If you have 2 images to use as reference for Qwen Edit , write stuff in prompt that will put attention on the target image

Prompt attention applies to reference images but also applies to images created from pure noise , as the generation happens over multiple steps. Since in T2i you start from pure noise (no stuff in image to reference) using terms like 'left' , 'right' , 'up' or 'down' will still place attention in the cardinal directions of the image. Here are more examples of prompt attention.

//---//

Back to more practical matters , consider we wish to merge these two images with character replacement using Qwen Edit

Image 1

To be placed in image 2

//---//

So low frequency noise = remove white background , she has a blue bodysuit , and she has yellow shoes , she has long hair

High frequency noise = she sits by a car in a parking lot at night

And usually top off with ' she has light hair '

so prompt is ``` remove white background , she has a blue bodysuit , and she has yellow shoes , she has long hair , she sits by a car on the right in a parking lot , the sky is dark , she has light hair ```

You can set as 512x1024 for 0.9 credits cost at 8 steps

Costs twice as much otherwise at 1.8 credits

So not great initially cuz I didnt describe the phonk girl by the car enuff in prompt

But that can be fixed by straight up captioning the phonk girl oarking lot image and appending it to the end to bring more attention to it

``` the night sky is black , parking lot , remove white background , she has a blue bodysuit , and she has yellow shoes , she has long hair , she sits by a car on the right in a parking lot , the sky is dark , standing next to a sleek, dark-colored sports car. The car has a distinctive front grille and is parked in a parking lot at night, illuminated by streetlights. The background features several other cars parked in a lot, with some trees and bushes visible. The person appears to be posing for the camera, with one hand resting on the hood of the car. she has light hair ```

Now I brought too much attention the the parking lot image

Likely cuz I decided to start with ``` the night sky is black ``` which brought attention to the girl too as she is also dressed in black

Final prompt after tinkering can be

``` remove white background , she has a blue bodysuit , she has yellow shoes , she has light hair , she sits by a car on the right in a parking lot , the sky is dark , standing next to a sleek, dark-colored sports car. The car has a distinctive front grille and is parked in a parking lot at night, illuminated by streetlights. The background features several other cars parked in a lot, with some trees and bushes visible. she has a blue bodysuit , she has yellow shoes , The person appears to be posing for the camera, with one hand resting on the hood of the car. she has light hair ```

Post: https://tensor.art/images/954799554404853126?post_id=954799012165218733

This Qwen Edit LoRa is optimized for 3 image merging: https://tensor.art/models/954755878882418264

Thats it for the tutorial on using prompt attention for Qwen edit generation.

/Cheers!