Flux LoRAs are a really powerful way to customize AI image generation for specific styles. You simply specify one or more Flux LoRA adapters during image generation, and they apply almost like "style filters" to generate images in a consistent style. The Together API now supports using Flux LoRA models for image generation – here's a blog post showcasing some of the most interesting LoRAs for you to experiment with.

In this post we’ll look at how to use the Together API to run Flux LoRA models, what kinds of images you can generate using some popular LoRAs, and we’ll even show you how you can run inference on LoRA models that you’ve trained yourself on the Together Platform.

Let’s get started!

{{custom-cta-1}}

Using a LoRA to create a specific look and feel

LoRA (Low-Rank Adaptation of Large Language Models) is a technique that offers a convenient way to reproduce specific visual styles when generating new images from text prompts.

For example, the Flux-Sketch-Ep-LoRA model generates the following images for prompts of different city buildings:

Try the Flux Sketch LoRA on the Together Playground →
Don't forget the trigger words: ep sketch

To use a Flux LoRA, you can specify the new image_loras option when using the FLUX.1-dev-lora image model with the Together API, along with the LoRA’s trigger word in your prompt.

For example, the trigger word for Flux-Sketch-Ep-LoRA is “ep sketch”. So, to make the image of the cafe, we could write this using the Python SDK:

from together import Together

client = Together()

image = client.images.generate(
    prompt="ep sketch, A vibrant, colorful, sketch illustration of a sleek, retro-style sports car with bold, expressive lines, bright, pop-art inspired colors, and dynamic, swirling patterns in the background.",
    model="black-forest-labs/FLUX.1-dev-lora",
    height=832,
    width=1280,
    steps=33,
    image_loras=[
        {
            "path": "",
            "scale": 1,
        },
    ],
)

print(image.data[0].url)

We can also use the TypeScript SDK:

import Together from "together-ai";

const client = new Together();

const image = await client.images.create({
  prompt: "ep sketch, a cafe",
  model: "black-forest-labs/FLUX.1-dev-lora",
  width: 1024,
  height: 768,
  steps: 28,
  image_loras: [
    {
      path: "",
      scale: 1,
    },
  ],
});

console.log(image.data[0].url);

As you can see, just by providing the URL of a LoRA and adding the trigger word to your prompt, you can easily generate images that all have the same look and feel.

Let’s explore some more popular LoRAs, and the various use cases they work for.

Logo Design

This model can create simple logos from one or two words:

Try the Flux Logo Design LoRA on the Together Playground → 
Don't forget the trigger words: logo, Minimalist

Model details:

• LoRA: huggingface Shakker-Labs/FLUX.1-dev-LoRA-Logo-Design · Hugging Face
• Trigger words: “logo, Minimalist”

Colored sketch

This model is great for sketches that have a hand-drawn feel:

Try the Flux Colored Sketch LoRA on the Together Playground → 
Don't forget the trigger words: ep sketch

Model details:

• LoRA: huggingface Shakker-Labs/FLUX.1-dev-LoRA-Logo-Design · Hugging Face
• Trigger words: “ep sketch”

Outfit generator

This model is trained on 20.5k samples from H&M, so it’s perfect for generating outfits in isolation.

Try the Flux Outfit Generator LoRA on the Together Playground → 
Don't forget the trigger words: For this model, you must detail the outfit’s color, pattern, fit, style, material, and type.

Model details:

• LoRA: https://huggingface.co/tryonlabs/FLUX.1-dev-LoRA-Outfit-Generator
• Trigger words: For this model, you must detail the outfit’s color, pattern, fit, style, material, and type.

Anime blockprint style

This unique model generates anime-style images with bold contrast and clean lines:

‍

Try the Flux Anime Sketch LoRA on the Together Playground →
Don't forget the trigger words: blockprint style

Model details:

• LoRA: https://huggingface.co/glif/anime-blockprint-style
• Trigger words: “blockprint style”

Icons

Use this model to generate vector-style icons:

Try the Flux Icon Kit LoRA on the Together Playground → 
Don't forget the trigger words: icon kit

Model details:

• LoRA: https://huggingface.co/strangerzonehf/Flux-Icon-Kit-LoRA
• Trigger words: “icon kit”

Using your own LoRA on Together AI’s serverless platform

You can also use the Together Platform to run inference against any LoRA you’ve trained yourself.

LoRAs can be trained with as few as 10 images, making them much easier to use for custom model creation than traditional fine-tuning techniques, which typically require datasets with thousands of examples for training.

Additionally, since the Together Platform supports any Flux LoRA, you can run inference against your own custom model without the need for any additional infrastructure.

Once you train your own LoRA, upload it to HuggingFace, then specify its URL and trigger words just like in the examples above:

from together import Together

client = Together()

image = client.images.generate(
    prompt="your trigger word, a cafe",
    model="black-forest-labs/FLUX.1-dev-lora",
    height=832,
    width=1280,
    steps=33,
    image_loras=[
        {
            "path": "",
            "scale": 1,
        },
    ],
)

print(image.data[0].url)

This endpoint uses the same pay-per-token pricing as all other serverless endpoints, so you never have to worry about starting or stopping a server just to use your custom models.

Getting started with Flux LoRAs

To get started with Flux LoRAs on Together AI: 

• Sign up for our platform, and upload existing LoRA adapters to try them out on our serverless platform. 
• We also have a python notebook you can follow along with to help you experiment with LoRAs. 
• Read the docs to get started via the Together API.

Ready to scale? Connect with us about dedicated GPU endpoints, and enterprise deployments for FLUX.

Happy Building!