LTX-2.5 Launches With Multi-Shot Continuity, Native 4K HDR, and a RAW Finishing Workflow

LTX-2.5 generates connected shots that hold character, environment, lighting, and voice across cuts, then outputs native 4K HDR and a RAW workflow for professional color and finishing. The open-weights model runs on hardware with a stated 16GB VRAM minimum and ships a pretrained base checkpoint built for fine-tuning.

• Multi-shot in one pass. LTX says the model produces connected shots that keep character, environment, lighting, and voice consistent across cuts.
• Native 4K HDR and RAW. The model outputs high-resolution HDR footage and a RAW workflow meant for professional finishing without compromising the master.
• Runs on your own hardware. The open weights carry a stated 16GB VRAM minimum, a permissive license with no mandatory branding, and free use under $10M ARR.
• Built to fine-tune. LTX-2.5 includes a pretrained base checkpoint teams can adapt to their own data and deploy on their own infrastructure.
• New video decoder. LTX calls its decoder Diffusion Fidelity Rendering, which the company says allocates rendering compute by scene complexity.

Multi-shot generation targets continuity across cuts

The headline change is native multi-shot generation. According to LTX, the model creates connected shots that retain character, environment, lighting, and voice from one cut to the next, rather than treating each clip as an isolated generation.

LTX also lists stronger prompt adherence, saying the model follows complex creative instructions from shorter, simpler prompts, plus automatic duration selection that sets clip length to the requested action. A beta precise-editing mode edits real footage inside the same pipeline.

Multi-shot coherence has become a common target across video models; we covered Kling 3.0's native 4K and multi-shot generation.

We also covered PixVerse V6's multi-shot storytelling push.

4K HDR and RAW aim at finishing pipelines

LTX-2.5 generates native 4K HDR footage built for professional finishing, and adds a native RAW workflow so teams can generate and edit inside color and finishing pipelines without altering the master file. The model uses a Gemma 4 12B text encoder with a custom prompt enhancer, according to LTX's page.

That finishing focus tracks with other work bringing generated and low-bit footage into a high-bit-depth grade; we covered ETC's 16-bit HDR pipeline for AI video.

Open weights run on a 16GB VRAM minimum and are free under $10M ARR

LTX-2.5 ships as downloadable open weights that LTX says run on-prem, at the edge, or via API, with the pretrained checkpoint available for fine-tuning. The license is permissive, carries no mandatory branding, and is free under $10M ARR.

API pricing is billed per second of generated video: $0.09/sec at 720p, $0.15/sec at 1080p, $0.19/sec at 2K, and $0.37/sec at 4K. LTX reports that a self-hosted setup on two GB200 GPUs generates a 10-second 720p clip in 6.8 seconds, which the company states is faster than real time. That performance figure is an LTX measurement and has not been independently verified. Open weights with fine-tuning put LTX-2.5 alongside other self-hostable models we covered, including MiniMax H3.