What does it do?

This workflow was designed for the "Weather Changing" challenge.

It involves uploading an image and describing the desired situation in the dedicated box.

For example, typing "Heavy snow" will result in an image depicting heavy snowfall, while typing "lightning storm" will display a powerful thunderstorm.

Similarly, entering "sandstorm" will transform the image into a location filled with sand both on the ground and in the air.

A unique feature of this workflow is its flexibility: in addition to the predefined weather options required by the challenge, you can input any word to modify the image's situation.

For instance, typing "flood" will portray the location in the photo as flooded, adding a dramatic and realistic element to the scene.

How does he do it?

This process is made possible by using the "instruct pix2pix" checkpoint.

To make this checkpoint work on Comfyui, a method combining Controlnet and the new Ipadapter was necessary.

The checkpoint is activated through positive requests, specifying the desired change.

For example, you can request to change the hair color of a person or add scenic elements to the photo.

Additionally, I've used LLM to describe the scene and create the prompt needed to generate a high-resolution image, also leveraging Ultimate SD Upscale to highlight details.

Research and Information

For further insights into using "instruct pix2pix," you can search for information on YouTube or Google.

Or go directly to Stable Diffusion's official website at this address:

https://stable-diffusion-art.com/instruct-pix2pix/

the model I used is the older model, but doing a search afterwards I found a newer model, which you can find at CivitAI

I tested the most updated model, found on CivitAI, it is lighter, only 2Gb compared to the 7Gb of the old model, I put here some images generated with this model.

The word "Ligthning Storm" was used.

The word "flood" was used.

In conclusion, this workflow offers a fun and flexible way to modify images according to personal preferences, including simulating natural phenomena such as floods.

below some extra challenge images

generated with old model

The word "Flood" was used.

The word "Autumnal" was used.

The word "Christmas" was used.