@vanman_1000@ingi_erlingsson Yeah, we have a compressed latent. You want fast paced? Use 60 fps for generating more frames by paying more tokens. Let me know how it does for you.
New IC-LoRA to use with @ltx_model I trained this one with Lightness maps. So you can use it to drive luminance and light sources in your generated videos. I find it especially fun to use for Audio Reactive workflows.
Grab it here : https://t.co/XWs5t3vg4M
ComfyUI LTX-2 VRAM Memory Management.
- reduces VRAM usage for LTX-2:
- possible to gen 900+ frame vids with 24GB GPU;
- chunks FeedForward layers to cut peak VRAM, zero quality loss;
- multi-GPU.
https://t.co/MytXTAYVj3
My read on LTX-2 is different from the technical coverage Iโm seeing.
This is a $1.8B company that just gave away their crown jewels. And it might be the smartest thing theyve done.
Hereโs the math on why open-sourcing makes strategic sense.
Lightricks has 6.6 million paying subscribers generating roughly $80M in annual revenue from apps like Facetune. Their core business model is consumer subscriptions on mobile.
But the AI video market is consolidating FAST.
Sora, Kling, Veo, Runwayโฆ all closed systems with proprietary weights. The big players are racing to lock in developers and enterprise customers through API contracts.
Lightricks canโt win that game at their scale.
So they flipped the table.
By open-sourcing LTX-2 with full training code, theyโre making a bet on ecosystem dynamics. If ComfyUI workflows and indie developers standardize on LTX-2, Lightricks becomes the de facto foundation layer for audiovisual generation.
The real product becomes LTX Studio. Which charges for the UI, collaboration features, and enterprise support.
Itโs the Red Hat playbook. Give away the engine. Sell the gas station.
What makes this interesting is the timing. They dropped this at CES 2026 alongside NVIDIA optimizations that cut model size by 30% and double inference speed on consumer RTX cards.
That NVFP8 quantization means a 19B parameter model runs on hardware people already own.
Compare that to Sora, which still has limited access and undisclosed pricing. Or Kling, which requires cloud API calls.
The constraints are flipping. Access used to be the bottleneck. Now compute is.
And Lightricks just positioned themselves on the right side of that shift.
The failure mode here is if the community fragments. If developers choose other open models or the proprietary players drop prices agressively enough to make local compute irrelevant.
But my guess is thatโs a 2-3 year window minimum. And in that window, Lightricks has a shot at becoming the default standard.
The 18x faster inference stat buried in the paper matters more than the 4K resolution. Speed determines iteration cycles. Iteration cycles determine adoption.
This is a product strategy disguised as a research paper.
Training LTX-2 is the real deal and a big deal ๐
The model is open for you to customize it, to create exactly what you wanted, not just what it already knows. But unlike heavier models, this one is blazing fast in training as much as inference. Get your dataset ready and TRAIN!!
A fun Tear FX LoRA I trained on the recently open sourced LTX-2 by @LTXStudio
If you want to tear some object, grab the LoRA here :
https://t.co/ESo7yEJZJG
LTX-2 provides unprecedent control and flexibility comparing to anything else in the space of video models today
It was lots of fun working with comfy on this one, now the fun is all for you ๐ซต
We introduced a proper way to deal with audio-visual generations in ComfyUI, to give you all that you needed in Core, but also lots of other controls and capabilities in our own https://t.co/rI6MRiYCKG
LTX-2 is natively supported in ComfyUI on Day 0 ๐ฌ๐
The next chapter in controllability for open-source video generation.
- Open-source audio-video foundation model
- Generates motion, dialogue, SFX, and music together
- Canny, Depth & Pose video-to-video control
- Keyframe-driven generation
- Native upscaling and prompt enhancement
Everything in a single pass.
LTX-2 from @lightricks is now live in ComfyUI!
Video generation in 4K / 50 FPS, with synchronized audio!
Lightening-fast generation speed.
Available now:
LTX-2 (Pro) and LTX-2 (Fast)
Text-to-Video & Image-to-Video
Check more examples ๐งต
@EddyLeeKhane@ltx_model TBH I would start with switching from cog to LTXV 13B 0.9.8 to get both speed and higher quality.
LTX-2 is going to be possible to run in many settings but we are counting on the great people of the open source community to build the right quantization and tools for it.
Yet again a bunch of stuff we were cooking, coming at you.
The big thing here - long coherent videos. Tile them as much as you want to make longer and longer ones.
But there's also much more inside:
* A detailer LoRA that can make crisp 1080p and even more.
* A better 2B model for blazing fast results.
* A future where video generation is done without limits of space and time.
p.s. If you're excited about such things and want to join our team, DM
๐ Long shot generation in LTXV-13B
Now supports up to 60 seconds of video!
โข Auto-regressive generation (up to 60s)
โข Standard generation (up to 20s)
โข Streamable on H100 with low latency
โข Compatible with control LoRAs (released last week)
โข Time-varying prompts supported
It's a good idea in general only that it will be a lower bound.
If you use fp8 it's less memory, so you need a model column too. If you use different attention implementations it will be different.
We will try to be more comprehensive, but I suggest you join our discord community (https://t.co/C36Vz6Dkiy) where lots of people are running on different cards.