Finally! New best local AI image editor is here transcript

Finally, we have a new open-source image generator and editor. So, Alibaba just dropped Quen Image 2.1, and this is an extremely flexible and powerful tool. In this video, we're going to go over all the incredible things that it can do, plus of course, I'm going to show you how to install this on your computer, so you can run it for free and unlimited times offline. The awesome thing is this can even run on very low VRAM.

Let's jump right in. First, let's go over its capabilities. Not only can it generate some super realistic photos, it's also great at anatomy and prompt understanding, as you can see from these comparisons, it's also really good at even creating some super complex stuff like diagrams and infographics with a ton of different elements. What you see here is just one textto image generation from Quinn image 2.1. Here's another example.

It's also great at designing interfaces. So, here are some examples for your reference. Now, probably the most powerful feature about Quinn image 2.1 is that it can generate transparent images. In other words, this can also output an alpha channel. So, here are some examples of different generations with transparency built in. Now, not only can this do text to image, but the really nice thing about this is it can also take in images as reference for you to edit further.

I don't know about you, but I've been waiting like months for another new image editor. The previous models that we got like creatu and ideoggram were only text to image. So you can't really edit existing images using natural language like nano banana. But finally for quen image 2.1 this is possible. So for example you can input the same transparent image and then just prompt it to change the text of the image. And here's our result on the right.

Or here's an even cooler example where we can input a photo like this and get it to extract only the branches and leaves in the foreground as a transparent layer. And as you can see, it can do a pretty good job of this. This can also take in up to 10 reference images for you to insert into your photos. So, here's an example of this in action. As you can see, it's great with facial and character consistency. Here's another awesome example where we can input a ton of different photos and just specify using natural language what out of each photo we want to extract.

For example, we want the woman in the first image wearing the pink jacket in the second image and the shoes from the third image. in this handbag and also wearing the hat from the last image. And it's able to combine whatever you specify into one seamless photo. Or here's another example where we can combine a ton of different furniture photos and add them into the same room. Or you can also just annotate directly on the photo that you want to edit.

For example, we can draw over these items that we want to change or remove. Or here's an example where we can draw directly over the image and get it to add something there. This is also great at character consistency, as you can see from this example. Or here's a really complicated outfit. Well, we can get it to generate a model wearing this outfit. And as you can see, it's able to preserve the design of the outfit very well.

Or here's another cool example where we could input this selfie photo and get it to generate a panorama from the photo. And after generating this panorama, you can then basically explore the scene in 3D. So, those are just some quick examples of all the cool things that Quinn image 2.1 can do. The best thing is this is open source, so you can download and run this locally. So, if you're interested, next let's go over how to install this.

Now, for this tutorial, we are going to use an interface called Comfy UI, which is like the most popular platform for running open- source image and video generators offline. It's completely free for you to install and use, and you can customize a ton of things with it, which is why I highly recommend you use Comfy UI for Quinn image 2.1. Now, if you're not familiar with Comfy UI, see this video first where I go over a full installation tutorial.

All right, so assuming you do have Comfy UI installed, the first step is to update this to the latest version. So, simply double click into this update folder and then double click on update Comfy UI.bat. So, this will proceed to update your Comfy UI to the latest version. All right, afterwards let's press any key to exit this window. And then next we can start up Comfy UI. So let me start this up. All right, once you open up Comfy UI, if you click on templates in the left sidebar and you search for Quen image 2.1, you should see several workflows.

Text to image, remove background, and image edit. So let's go over the text to image workflow first. Once you click on it, it should look like this. By the way, if you don't see these workflows, I will also link to this page where at the bottom you can download the workflow files and then just drag and drop them onto your company UI interface. All right. Now, once you open this up, you might get this red outline with some missing models.

So, let's also proceed to download the models. I'll link to this page in the description below. If you click on files and versions, here are the models that you need to download. First, let's click into diffusion models. And here you are given the option of either the full BF-16 one which is 14.2 GB in size or this smaller INT8 Conrot version which is 7.2 GB in size. This will fit on 8 GB of VRAM or even potentially less.

