Epic LoRAs for Qwen Image Edit! (Patreon)

Video - https://youtu.be/KYKKpVcdrqg

Overview

Loads of Qwen Image Edit LoRAs are being made available for download from Huggingface, so I figured it was time to try out a load and see which are the best!

This includes LoRAs for:

Relighting

Face Swap

Retexture

Multi-Angle

Group Photos

Photo-to-Anime

Though I did try a few more, such as an upscale LoRA, I think these ones are some of the best.

Links for all the LoRAs and the required key phrases are in the "LoRA Notes" section of the workflow.

UPDATE

Also included is a SIMPLE version of the workflow which doesn't have expert options such as AI enhanced prompts, image compare and VAE up-scaling :)

Relighting

For example, and input like the one below with prompt "Relight the image using sunlight filtering through the blinds (creating striped shadows)":

Will "relight" the image adding striped shadow:

It has a whole load of relighting options and for the most part things work with English prompts, however using the example text from https://huggingface.co/dx8152/Qwen-Image-Edit-2509-Relight/tree/main/%E6%95%B0%E6%8D%AE%E9%9B%86 can give better results.

If you've got a badly edited image with a couple of layers like in the examples...

... then this LoRA can help give that more "integrated" feel to the image!

BFS Face Swapping

There are various "BFS" models models and I did most testing with V3 as that is allegedly the best ;) The body image goes first and the face image goes second, like this:

Which then results in a fairly decent face swap... even if the jack does go a bit strange!

It didn't work every time, however, and sometimes gave some... interesting results, such as this example input:

Which ended up giving this result:

Which is still perfectly acceptable to me, given how weird it is.

Often the prompt needed changing from their default example to maintain hair or expression, for example. Plenty of fun to be had either way though.

Retexture

Applying a specific texture from one image to another was also pretty good, like in this example:

For the prompt I asked it to apply the texture to her skin:

which certainly isn't creepy looking at all. Also worked well on clothing, trees, mugs, etc.

They also provide a texture extract LoRA as well, but I didn't have quite as much success with that for extracting useful texture images.

Multi-Angle

The Multi-Angle LoRA worked pretty well with English phrases, unlike the relight LoRA by the same person.

This works best with a single image such as this example:

which can then be "viewed" from various different camera angles such as left:

or from above:

plenty of other options available too, allowing you to "explore the world" of a single image.

Group Photos

Qwen can do group photos without a LoRA, but if you've been having issues with that, then the "PhotoUs" adapter may help!

Pick your two or three people to put together with an appropriate prompt...

... and it should do a fairly reasonable job of keeping the likeness, along with a new pose and scenario, like in the result below:

Photo-to-Anime

Making anime versions of images is always popular, so why not try out a LoRA which does just that? And yes... it works with rodents!

One thing I found with this LoRA, however, is that it only really worked if you "enable images via prompt", or else the style wouldn't get applied...

With that enabled, then your input image... such as this example:

would get changed into a very good anime style version:

Workflow

Some small changed to my Qwen Image Edit this time. The options have been split out (like above) so that it#s easier to pick the 1, 2 or 3 image latent encode options without forgetting to turn the unused ones off...

There is also a VAE switch now as well:

This means you don't have to load the non-upscale VAE if you're only using the upscale version. It will also use the same VAE throughout for all the various options.

Also a slight update to the AI prompt generation text, which should help produce some slightly better prompts. Give it a go and let me know if you prefer the new one!

The updated workflow is attached, and the anime-example one is exactly the same, just with the options to make sure the anime LoRA does it's thing by default (images get encoded via the prompt method).

All the links and trigger phrases are in the LoRA note for easy access:

Enjoy!