Run LTX 2.3 WITHOUT ComfyUI on 8GB VRAM! Full Local Setup 🔥

Indexé : 2026-05-16

[00:00:00]A lot of you probably started your local AI journey full of excitement, only to end up staring at a terminal window spitting out red error codes. It's a familiar story, right? But by the time you finally get ComfyUI to launch, you're way too drained to actually create anything. But there's a better way. Imagine a local generative AI platform built directly on the ComfyUI architecture that just works. It's 100% unrestricted, runs entirely on your own hardware, and comes preloaded with all the top-tier local models you actually need. We're talking about a smart prompt builder to perfectly dial in your vision, the ability to stack multiple LoRAs, and absolutely zero credit limits. You own it for life. Every single update and new feature is yours without a monthly or yearly subscription fee ever sniffing your bank account.

[00:00:45]Once you get inside, the very first thing you'll notice is the interface. It doesn't feel like you're staring into a chaotic web of spaghetti nodes. It actually feels like you're using a high-end polished cloud platform. It's clean. It's incredibly intuitive. And everything is exactly where you'd expect it to be. You get that premium professional aesthetic, but with the total peace of mind knowing every single operation is happening right there on your own hardware.

[00:01:10]The beating heart of the app is this massive library of presets. It's a literal goldmine of preconfigured models, whether you're looking to generate stunning high-res images or completely push the limits of AI video.

[00:01:21]But here's the best part. It's a living platform. They aren't just building another app. They're building a tool based heavily on our actual needs. So definitely jump into their Discord server and tell them what you want to see next because they are clearly paying attention.

[00:01:36]The generation process is super straightforward. First off, download the preset you want from the preset store. I highly recommend grabbing the Q5KM preset if you're rocking 8 gigs of VRAM and 32 gigs of system RAM. It'll throw a warning saying the preset is too heavy for your GPU, but you can run it without breaking a sweat if you have that 32 gigs of system RAM backing up your 8 GB VRAM.

[00:01:59]I stress tested it, as you can see right here, and there wasn't any significant SSD swapping or offloading during the sampling process. After downloading the preset, just click on the SimplyGen logo up here, select the video option, and then lock in your LTX 2.3 GGUF preset.

[00:02:17]Now, you've got three options here. You have the first to last frame option to give you way more granular control over your image to video generations. Then, you've got your standard image to video and text to video generation modes.

[00:02:30]Over here is where you access your past generations. These are a few of the clips I threw together using the Q5KM version. Check these out.

[00:02:39]Hi. Turn your rig into an AI powerhouse.

[00:02:44]Don't miss the next workflow. Hit subscribe and join the crew.

[00:02:53]I rode a thousand miles just to wash the blood off these hands.

[00:02:58]Turns out it's baked right into the soil.

[00:03:04]Structure confirmed at grid 309er. No movement for 40 minutes. Are you set?

[00:03:09]The cold doesn't care about the objective. It waits for you to lose focus. I am set. Let's keep wait You don't believe we can reach the deadline.

[00:03:18]You think I'm underestimating the complexity. I know you can reach it.

[00:03:23]I'm just making sure you understand the consequence.

[00:03:27]They said we couldn't build on this scale. They didn't think it was possible.

[00:03:31]Well, they were wrong because tonight the impossible becomes inevit All right, let's talk about the prompt builder.

[00:03:38]Prompting is typically a massive headache when you're dealing with LTX 2.3. You normally have to micromanage every single tiny detail you include in the prompt. But, here, you literally just type what you want and jump straight into the generation. First, you define the subject and the action, and then you can easily select the type of shot and camera angle right from this drop-down menu. Down below, you set the environment, and next to that, you have the option to enable character dialogue and select the delivery style from the drop-down.

[00:04:08]We can describe the lighting and the overall mood in the very next section.

[00:04:11]You've got some pre-made styles here to choose from, and further down you can dial in the specific atmosphere and mood. The next section is for selecting your aspect ratio, and you get four solid options here. Down below that, you choose the resolution. Now, full disclosure, I really don't recommend using 480p with LTX unless you're strictly generating tight close-ups. For text to video, you can actually push 1080p to generate up to 15 second clips, even if you're bottlenecked by an 8GB VRAM card. But when it comes to first and last frame-based image to video generation, stick with 720p since it demands significantly more processing power.

[00:04:49]Then right here we can set the duration of the clip, and down at the bottom, you have the option to stack up to five different LoRAs.

[00:04:56]Now, let's look at a prompt I actually used for one of my own generations. Just click the reuse settings icon right here to load the prompt up. Here you can see the exact settings I used to generate that specific clip. If you want to dig into the details of your generation, just click this icon. On the top, you have the basic prompt you entered, and down below is the final assembled prompt the system put together. After scrolling down, you'll see your generation settings, and at the very bottom, you've got all the technical details.

[00:05:25]Let's jump over to the local tab. Right here is where you have the option to stop or start the app itself. If you run into any errors during the process, this is where you'll find the logs, which you can easily copy and send over to Simply Gen's Discord server for quick troubleshooting.

[00:05:40]At the bottom, you have your installation path, models directory, and output directory. Now, if you're already a ComfyUI veteran and have your models downloaded, you can just move them over from your Comfy UI directory or simply change the path right here in the settings. I really only recommend doing that if you're an advanced user though, because it can absolutely throw compatibility issues at you if you don't map it properly.

[00:06:03]Right above that, you have some advanced settings if you're someone who needs absolute granular control over your generations.

[00:06:11]Just a quick heads-up, you will need a Hugging Face token to automatically download these models. Just click the add key option here and this little menu will pop up. Click on it and it will shoot you straight over to Hugging Face.

[00:06:23]Make sure you're logged in and then hit the create new token option. Select the read option right here, type in a name for your token and click on create token. It'll generate your token, but keep in mind you can only see this once, so make absolutely sure to save it somewhere safe. Then, head back to the app, paste your token in this box and hit save.

[00:06:44]For the record, I tested this app on CUDA 13.2 with the latest Nvidia drivers installed, so make sure your system is fully up to date before diving in.

[00:06:54]Now, let's talk pricing. This is just a one-time purchase of 30 bucks and you can grab it over on Gumroad. One-time purchase, that's the key phrase here.

[00:07:03]You pay that initial fee and after that, everything is entirely free, unlimited and fully unfiltered. It's basically the exact opposite of getting bled dry by cloud subscriptions.

[00:07:14]So, the big question, does this replace Comfy UI?

[00:07:18]Honestly, I don't think so. Comfy UI is our beloved playground. It's where we build our own workflows, run wild experiments and squeeze every single drop of performance out of our hardware and nothing is going to replace that feeling. But, the honest truth is that Comfy UI isn't for everyone. It's definitely not for casual users who just want to see their creativity come to life without spending hours troubleshooting after a long, busy day of work. So, here is the perfect solution for users who just genuinely value simplicity and their own free time. See you in the next video.