Editorial illustration for LTX-2.5 Creates 10-Second AI Video From Image in 6.8 Seconds
LTX-2.5 Generates 10-Second AI Video in 6.8 Seconds
Lightricks' open-weights offshoot LTX released LTX-2.5 today, a video and "world" model that turns a still image into a 10-second clip in 6.8 seconds when running on Nvidia's newest superchips. The model launches natively inside ComfyUI, the node-based tool that most people building open generative media pipelines already use, through a day-one partnership between LTX and ComfyUI's team. Weights are posted on Hugging Face, and the model is also reachable through ComfyUI directly or the LTX API for teams that want someone else to handle the infrastructure. Pricing is straightforward: free for any organization pulling in under $10 million in annual recurring revenue, with bigger companies expected to negotiate a license.
LTX says its model family has now cleared 33 million downloads, a number the company points to as proof it's the most-used open world model line around. VentureBeat spoke with LTX co-founder and CEO Zeev Farbman and ComfyUI co-founder and CEO Yoland Yan ahead of the release, covering what changed in this version, why the two companies chose to integrate so tightly, and their shared wager that open weights beat closed APIs in the long run for video and world modeling.
"With LLMs, the surface area of the API is pretty narrow, we're typically asking some kind of question, passing words and getting words back," Farbman said. "With video models, world models, there are so many different use cases that require people to get access to the weights and create flows that really work for them."
Why this matters
We've watched the open-weights video model race turn into a speed contest, and LTX-2.5 pushes that further: 6.8 seconds to render a 10-second clip is fast enough to change how prototyping actually feels, not just how a benchmark reads. The ComfyUI integration matters more than the raw number, though. LTX is betting that distribution through an existing workflow tool, one built by a pseudonymous developer back in January 2023 and now the default sandbox for open generative media, beats trying to pull users into a new interface.
That's a smart read on where the actual users are. For developers and founders, the open weights plus API combo means you can prototype inside ComfyUI for free and then scale through LTX's hosted service without switching stacks, which lowers the cost of testing ideas before committing engineering time. Researchers should watch whether "world model" claims hold up under scrutiny once people start poking at consistency and physical plausibility across longer sequences.
Speed is easy to demo. Whether LTX-2.5 actually understands the scenes it's generating is the harder question, and one worth testing directly rather than taking on faith.
Related Reading
Open Source D&B rebuilds 642 million‑business database after AI agents hit limits
Open Source OpenAI's GPT Transcribe Cuts Error Rate to 3.31%, Improving on GPT-4o
Open Source Study: 6.7% of Deepfake Requests on Hugging Face Targeted Children
From Archives NVIDIA's Nemotron 3.5 Lightning Model Speeds Up Agent Tasks by 30%
From Archives webAI Releases TwIL-LM Logic Models for Autoformalization on Local Hardware
Common Questions Answered
How fast can LTX-2.5 generate a 10-second AI video from a still image?
LTX-2.5 can create a 10-second video clip from a static image in just 6.8 seconds when running on Nvidia's newest superchips. This speed represents a significant advancement in the open-weights video model race, making it fast enough to meaningfully change how prototyping feels rather than just improving benchmark numbers.
What is the primary distribution method for LTX-2.5 and why does it matter?
LTX-2.5 launches natively inside ComfyUI through a day-one partnership, integrating directly into the node-based tool that most open generative media builders already use. This distribution strategy through an existing workflow tool is considered more important than the raw rendering speed, as it enables developers to create custom flows that work for their specific use cases.
Where can developers access LTX-2.5's model weights?
The model weights for LTX-2.5 are posted on Hugging Face and can be accessed through multiple channels including ComfyUI directly, the LTX API, or by running the model locally. This open-weights approach allows developers to integrate the model into their own generative media pipelines and workflows.
Why does LTX believe open-weights video models need different API approaches than language models?
According to LTX's perspective, language models have a narrow API surface where users typically ask questions and receive text responses, but video and world models have vastly different use cases requiring developers to access the weights directly. This difference means creators need the ability to build custom flows that specifically work for their unique requirements rather than relying on a standardized interface.
Further Reading
- Lightricks Releases LTX-2: The First Complete Open-Source AI Video Foundation Model - PR Newswire
- LTX | Open Foundation Models for Video, Audio, and World Simulation - LTX
- AI image to video model - LTX - LTX
- LTX-Video - Official repository for LTX-Video - GitHub
- LTX-Video - Hugging Face model card - Hugging Face
