[LTX 2.3] Singularity: Cinematic Motion & Shot Control (Alpha)

The 10% Miracle. This LoRA is a comprehensive optimizer for LTX-2.3 Image-to-Video and First/Last Frame workflows. Even at only 11,500 steps (10% of training), it delivers a massive leap in physical coherence and cinematic quality.

🚀 What's New in This Version?

• 🎬 Shot Injection & Timing: Precise control over camera cuts and motion flows using the 0-X seconds format.

• 🖐️ Physical Correction: Significantly improved finger/toe stability and limb consistency during movement.

• 🗣️ Human Expression: Reduces "AI Stiffness." Characters feel alive with natural micro-expressions and optimized Chinese lip-sync.

• 📽️ Cinematic Lighting: Injected professional film-grade lighting and shadow contrast.

• 🚫 No More Subtitles: Effectively suppresses the hardcoded AI subtitles from the base model.

• ✨ Cross-Style Potential: Outstanding performance on Anime/Stylized conversion (even without specific anime training data).

🛠️ Usage Tips & Parameters

• Base Model: ltx-2.3-22b-distilled-1.1 (Works best with transformer-only fp8 scaled version)

• LoRA Weight: 0.8 - 1.0 (Start with 1.0)

• You can click on the link below to try it out directly. If the effect is good, you can deploy it locally

  https://www.runninghub.ai/post/2053064469290463233/?inviteCode=sdhs0trb

• Note: High-dynamic physics (extreme motion) is still being refined due to the early training stage.

📜 Prompting Framework (Singularity Protocol)

To unlock the full power of this LoRA, follow this structure: (English for visuals, Chinese for Timing/Dialogue)

1. Scene & Style: Cinematic Wuxia style, indoor dim lighting...

2. Action Timeline: 0-5s, young man looks down then raises his head...

3. Camera Timeline: 0-5s, tight close-up, slight handheld movement...

4. Dialogue: 0-5s，男子说：这都什么呀？ (Use Chinese here)

5. Technical: film grain, cinematic bokeh，无字幕

Full Example: Cinematic wuxia style, indoor dim lighting, mysterious mood. 0-10s, young man in ancient white robes looks down with a confused expression. 0-10s, tight close-up, static camera with slight handheld movement. Dark stone background, warm candlelight bokeh. 0-10s，男子说：这都什么呀？都没听说过呀。Voice: low, Pace: slow. Precise lip-sync, film grain, cinematic bokeh，无字幕。

❤️ Community Feedback

This is an Early Alpha (10%) release. I decided to release it now because the initial results were too good to keep hidden.

If you like this model:

1. Post your generations! I need to see how it performs across different workflows.

2. Leave a Review. Your feedback directly shapes the 100% full version training set.

3. Follow me for the upcoming full release!

If you have questions, feedback, or want to collaborate on AI video workflows, feel free to reach out:

• WeChat (微信): aigctyd

• QQ Group (QQ社群): 1058747239

• Email: [email protected]

I'm actively looking for community feedback to refine the full version of this LoRA. Let's push the boundaries of LTX2.3 together!