Now if you have lower than this if you have like 4 GB of VRAM. Don't worry later in the video I'm also going to show you another method to run this with even just 4 GB of VRAM. So I'm going to download this one and this goes in Comfy UI in models and then diffusion models. Let's click save. And then afterwards, we also need to download a text encoder. Now, the first two files here are used to enhance your prompt. So, this is an additional AI model which takes your prompt and enhances it further before it plugs it into the image generation node.

It's supposed to make your image quality better, but note that each file is like almost 9.5 GB in size. So, especially if you're running low on VRAM, I would not recommend this step. It's better if you just, you know, manually write a good prompt. So, the bare minimum that we need to download is one of these text encoders down here. Now, there's different compression levels here. The full BF-16 one is 17.5 GB in size.

We also have a W4 A8 one, which is the smallest. This is only 6.3 GB in size. If you have low VRAM, then this is highly recommended. So, let's click on download. And this goes in Comfy UI in models and then text encoders. Let's click save. All right. Afterwards, we also need to download the VAE for this. Now, this is fairly tiny at only 676 megabytes in size. Let's click download. And this goes in Comfy UI in models and then VAE.

Let's click save. All right. After you've downloaded all the models, going back to our Comfy UI workflow, simply press R to refresh our model list. And then down here is where we select the downloaded models. So, first of all, for the unit, let's select this Quinn image model that I just downloaded. And then for the clip, I will select this Quen 3 VL W488 model which I just downloaded. And then for the VAE, yes, I'm going to select the VAE that I just downloaded.

And that's pretty much it. So, let's go over how to run this. It's pretty straightforward. So, here is where you would choose the aspect ratio. Here are the options you can choose from. Here are the megapixels, so you can increase this for a higher resolution. Here is where you would enter a positive prompt or an optional negative prompt. Here's the CFG. So, this is like how literally you want the AI to follow your prompt.

If you find that it's not following everything you specified in the prompt, you could try increasing this value a bit more. Or conversely, if you want it to be more creative and do its own thing and introduce some variation, then you could decrease this value. Note that if you set the CFG to one, then the negative prompt won't actually work. You'll need to set the CFG to a slightly higher value for the negative prompt to take effect.

And then this is the number of steps it takes to generate your image. The more steps you have, the higher quality it will be, but it'll also take longer to generate and vice versa. And then here's the algorithm that's used to generate the image. The seed is like the unique ID of the image. So if you keep all the settings the same and you use the same prompt and the same seed, you're going to get the exact same image as before.

So if you want to use the same prompt but get a slightly different image, then you'll need to change up the seed. And that's pretty much it. So let's click run and see how long it takes for me to generate an image. Perfect. So, here's what we got. And if I pull up my window here, you can see that it only took like 29 seconds to generate this image. Note that I'm using an RTX 5000 ADA, which has 16 GB of VRAM, and I'm just running this on my laptop.

So, it's a pretty fast and lightweight model. So, that's text to image. Next, let's go over image editing. Again, I'm going to click on templates, and then at the top, search for Quinn image 2.1. And then here is the image edit option. So let's click on this and it's going to open up a new workflow like this. So same as before down here we need to load the models that we just downloaded. So for unit I'm going to select this one.

For clip I will select the W4 A8 one which I just downloaded. And then for VAE I'm going to select this one. Over here is where we can choose to upload different reference images for it to edit or refer to. And note that here you can input up to 10 images. Let's do a simple one. I'm just going to upload one image. So to remove the second image, I can either click on it and press Ctrl +B to bypass it. Or I can just click on it and delete this al together.

And if I need to add a new image, I can just double click anywhere on the interface and then type load image to insert a load image node. And then I just need to connect the output to over here. It's as simple as that. So again, I'm going to just use one input image for my demo. Here I'm going to upload this car image. And then here is where again we would set the positive prompt and the negative prompt. Here are the same settings as what we saw from the text to image workflow.

