Training a DLSSNR cartoon NPR rendering style: new lines data visibly closer to target
bdsqlsz · x · 2026-10-13
bdsqlsz shared progress on training a DLSSNR cartoon (NPR) rendering style. The comparison image shows the previous unit, a new unit trained with line data, and the unchanged NPR target — the new unlit+lines version is visibly closer to the target style in hair line detail and overall texture.
More from Multimodal
- Creator claims breakthrough in character consistency using MiniMax H3 with custom LoRA and masking — linoy_tsaban · 2026-10-13
- Google's VIS-Ground adds contextual grounding to interactive video storytelling — GoogIe · 2026-10-13
- Scale AI's Alexandr Wang recreates classic video in pure code with Muse Wonder — alexandr_wang · 2026-10-13
- Finale of AI-Generated Gay Romance Short Drama "The Mafia Boss Can't Resist Me" — Business-Fun-1999 · 2026-10-13
- Single-prompt short film made with Seedance 2.5 on Runway goes VHS-horror — umesh_ai · 2026-10-13
- Claude can finally generate images, per early user reports — nikola_mr64990 · 2026-10-13