0:00 Okay, guys, so today I'm showing you how, in two minutes, using your computer and ConfiUI with new Hi3D and MV Adapter, using one image, we can create a textured 3D model absolutely for free. Okay, guys, so today we're taking a look on Hynion 3D Flow together with MV Adapter. So, I mean, something amazing happened just recently. And you probably heard already that Huneon 2.0 is available for you to use in ConfiUI. So, basically free. You can create 3D models. And recently, there has released a new custom node for ConfiUI called ConfiI IG2MV, which is basically an MV adapter that we can use inside ConfiUI. So, big thanks to Hans Music, the guy who actually developed this. And we're going to use this together with Huneon 3D custom nodes and Huneon 3D models.
0:54 So flow itself looks like this. We actually add an image and then we create a 3D model. We create some multi views from the model and then using stable diffusion model we create textured version and then we apply the texture to the 3D model and here we are. So let's start with the installation. So as usual, I prepare short instruction in Notion for you to use with all the models, with the workflow, with all links that you might need. How I usually do the thing. So first thing you need to do is open the ConfiUI and let's drag and drop the workflow right here. So basically, what is going to happen is that it will open a flow and if there are some missing notes, it will actually not let you know. So there is a link to official workflow from the guy who developed...
1:41 the custom node for MV Adapter. And there is my edition where I did add a little bit of optimization so you can use it on low RAM. And I actually fixed it so it works in updated Conf UI. So first thing you do is you update Conf UI and then let's drag and drop. Okay, whenever you drag and drop it here, you might see something like this. So you might see that some custom notes are missing. So if that happened, it's very easy. Go to manager, go to install missing custom notes, and you will see something like this, where you can actually select all and press install. After the installation, it will actually suggest you to reboot. Please do that. And basically, that's it. So the last part is to try and run the flow.
2:27 So whenever you install all the nodes, then you go to the Conf UI and you might not see the errors when you open it. Then you can add an image and actually press run. But then something else might happen and you will see the window something like this, which means that you're missing some custom models or so. So you might have them installed or not. If not, just go to the Notion file and I actually explain. with all details what models you need to download and what folder you need to put. If that's done, you might need to go to nodes that will be marked red. And then you have to select the model here. So for example, you go here and you will select it like this. For your convenience, nodes with missing models will be framed.
3:15 with a red color. Okay, if you did all things correct, you must be able to run. And you can press in around two minutes, depending on your GPU, it will create a 3D model with textures. On my 4070 with 12 gigs, it takes two minutes approximately to create a model. So there is a strange case when for some users, and for me including, there was a problem with custom rasterizer. That happens in high 3D or in the multi-view node. So if you're facing a problem like that, please go check my Notion and I explain what helps me to do that. But my friends tried as well and they didn't face the same issue. It might happen on the Windows only.
4:01 Okay, guys, now let's see what actually this flow is about and what we can do. So there is two parts. So in the very beginning, we can see that there is an image. So the only thing you need to know is that the image should be a one-to-one ratio. So this workload, do not do that automatically. So just make sure when you add image here, it is like one-to-one ratio. The other thing here is that here you can set some constants about what image size we give to high 3D and what image size we give to our SDXL model that will create multi-views, right? The important thing here is that this one you can... leave 1K, that's totally fine. But this one will create six images. And if you have low RAM, it is good that you can maintain this value like 512 or maybe less. But if your VRAM is more than 12, like 24, you can try 1K or 768, which will give you better results on texture.
4:59 Okay, and one more small thing here is that this node actually creates an image to be used as reference for multi-view painting. And here you can set a background color. So the only rule here is that just select the color that will make your character distinguish from the background, right? So, I mean, basically, you can actually use this part individually. So you can actually bypass all the rest and you just... can create 3D models themselves. So this group is basically about this. A few things that you might tweak, you might need to tweak here, is that you can control actually the face amount. So amount of polygons can be controlled here. 50,000 faces, pretty much. If you need something like low pole, middle pole, you can reduce the value. Or if you want more, you can add. I can also smooth normals, etc.
