LTX-2.3 long-form lessons: CFG=1 kills negative prompts, brightness drifts per clip
Chance-Thanks-1278 · reddit · 2026-11-19
A creator building a 20-second-per-clip AI music video with LTX-2.3 + ComfyUI (RTX 4060 8GB, 576×1024, 24FPS) shares lessons and a free community toolkit with a workflow and an experimental lighting-correction node.
Unwanted subtitles/text: negative prompts (text, subtitles, watermark...) appear to do nothing because both CFGGuider stages run at CFG=1, where the negative conditioning branch is effectively ignored; check your CFG config, and consider LTX2 NAG from ComfyUI-KJNodes as alternative negative guidance (untested by the author).
Continuity: using each clip's final frame as the next clip's start image, plus locked camera, no cuts/zoom, accurate lip-sync and consistent-appearance prompting.
Progressive darkening: every reused final frame gets slightly darker, visibly darkening faces after several segments — the motivation for the experimental lighting correction node.
More from Multimodal
- HyperFlow: 14B distilled MiniMax H3 runs video gen on 6GB VRAM at 2x speed — cgpixel23 · 2026-11-19
- One Neon prompt turns a monstera plant into a walking, waving character — fofrAI · 2026-11-19
- Kezek splits mono two-person conversations into per-speaker tracks for full-duplex dialogue training data — manifestro · 2026-11-19
- 6-minute Hubble documentary made with Claude and Jimeng CLI workflows — xiaohu · 2026-11-19
- Seedance 2.5 4K hands-on: AI video finally loses its plastic look — xiaohu · 2026-11-19
- Eerie AI-generated portraits: neon grimace faces and glowing-eyed shadows — azed_ai · 2026-11-19