Video style transfer, built on LTX-2.5
Apply a consistent look to a clip while preserving structure, identity, and motion. Use a pre-trained transform like day-to-night or add-water, or train a custom IC-LoRA for a branded look. Open weights, self-hosted.
.svg)
Key capabilities
Structure-preserving video-to-video
Reference-conditioned IC-LoRA transforms the look while subject, framing, and motion stay locked to the source.
Pre-trained transforms
Day-to-night relighting, add-water VFX, and colorization are ready to run. No training required.
Relight without a reshoot
Change time-of-day from daylight to night using a reference-conditioned transform. The scene structure and motion stay intact while the lighting changes.
Add environmental VFX
Introduce water, rain, floods, or splashes to a dry plate. The transform holds the original composition while layering in the environmental effect.
Apply a branded look at scale
Train a custom IC-LoRA on your proprietary style and apply it consistently across a campaign or sequence.
Restyle with structural control
Combine style transfer with Union Control using Canny edges, depth, or pose to restyle while holding compositional structure.
Built for
VFX and finishing teams applying a consistent look across shots without reshoots. Generative AI leads training branded, reusable styles into a pipeline.
Product teams, AI startups, and developers building AI-powered video features. Add production-grade video generation as a product capability, not a research project. One API, production-ready results, and no custom orchestration.
Brands, agencies, and creative teams producing high volumes of content. Turn existing assets into video at scale. Faster iteration, lower production cost, and more output from what you already have.
Teams that require full control over deployment and data. Run video generation in your own environment. On-premises, no cloud dependency, and full infrastructure ownership.
Platforms powering creative tools with multiple AI models. Upgrade your video output with a best-in-class engine. Improve generation quality, retain users, and differentiate with a model built for production, not prototypes.
How it works
Input
Technical characteristics:
- Reference clip — VAE-encoded, downscale factor 1 for these transforms.
- Dual-panel prompt — Includes the transform trigger (e.g. "ADD WATER", relight-to-night, or "COLORIZE"); the reference drives structure, the prompt drives the look.
- Base model and LoRA — LTX-2.3-22B plus the relevant IC-LoRA, or train a custom IC-LoRA on paired reference-to-target clips with ltx-trainer.
Output
Technical characteristics:
- Transformed clip — Matches the source structure and motion; only the targeted look changes.
- Recipe — Use the stage-1-only native-resolution recipe for identity-safe results.
- Self-host — Run via ComfyUI or ltx_pipelines.ic_lora on GPU, under the LTX-2 Community License.
What users are building with LTX
FAQs
What transforms are available out of the box?
Day-to-night relighting, add-water VFX, and colorization. Restoration transforms like decompression and deblur use the same IC-LoRA framework; see the restoration page for those.
Can I apply my own style?
Yes. Train a custom IC-LoRA on paired reference-to-target clips with ltx-trainer.
Does it preserve the original motion and structure?
Yes. These are reference-conditioned video-to-video transforms. Subject, framing, and motion stay locked; only the targeted look changes.
How do I run it?
Open weights on HuggingFace via ComfyUI / ltx_pipelines.ic_lora. Use the stage-1-only native-resolution recipe for identity-safe results. LTX-2 Community License.
