LTX 2.5 Full-Resolution Workflows: No Downscaling, Better Detail but Slower
nickinnov · reddit · 2026-08-16
The author shares full-resolution ComfyUI workflows for LTX 2.5 (Lightricks' open video generation model), avoiding the downscale-render-upscale technique used in official templates. Full-res versions preserve detail in image-to-video when action departs from first frame, and improve sea/sand detail in text-to-video. Performance: official template 8x8s+3x33s=163s, full-res text-to-video 8x33s=264s (rendering only). Author advises checking JSON files before running.
More from Multimodal
- Seedance 2.5 in CapCut: AI Video Workflow Breakthrough with Consistency and Editing — thetripathi58 · 2026-08-16
- MiniMax H3 First Impressions: Generating Manifest-Style Video Clips — dominic__612 · 2026-08-16
- Request: Recommended Tools and Params for Minimax H3 Character LoRA — Apart_Luck700 · 2026-08-16
- Full pipeline for building a $2M AI video production workflow revealed — EXM7777 · 2026-08-16
- Local AI music video generation workflow on RTX 3060 — No-Victory6860 · 2026-08-16
- MiniMax H3: Issue with poor face details on wide shots — ItsLukeHill · 2026-08-16