5:52 And the second part is that we actually take a triangulated unwrap model, then we create with union 3D nodes, we also create multi-view with normal and position map, and then we use them as a reference together with the reference image. Using SDXL model, we render actually six views, and then these six views are used to... create a texture. So in this group, there's not much you actually can tweak but these settings. So here you can see Step, SFG, Reference, Condition, Scale. The steps is something you can tweak if your result is not decent because here in example, I use TurboModel and sometimes for some prompts, less amount of sales might give the better results. But again, like if you have quite good GPU, you can try more. And again, like you should play around the value between.
6:46 6 to 10 is actually something good. You can play around with SFG, but value around 2 gives best results. And of course, in my flow, their seed is fixed, but you can set it to randomize. So each time you rerun, it's like a slightly different result. Same actually goes for Hunyuan 3D mesh, but there I don't think like this actually works to tweak. And one more thing. So in the bottom, there are two nodes that are currently bypassed, but you can actually select them with control or press control and select them like this. Press bypass and you will activate the upscaler. This one is upscaler using model that actually takes the ready result and upscale them before texture is generated.
7:37 I bypass them by default when I test it. And whenever I'm happy with the result, I activate this node, rerun, and create the final model. About the outputs. So your Hunian 3D models are saved twice whenever the 3D mesh is generated. So you can go here and see that they are saved in 3D slash high 3D, which can be found easily if you go to your Confioi folder, go to Confioi, go to the output, and go to 3D. Here you can see all the models and they are saved on each run. Same happens when model is textured. It saves you 3D slash high 3D underscore textured. So don't get lost. If you need the models, take them from there. Okay, now when that's clear, let me tell you what more is possible with this flow. As you understand already, here we create a model and we texture model. And we actually can texture...
8:33 our own model or any other model or retexture the model if we don't like it. We can use not only image from, but also text from. So let me show you how it's done. So basically what we need to do is to disable this flow, right? You can press on group and you can press bypass group nodes, which will bypass it. And here is a node I prepared, but you can easily add it from high 3D. And there is a load mesh node, right? So this load mesh will actually use a GLB pass. So it can't be the model we generated previously. So for example, here is a model I generated just right now. I can actually copy as pass and I put it here. The only thing you need to do is to delete this comma in the end, in the very beginning. Press OK and that's it. So the another thing is that you need to know if this model is unwrapped or not. So as you can see here in this flow,
9:29 There is one more node. We can copy this node, add it here. If your mesh is not UV unwrapped, and as we can see, it is saved here before unwrapping, then we need to unwrap it here. Same goes for any other models that you get from elsewhere. So you can take GLB from elsewhere, add it here, unwrap it, and do texturing. So the reference image for this will be taken from here. Otherwise, if you don't want to use reference image, you can disconnect this node. from here, and you can actually use a text prompt, which is amazing, and you can experiment and play around with other options. Okay, guys, and in the end, I'll give you a few tips. If you have low VRAM, like 6 or 8 gigabytes, what things you can do to speed up the flow and make it work without memory.
10:24 So the first thing is to take a look on the image resolution we use here that is used to create six views in SDXL models. You can try 256. It will definitely reduce the VRAM peak values. Another thing is that if we go here, you can disable Upcast FP32. You can disable this option and you can enable via Tiling here. Also, The more steps you add here, the longer it will take and potentially increase the DRAM peaks. So if your computer is really weak, probably try to reduce steps. Yes, this will reduce the quality in some cases, but at least it will give you the option to play around with this flow. And of course, don't forget about the upscale. Upscale might take a lot of DRAM at some points, depending on the resolution of images you're creating.
11:18 So make sure to disable them if your VRAM is not good enough. And last but not least, very important tip, do VRAM cleanup after each run if you have low VRAM. So simply press right button on any free space and you will see cleanup of VRAM usage. You can alternatively press Shift-R and it will actually do the same thing. So it will... Actually clean up the VRAM and you will not see that your workflow is stuck or something happening. And one small tip in the end, when you run ConfiUI, you might know the terminal that is opened. And sometimes this one actually gives more information about what's happening, what's rendering, and what time it took than the ConfiUI if you didn't install a lot of custom nodes. So here you can get the idea how long it took for you to create images or mesh, etc.
