Testing LTX 2.5: Model Cannot Generate Specific Speech and Suffers from Template Bugs
Lost_Lab_739 · reddit · 2026-08-14
The author tested the LTX 2.5 video model to generate videos with specific spoken lines, only to find it produces background music or human-sounding gibberish instead of the intended text. A code review revealed that audio and video are both generated from a single text prompt, lacking any input node for specific scripts or transcripts.
During troubleshooting, two major issues were identified:
- Missing Node in Templates: The official ComfyUI templates are missing the LTXVModalityGuidance node, which couples audio with video and handles lip-sync. Without it, the model only outputs music.
- Broken Prompt Enhancer: The built-in prompt enhancer silently discards the original prompt and outputs garbage characters in some setups (likely due to fp16 overflow), feeding invalid input directly to the video model.
More from Multimodal
- Kling 4.0 fully available, shows markedly better motion coherence in fast-action scenes — alifcoder · 2026-10-16
- MJev-Spatial: Single-Pass Classification Beats Autoregressive VLMs on Spatial Reasoning — mindlab-research · 2026-10-16
- H-DDCM: Hadamard Codebooks Make Diffusion-Based Image and Video Compression Scalable — Technion · 2026-10-16
- ComfyUI v0.41.0 ships with Flux3 image model and MiniMax-H3 LightVAE support — Fcking_Chuck · 2026-10-16
- AI-generated short film 'The Dragon Rider' glides cinematically above the clouds — jijo6686 · 2026-10-16
- Realistic candid Midjourney portrait prompt: natural skin, lived-in bedroom light — tisch_eins · 2026-10-16