Lightricks, a leader in consumer-friendly creative software, has achieved a major milestone in generative AI: its open-source LTX Video (LTXV) model now supports 60-second, real-time controllable video generation. This marks a dramatic shift from prior eight-second limits, ushering in a new era of interactive AI video, where creators precisely guide their content as it unfolds.
🔑 Key Highlights at a Glance
- Full 60‑second video length: A six-fold increase enabled by updated model architecture.
- Real-time prompt control: Add or adjust inputs (e.g. pose, depth, style) mid-generation.
- Streaming output: First second appears in under one second; subsequent frames update continuously.
- Open source, hardware-friendly: 2B and 13B versions run on consumer GPUs; code and weights available via GitHub and Hugging Face.
What Makes This a Generative AI Breakthrough?
1. Extended Duration—Why It Matters
Most AI-generated videos cap out at a few seconds—rarely enabling coherent storytelling. Lightricks’ update shatters this limitation: 60 seconds of cohesive, fluid, creative video lets creators build narratives, character arcs, and even interactive sequences.
2. Real-Time Streaming Output
The new model uses an autoregressive, streaming-first architecture—where chunks of frames are generated based on earlier segments. Users see the beginning instantly (~1 second) and can direct the unfolding video in real time.
3. On-the-Fly User Control with LoRA
LTXV supports IC‑LoRA controls (e.g. pose, edges, depth) that can be adjusted mid-generation. This allows creators to refine content dynamically, blending technical flexibility with real-time artistic expression.
4. Multiscale Rendering in the 13B Model
The LTXV‑13B model uses multiscale rendering, where rough motion is captured first and then refined layer by layer—like an artist refining pencil sketches into full-color animation—with no compromise on speed.
5. Accessible, Open-Source, and Efficient
All LTXV versions, including 13B and mobile-friendly 2B, are open-source. They run on a single H100 GPU—or even consumer cards like RTX 4090 with ~8 GB VRAM—without requiring costly GPU clusters.
🚀 How It Works: Under the Hood
A. Autoregressive Chunked Generation
The model processes videos in overlapping frame “chunks.” Each chunk is generated based on the preceding one, ensuring smooth progression. Like a novelist building chapter by chapter, LTXV maintains motion continuity and narrative cohesion.
B. Multiscale Rendering Architecture
LTXV-13B begins with a low-resolution pass to establish structure and motion, then progressively refines details and color. This ensures film-like quality—layer by layer—without compromising generation speed.
C. LoRA Integration for Fine Control
Integrated Low-Rank Adaptation (LoRA), especially IC‑LoRA, provides real-time control during streaming. Users can adjust attributes like depth, pose, and edges to influence the evolving scene.
D. Hardware Efficiency
Thanks to kernel optimization and precision quantization, LTXV-13B runs 30× faster than comparable models and can generate 5 seconds of 24 FPS video in ~4 seconds on H100—outpacing many alternatives.
🎞 Practical Applications: Beyond the Clip
1. Filmmaking & Animation
Create fully directed scenes—character dialogue, action sequences, stylized cutscenes. The combination of length, rendering fidelity, and interactive control makes LTXV a tool for narrative filmmakers and animation studios.
2. Live & Interactive Media
Integrate real-time visuals in live performances and AR. Artists can paint scenes on-the-fly, performers can trigger visual effects mid-stream, and presenters can tailor visuals in response to audience feedback.
3. Gaming & Cutscenes
Generate in-game cutscenes on demand, synchronized with player actions or events. Dynamic scene control ensures narrative alignment and immersive coherence.
4. Education & Training
Develop adaptive video modules that evolve based on learner input—ideal for interactive training, branching education paths, and personalized e-learning content.
5. Marketing & Advertising
Produce vertical format ads of 15–60 seconds tailored in real time. Streamline production workflows and enable dynamic campaign adjustments.
🧩 suitable Technology & Open Access
Lightricks & Model Lineup
- LTXV-13B: flagship model with 13B parameters, multiscale rendering, IC‑LoRA.
- LTXV-2B (distilled): mobile-efficient, still offering real-time responsiveness and controllability.
- Distilled versions reduce VRAM needs while retaining fidelity, enabling workflows on consumer hardware.
Training Data & Ethics
Lightricks trained its models using licensed assets from Getty and Shutterstock, ensuring visual quality and legal safety.
Community Tools & Workflows
Lightricks provides support for ComfyUI, LTX-Video-Trainer, and Hugging Face pipelines. The 0.9.8 release added control models for pose, depth, and canny edge fine-tuning.
✅ Why This Breaks the Mold
| Feature | Impact |
| 60-second real-time length | Enables storytelling, dialogue, and interactive media creation |
| Autoregressive streaming | Live, flowing narratives without stutter or abrupt transitions |
| LoRA-based control | Live art direction: refine motion and style as the video unfolds |
| Open-source & hardware-efficient | Democratizes generative video for individuals and small teams |
This evolution transforms AI video from short demos into directed storytelling tools, offering creators unprecedented creative freedom.
🔜 Looking Ahead: What’s Next?
- LTX Studio integration: Streamlined workflows, enterprise licensing options, and cloud deployment upcoming.
- Community innovation: Open‑source release supports plugins, community features, and non-commercial extensions in AR, live streaming, and more.
- New content formats: Experiments in interactive film, AR performance, live education, and AI-driven gaming cinematic experiences are already underway.
🎯 Final Takeaway
Lightricks’ update to 60‑second, real-time, user‑directed AI video signals a seismic shift: generative video is evolving from novelty to narrative tool. Accessible, interactive, and high-quality—LTXV empowers creators, startups, educators, and developers with cinematic potential previously out of reach.
If you’re exploring AI-driven visual content, now is the moment to experiment with LTXV. This isn’t just about AI generating clips—it’s about AI empowering storytelling in real time, for everyone.
🔍 FAQ
What is LTXV?
LTX Video (LTXV) is an open-source AI video generation model by Lightricks, now capable of producing 60-second, real-time directed videos with interactive editing controls.
Can I control the video as it’s being generated?
Yes—with real-time LoRA-based controls, you can adjust pose, depth, and edges on the fly while the video generates.
Do I need expensive hardware?
No—a single H100 GPU or a consumer GPU like RTX 4090 with ~8 GB VRAM can run the 2B or even distilled 13B model efficiently.
Is it free?
Yes—all LTXV models, tools, and weights are open-source and free under Lightricks’ open‑weights license via GitHub and Hugging Face.