Flux 1: A New Image Generator for Stunning Visuals

Flux 1 is a powerful new image generation model developed by Black Forest Labs, the same team behind Stable Diffusion. It excels in generating realistic images, surpassing other leading models like Midjourney and Dolly in overall quality and realism. Flux 1 is especially noteworthy for its exceptional text rendering capabilities, even producing impressive results when generating human hands.

A bedroom with a window on the left and an archway leading to a misty forest on the right. — An example of an image generated using Flux 1.

Getting Started

To use Flux 1, you’ll need to install CompVUI on your computer. You can find a separate tutorial on installing CompVUI if you haven’t already. Once CompVUI is installed, you need to download the Flux model files. There are two primary versions for local use: the dev version and the schnell version.

The dev version prioritizes quality, while the schnell version focuses on speed. For this tutorial, we’ll use the dev version. Download the model you choose and place it in the following directory: ConfuUI/models/unet.

A screenshot of a file explorer showing the directory structure for saving the Flux model files. — Saving the Flux model files in the correct directory.

You’ll also need to download three different encoders. Links for these files will be provided in the description. Save these files in the following directory: ConfuUI/models/clip.

The final file to download is the VAE model. This also comes in two versions, depending on whether you’re using the dev or schnell version. Download the appropriate VAE model and place it in the following directory: ConfuUI/models/vae.

Setting Up the Workflow

Once all the necessary files are downloaded, open CompVUI and update it to the latest version. Next, go to openart.ai and download the simple workflow file for Flux. Drag and drop this file onto the CompVUI interface to load the workflow.

In the workflow, select the correct Flux model you downloaded. You can choose between the FP8 encoder for speed or the T5-XXL-FP16 encoder for higher precision. Next, write your prompt describing the image you want to generate. To find ideas or write good prompts, refer to the prompt writing video linked in the description.

Generating Images

For a realistic example, let’s generate an image of an elderly woman sitting in a garden with flowers around her. Adjust the width and height of the output image as desired. Remember that higher resolution images take longer to generate.

Ensure the seed is set to randomize for diverse results. You can experiment with different sampler names and schedulers, but the default settings (euler and sgm uniform) are a good starting point. The steps value controls the quality and speed of the generation, with higher values resulting in better quality but slower processing.

A screenshot of the CompVUI interface showing the settings for generating an image with a specific prompt. — Setting up the parameters for image generation.

Finally, select the VAE model you downloaded from the list. If you renamed the file, make sure it’s placed in the correct folder. Now, you can click on “queue prompt” to start generating the image.

Flux 1 is not the fastest model available. Depending on your GPU’s VRAM capacity, you might encounter issues with processing speed. If your GPU has limited VRAM (12GB or 8GB), consider using the schnell version, FP8 encoder, or a cloud-based solution like ThinkDiffusion.

A screenshot of the CompVUI interface showing the progress of image generation. — Generating the image using the Flux 1 model.

After the process is complete, you’ll see a preview of your image. The output will likely be stunning, featuring impressive detail, depth of field, and realism.

Since the seed is set to randomize, you’ll get a different result each time you click on “queue prompt.” This allows for exploring various variations and potentially finding even better images with the same prompt.

Text Generation

Flux 1 also excels at text generation. To test this, let’s create a cover for a children’s book with a character and a large title.

A screenshot of the CompVUI interface showing the prompt for generating a children's book cover. — The prompt for generating a children’s book cover.

As you can see, Flux 1 generates accurate text, letter for letter, with a design that matches the prompt. This demonstrates the model’s impressive text rendering capabilities.

Hand Rendering

Flux 1 has demonstrated its ability to render hands realistically. For instance, one prompt generated an image of a hand with a correct finger count, showcasing the model’s detail and accuracy.

A close-up image of an elderly woman's face. — An example of an image with well-rendered hands.

Image Upscaling

For even higher quality images, you can use a third-party image upscaler like Topaz Photo AI to sharpen and upscale the generated images. This process effectively reintroduces details and enhances the overall quality of the image.

Conclusion

Flux 1 is a powerful and versatile image generation model with impressive capabilities. Its realism, text rendering, and hand rendering abilities make it a valuable tool for creative projects and visual exploration. Experiment with the different settings and prompts to discover the full potential of this innovative model.