✨ Say goodbye to the tedious chore of splitting videos into pose images! LTX 2.5 Motion Sync is officially here, and it is a massive upgrade over LTX 2.3. This advanced workflow effortlessly transfers precise, lifelike motion from any source video directly onto your chosen characters in a single, fluid process! 🔄
⚠️ Heads up, creators: This setup handles realistic and 3D styles like a dream, but it may struggle with flat, 2D anime images.
🛠️ Changelog & Bug Fixes
Fixed in FlyV2.0: Better fidelity & motion adherence. Added a VideoStitch Node for comparison in Output & Original Input. Also added a Metadata Tracker node, for those who want to share their creations on CivitAI.
🎨 Why This Workflow Rules
Direct Motion Transfer: No more breaking down videos frame-by-frame.
Superior Precision: Noticeably cleaner and more accurate than older LTX 2.3 pipelines.
Hardware-Friendly Optimization: Built natively around the official, highly optimized distilled INT8 ConvRot architecture.
🧠 The Engine: Required Models
To make this magic happen, you will need to download and equip the following official models:
📁 LTX 2.5 Video VAE Conv BF16 — The core video VAE.
🎵 LTX 2.5 Audio VAE BF16 — Audio processor for complete sync.
🔤 Gemma 4 12B Text Encoder (Comfy INT8) — Smart prompting alignment.
🤖 LTX 2.5 22B Distilled Transformer (Comfy INT8) — The heavy-lifting generation model.
🔍 LTX 2.5 Latent Spatial Upscaler X2 — For crisp, high-fidelity output.
🎮 LTX 2.3 22B IC-LoRA Union Control (Ref 0.5) — The secret sauce for fluid control.
LTX 2.5 Huggingface repository - https://huggingface.co/Lightricks/LTX-2.5/tree/main
💬 Give it a Spin!
Take it for a test drive and push it to its limits! I would love to hear your feedback on the motion fidelity.
👇 Show off your motion transfers! Drop your original videos and your newly animated characters in the comments gallery below! 👇
Description
FAQ
Comments (3)
Quick with V2. Will share review & output.
When is the version for Minimax H3 coming out?
In the works. Minimax H3 needs dwpose while LTX reads frames with pixels as it is yet delivers accurate motion transfer, but with Minimax H3 motion adherence achieved till now is not accurate enough. Needs more testing. Probably a week or so.