePrompt1 ImagePrompt2 --sref stylePrompt1 stylePrompt2
Another way to look at this is that Midjourney is colour grading your image that it generates using the colour scheme of the reference image provided.
Let’s explore this further with a reference image from Mad Max: Fury Road
Let’s create an image using this prompt:
a old 1965 Post-Apocalyptic mustang in a dystopian scene, mustang is heavily modified with armour and weapons to protect it's driver, shot from behind the car overlooking a desert landscape and rocky mountains, camera is placed at a very low angle --ar 239:100 --sref https://s.mj.run/C76XE6JqaTA --stylize 200 --v 6
You see the resulting image has the same colour grading ie. aesthetics and style as the source image. Yet we did not reference anything that relates to the colour scheme in the prompt. Now to prove that we can run the same prompt without the
a old 1965 Post-Apocalyptic mustang in a dystopian scene, mustang is heavily modified with armour and weapons to protect it's driver, shot from behind the car overlooking a desert landscape and rocky mountains, camera is placed at a very low angle --ar 239:100 --stylize 200 --v
You can see these images have different aesthetics and its based on Midjourney’s own interpretation of what it should look like. So using the –sref style reference has immense benefit to the results.
Let’s try from another movie – Tenet
here is the result with prompt:
a african american man in a well dressed suit standing in a secret research facility --ar 239:100 --sref https://film-grab.com/wp-content/uploads/photo-gallery/Tenet_027.jpg?bwg=1614340151 --sw 300 --stylize 200 --v 6
I mean this is incredible what Midjourney does. SREF style reference let’s you replicate the aesthetics of a reference image in the image you are creating with the prompt. It pretty much replicates colour grading of the reference image, so its a cool way to create consistent images using Midjourney in a particular style so you can have consistency in their appearance and aesthetics.