The new EDIT function in Midjourney

Hopefully you are prompting on the web already instead of in the Discord environment? As below is focussed on prompting on the web.

 

Relating back to the last post about Random Style References, someone has been really sweet and offers us an easy visual website where you see the code and some example images. The total number of entries was 95000 at the the 2nd of November. Check it out here The SREF Project.

 

Then if you like, you can even give weights to the style references on the web. Here is an example prompt, just for fun:

medium close-up photo in style of Vogue photography --chaos 20 --ar 1:1 --style raw --sref 3095667286::3 3219755275::4 4222940597::10 --stylize 50 --v 6.1

 

There is a new function launched in Midjourney and is called ‘Edit’, here you can upload an image or use an image of the internet and change it as you like (you do need a yearly subscription, or you had a monthly subscription during the last 12 months or prompted > 10.000 images).

 

A little run through:

Upload a file from your own archive or from the web with a URL.

If you click the button “Suggest Prompt”, you will find out it has the same function as /describe on discord. You only get one prompt now instead of four.

If you like the layout or the composition of the image you could make a selection to change something in the image. If you like the composition, but you don’t like the style, you can go to ‘Retexture’ and add a prompt with some style references, or just leave the prompt as it is and change some colours descriptions of the furniture, as I did here.

 

Start image:

 

Restructure:

 

Now I want to change the sofa to an (ugly) green sofa. Add the sofa of your likings to the prompt as an image reference and use –iw 3, make sure the sofa is on a white background:

 

Then I want to add some (ugly) pillows. Make the selection, add the pillow on a white background as an image reference and use the maximum image weight –iw 3. If Midjourney does not deliver add the subscription of the pillow to the text prompt and make your selections a bit bigger:

 

Now I would like to add a coffee table. Again find yourself a coffee table on a white background, add the text to the text prompt and use the image reference with –iw 3:

Not perfect, so make a small selection to give Midjourney the opportunity to fix it:

Now I want to change the colours by using ‘Restructure’ again and now I try a colour palette as a style reference with a low style weight of –sw 15 (make sure the image has the paperclip symbol). As you will notice, the composition of the image stays the same, the object on the table did change. The pillows feel more integrated in the whole:

Then I was curious what will happen if I decided to select some of the same style of images as at the start of this post as style references:

 

And you could do this with an image of your own living room, bedroom, whatsoever. This tool is very powerful. You can also use an image of a person and select everything around the face and change that. Some kind of InsideFaceSwap, but a little different. You do need some photoshop skills to integrate the uploaded face with the results of Midjourney.

If you used the ‘Full Editor’ on an image, you cannot use the simple ‘Editor’ anymore, but you can go back to the ‘Full Editor’ again, to keep making changes. When you like the final result, you can upscale it to your gallery.

 

Here a little overview of some other AI tools you might find interesting:

The most interesting one I found is Letz.ai

Here you can train a model for a person, a style or an object. You only need to upload 10 to 20 images of the person, style or object and the model will be trained. If the model does not deliver what you prefer you can change it and add or subtract images you used. Still have to test it intensively but I was thinking about trying to combine a product with a person and a style. More to come next time.

 

Other interesting tools:

Ideogram is leading in text generations.

Reshot AI is handy when you want to change the expression of your character. The quality for extreme changes in the expression is not as good as I hoped.

And when you have some nerdy characteristics try Flux on replicate.com.

For video Kling AI is the one I like best, especially if you use distortion in the negative prompt. The quality is insane.

 

Happy Prompting!

#keeponlearning