How I fixed skin compression and texture artifacts in LTX‑2.3 (ComfyUI official workflow only)

A significant breakthrough for users of the LTX-2.3 video generation model in ComfyUI has emerged from the community, solving widespread issues of skin compression, muddy textures, and blocky details. The solution, discovered by a user and shared on Reddit, debunks the common advice of switching models or installing extra nodes. Instead, it involves three precise adjustments within the official ComfyUI workflow itself, proving the problem was rooted in interpolation and sampler choices rather than the core model.

The first two changes target the resizing and intermediate upscaling steps. By setting both the 'Resize Image/Mask' and 'Upscale Image By' nodes to use 'Nearest (Exact)' interpolation, users prevent the early introduction of blurring or smoothing that LTX-2.3 later amplifies into visible artifacts. This directly combats the 'plastic skin' effect by keeping edges clean. The most impactful fix, however, is in the final 2x upscale stage: switching the sampler from 'Gradient estimation' to 'Euler_CFG_PP'. This change prevents the smearing of micro-details and stops the exaggeration of compression on darker skin tones, resulting in clean, stable, and natural-looking final frames.

This discovery is a major efficiency win for AI video creators. It streamlines the troubleshooting process, saving hours previously spent experimenting with alternative VAEs, models, or custom node packs. By providing a clear, reproducible fix that maintains the integrity of the official workflow, it lowers the barrier to achieving high-quality, artifact-free video generation with LTX-2.3, making advanced AI video more accessible and reliable.