Summary
Veteran AI presents a detailed tutorial on using the ‘Edit Anything’ LoRA with LTX Video 2.3 inside ComfyUI to perform AI-powered video editing. The LoRA, available on CivitAI and trained on 8,000 samples with ongoing updates, supports four prompt modes — Add, Remove, Replace, and Transform — allowing users to insert new objects into existing footage, delete elements, swap content, or apply full style transformations.
The tutorial walks through a custom ComfyUI workflow that improves on the official CivitAI template by adding a two-stage upscaling pass for better output quality. Key technical details include loading the LTX 2.3 base model from Kijai’s frequently updated releases, attaching the Edit Anything LoRA, using the ‘Add Guide Multi’ node to condition generation on the reference video, and a specific fix for the upscaling stage: rather than using the original reference image, the workflow extracts the first frame of the first-pass generated video to avoid mismatches caused by visual changes introduced during editing. CFG settings are discussed, with a default of 1 suitable for most cases and higher values recommended when prompt adherence is weak.
Worked examples demonstrate clothing replacement on a dancer, object removal with convincing background infill, and the relative difficulty of each mode — Replace being most reliable, Add being most demanding. RunningHub is highlighted as a cloud platform for running these workflows without local GPU hardware.
📺 Source: Veteran AI · Published April 21, 2026
🏷️ Format: Tutorial Demo
