LTX-2.5 Open-Source Video Model Rebuilt to Cut Retries and Artifacts
aigleeson · x · 2026-08-12
Lightricks has released LTX-2.5, an open-source video world model that overhauls the generation pipeline to minimize prompt retries, broken textures, and inconsistent shots.
Core Pipeline Design:
- Instruction Parsing: Uses a custom Gemma 4 12B backbone to read prompts, paired with a separate Gemma 4 E2B prompt enhancer to preserve the user's original intent.
- Duration Head: Introduces a duration head that predicts the optimal video length based on the prompt before diffusion begins.
- Native Multishot: Supports generating multiple camera shots within a single continuous sequence while maintaining consistent characters, environments, and voices across cuts.
- Video Decoder: Employs a diffusion video decoder to convert results into visible frames, improving texture reconstruction and reducing artifacts in high-motion scenes.
Training & Availability:
Trained on a larger dataset with RL post-training based on human preferences to improve first-generation quality. The entire pipeline is open for builders, supporting local execution, fine-tuning, on-prem deployment, and integration with ComfyUI and Diffusers.
Related event: Lightricks Releases Open-Source Video Model LTX-2.5(16 posts)→
More from Models
- GPT and Claude Settle a 25-Year-Old Information Theory Problem — weijie444 · 2026-08-12
- Enterprise AI Shift to Specialized Small Models: Generic LLMs Waste 99% of Compute — blaizedsouza · 2026-08-12
- Old Mistral Model Resurfaces: Illegible CoT Seamlessly Transitions to Clear Responses — aiamblichus · 2026-08-12
- Bizarre ChatGPT Bug: Sends Unsolicited Notification and Prompts Itself — sennepo · 2026-08-12
- Anthropic to Embed Invisible Watermarks in Generated Text to Comply with EU AI Act — EricBuess · 2026-08-12
- Local Benchmark: Muse Glimmer 30B vs Qwen 3.6 27B vs Gemma4 31B — WonderRico · 2026-08-12