And then down here, I don't know why it's like hanging all the way down here. So let me drag this up a bit. Here is where you would select the aspect ratio and the resolution. But here we have this extra toggle called custom size. Right now it's set to false. So what'll happen is it'll actually ignore this and use the size of your input image. If you want to use this specified aspect ratio and size, then you'll need to turn this on.

All right. So now it's going to set my image at 1 one aspect ratio at 1 megapixel instead of this original resolution over here. Now for my positive prompt, let's do something simple like make this car drive down a windy forest road motion blur professional photo. Down here, all the settings are the same as before. And that's pretty much it. Let's click run. All right. And here's our result. If I pull up my stats, note that this took around 81 seconds.

I would say it's a bit slower than Flux Klein, which is the previous best edit model. All right, so that's how you can do image editing. If you want to supercharge your content creation, definitely check out Luma, the sponsor of this video. Instead of manually doing everything one prompt at a time, Luma agents can autonomously work with you across entire creative workflows. You can start with an idea and have Luma help you develop the concept, generate assets, experiment with different directions, and iterate on everything in one place.

For example, I can give Luma the initial idea, and then work with the agent to actually develop it step by step. It remembers the context of the project, so I can keep refining the results instead of starting over with a new prompt every time. You can access the best image generators and the best video generators out there. In fact, one of the most impressive ones is Ray 3.2, 2, Luma's latest video model. What really stands out about Ray 3.2 is its understanding of motion, physics, and 3D space.

As you can see, it can handle complex movement, camera motion, and interactions between objects while keeping the scene coherent. So, instead of just generating everything that looks good in a single frame, you can create genuinely cinematic sequences where everything actually moves through the scene in a believable way. If you're doing any type of content creation, Luma is basically like an AI co-pilot that can work with you across your creative workflows.

Try Luma today using the link in the description below or by scanning the QR code here. Next, let's pull up the third workflow. Again, I'm going to type Quinn image 2.1 and then let's select this remove background workflow. Now, once we start this again, it's going to say we have a few missing models. So, I'm going to click on each dropown and select the appropriate model. All right, after that's done, over here is where we would input an image which we want to remove the background of.

So, let me upload this image. And then here is where we would enter the positive prompt. So, let's write remove the background, keep only the woman, and output a PNG image. All the settings are the same as before. Again, here we have this custom size toggle. If we set this to false, it's going to use the same dimensions as our input image. If we set this to true, then it's going to use the width and height that we specified down here.

This time, let's set this to false. So, it's just going to use the original resolution. And then let's click run. All right, here's what we got. And this took around 116 seconds. So, by default, this is saved into my output folder. If I drag this into like a photo editing tool, you can see that indeed it gives me a transparent background. So, for example, let me change the background to a red. And here's what we get.

All right, so that's how you can remove the background of any image using Quentyn image 2.1. All right, so that sums up the three standard workflows that are available right now. Next, let's go over how to load loras to this. If you're not familiar with loras, these are like fine-tuned models which you can add on top of your workflow to generate a certain effect or character or action. For example, there's one for like grainy photography, there's one for phone photography, one to enhance the details of skin.

Now, for Quinn image 2.1, since we are really early to this, this has only been out for a day, there aren't lures available yet. The only one I can find so far is this one, Quen 2.1 anime consistency. And this is supposed to enhance character consistency during anime style image editing. So, just to show you a quick example of how to load a Laura to your workflow, I'm just going to download and try out this Laura. So, let's click download.

So, this goes in Comfy UI in models and then in Lauras. Let's click save. Now, it depends on the use case of your Laura. Here it says this one is for image editing. So, I'm going to pull up my image editing workflow again. And to load the Laura, what you need to do is actually click on this button at the top here to open up this workflow. And you need to place a load Laura node between this node and what comes after it.

So, let me double click anywhere on the interface and type in load Laura. And let's just select this one by Comfy and place it over here. And if we click on this dropown, here's where we would select the Laura that we just downloaded. Here is the strength of this. So how much influence you want the Laura to have on your image. Right now it's set at 100%. Let's set this to something like8. And then I just need to connect the output from this load diffusion model over here.

