How to Run Flux Kontext Dev on a GPU Droplet

To get started with the demo, follow along with the steps shown below.

Setting up the GPU Droplet

To get started, we will need a DigitalOcean account to create a GPU Droplet. For this example, we recommend using an AMD MI300X or NVIDIA H100 powered GPU Droplet to run this demo. These GPUs will have the sufficient power to quickly run the model.

Follow the instructions in this tutorial to see how to get started setting up your environment to run the ComfyUI. Once you have set up the GPU Droplet, navigate to the next section of this tutorial.

Setting up the ComfyUI

Follow the instructions on the ComfyUI Github repo to install the repo correctly onto your GPU Droplet. If you are on an NVIDIA GPU, you can paste in the following to automate the process.

git clone https://github.com/comfyanonymous/ComfyUI
cd ComfyUI
python3 -m venv venv_comfy
source venv_comfy/bin/activate
pip install -r requirements.txt

This will install all the required packages for us. Next, we will get the required model files. Make sure that you have logged into HuggingFace on your machine before continuing. Paste the following code in to install the model files.

huggingface-cli download black-forest-labs/FLUX.1-Kontext-dev flux1-kontext-dev.safetensors --cache ./models/diffusion_models/
huggingface-cli download black-forest-labs/FLUX.1-Kontext-dev ae.safetensors --cache  ./models/vae/
wget -O ./models/text_encoders/t5xxl_fp8_e4m3fn.safetensors https://huggingface.co/comfyanonymous/flux_text_encoders/resolve/main/t5xxl_fp8_e4m3fn.safetensors
wget -O ./models/text_encoders/clip_l.safetensors https://huggingface.co/comfyanonymous/flux_text_encoders/resolve/main/clip_l.safetensors

This will download all the model files to run Flux Kontext dev. Now that this is complete, we can move on to the next section. Run the ComfyUI by pasting the following command into the terminal.

python3 main.py

Then open the link, and paste it into your VS Code (or now Cursor!) simple browser window, as shown in the tutorial we linked to earlier.

To get started, we recommend using the template provided by ComfyAnonymous in their ComfyUI examples repository, shown below.

Save the image file above, and you can upload it to the ComfyUI as a template by clicking Workflow>Open in the top left or ctrl+o. That should give you something like this:

From here, we can begin generating edited images! Simply upload your image into the modal on the left side of the workflow to get started, and enter a good prompt in. Check out the next section for some examples.

Above we have an example of several manipulations and edits we made using a photo of the author. As we can see, the capabilities of the model are wide ranging: from simple style transfer and background removal to full on translation and transformation. We had great success doing popular internet techniques like transferring the style of popular animation studios, and doing complex image editing. The model also seems capable of extrapolating smaller details from the larger image as needed.

In short, the model excels at simple editing tasks. The more descriptive the input prompt, the better the model can do. It is also excellent for tasks like inpainting and outfilling, which we found testing the pro model in the Black Forest Labs playground, as well. Overall, Flux Kontext is a very potent model for both image editing and generation, and we encourage everyone to try it out on a DigitalOcean GPU Droplet.

Flux Kontext dev is a truly powerful image editing model. In our experiments, we found it to be as capable as GPT-4o and other competitive image editing models at corresponding tasks. Be sure to test out Flux Kontext on a DigitalOcean GPU Droplet soon!