SDXL Workflow Bottlenecks: How to Fix Eye and Detail Artifacts During Upscaling?
TheJesusGuy · reddit · 2026-08-05
The author shares their SDXL text-to-image workflow running on an RTX 3080 10GB: generating an image with a DPMPP2DSDE sampler, upscaling 4x, downscaling while adding noise for detail, and upscaling again.
While producing high-quality results, the workflow has artifacts like off-center or non-round pupils and unnaturally sharp specular highlights (e.g., water drops). The author seeks community advice on whether this workflow is outdated and how to optimize it.
More from Multimodal
- Running MiniMax H3 on Dual RTX 5060 Ti: VRAM Splitting and Unloading Strategies — Kahvana · 2026-08-06
- Oscar-Nominated Cinematographer to Judge First-Gen AI Filmmakers at Higgsfield Festival — SimplyAnnisa · 2026-08-06
- MiniMax H3 Test: Generates Video with Sound Effects in 120s on RTX 5090 — R34vspec · 2026-08-06
- GradiumAI Details Methodology for Curating High-Quality AI Voice Catalog — mattturck · 2026-08-06
- Running MiniMax H3 on AMD GPUs Hits a Wall with SageAttention Dependency — itiswhatitiswgatitis · 2026-08-06
- Local Test of MiniMax H3: Flawless Cloth Physics and Temporal Consistency — papjak · 2026-08-06