And then this output to over here. And that's pretty much it. So if we press on escape to go back to the original workflow here, let me upload an animate character. In fact, what I'm going to do is set the aspect ratio to 16 to9. Let's increase the megapixels to two. And for the prompt, let's write make a character reference sheet with the front, back, and side view of this character. And let's press run and see how good this Laura is at preserving the consistency of this character.

All right, here's what we got. Let me pull up the stats for you. So, this took 128 seconds. So, that's how you can load a Laura into this workflow. By the way, if you need to stack multiple lures together, then you can just click on this node and copy and paste it and then just string them together like this. Or there are also some other Laura stacker nodes like this which allow you to stack multiple Loras in the same node.

Now, like I said, for this quantized coin image model, you should be able to run this with like 8 GB of VRAM or potentially even lower. But what if you're poor as hell and you only have like 4 GB? Well, in that case, maybe this Conrot version might not even work. The good news is you can still run Quinn 2.1 using more compressed models called GGUFS. So there's already a ton of Quinn image 2.1 GUFS out there. I'm just going to show you this one by Abray here.

They released multiple compressed versions of this and the smallest one is only 3.19 GB in size. So if you have less than 4 GB of VRAM, I would suggest you download this one. Note that for these ones like Q6 and Q8, they're not actually significantly smaller than this int8 convert version. So if you have like 6 to 8 GB, I would actually recommend you try this Convert version first. But if you have like 4 GB of VRAM, then you can download one of these.

So I'm just going to show you an example with this Q3 one. Let's click download. And this goes in Comfy UI in models and then in unit. Let's click save. Afterwards, back into our workflow. You can use this GGUF for all three workflows. Text to image, image editing, or background removal. I'm just going to show you an example with text to image. What you need to do is at the top corner, click on this button and over here instead of load diffusion model, you need to replace it with a gguf loader.

So simply double click anywhere on the interface and then type unit loader and you should see this unit loader gf node. So let's insert this somewhere here. By the way, if you don't see this option, what you need to do is click on this extensions button at the top and then type in gguf and you'll need to install this one. Comfy gguf by city96. Afterwards, you should see this node. All right. So, after placing this node here, we just need to take this input and drag it here.

And then I'm going to hold down shift to take this output and drag it here. So, right now, this is going to load our GGF instead of the diffusion model. In fact, for this node, I can just click on it and then press Ctrl +B to bypass this model. And then, let me press escape to go back to our main workflow. Down here, for the unit name, instead of the main Quen image model, we need to click on this dropown and select the GGUF that we just downloaded.

And that's pretty much it. Let's click run. All right, so here's our result. As you can see, even the most quantized Q3 GGUF isn't too bad. You still get a similar quality to the full model. Finally, it's also important to mention the license of this. So here, at least for their official license agreement here, it says this cannot be used for any commercial purpose without obtaining a separate commercial license. Now, this has raised a lot of questions and concerns from the community.

But the good news is Kquen has replied back and they said that outputs, in other words, your image generations are not part of the licensed materials. So you retain the rights to the images and other content they generate using the model. So, at least from this post, it seems like you can use your generations even for commercial purposes. All right, so that sums up my review and tutorial of Quen Image 2.1. It's a really flexible tool, and I really love how they included the ability to create transparent images.

And I also love that we finally have a new image editing model. Let me know in the comments what you think of this. And if you run into any errors during the installation, welcome to copy and paste the exact error message in the comments below, and I'll try to help you troubleshoot as much as possible. As always, I will be on the lookout for the top AI news and tools to share with you. So, if you enjoyed this video, remember to like, share, subscribe, and stay tuned for more content.

Also, there's just so much happening in the world of AI every week, I can't possibly cover everything on my YouTube channel. So, to really stay uptod date with all that's going on in AI, be sure to subscribe to my free weekly newsletter. The link to that will be in the description below. Thanks for watching and I'll see you in the next one.