cd /news/generative-ai/domain-tuned-diffusion-super-resolut… · home › topics › generative-ai › article
[ARTICLE · art-171548] src=arxiv.org ↗ pub= topic=generative-ai verified=true sentiment=↑ positive

Domain-Tuned Diffusion Super-Resolution for 4K Facial Asset Maps

A domain-tuned diffusion upscaler converts 1K facial asset maps to 4K (1024 to 4096 pixels) by adapting a frozen SDXL backbone with lightweight LoRA modules per modality while a Tile ControlNet preserves input structure, according to an arXiv paper (2611.21890v1). On held-out maps the method reduced albedo LPIPS from 0.21 to 0.08 and produced sharper, more faithful detail than bilinear and off-the-shelf GAN and diffusion super-resolvers, while avoiding misplaced or excessive displacement relief. The authors conclude that lightweight domain adaptation can yield render-ready 4K facial asset maps without a custom super-resolution architecture.

key takeaway

Adapting a frozen SDXL backbone with per-modality LoRA modules plus a Tile ControlNet upscales facial albedo and displacement maps 4x from 1024 to 4096, cutting albedo LPIPS from 0.21 to 0.08 versus bilinear and off-the-shelf super-resolvers.

by read1 min views7 publishedNov 18, 2026

arXiv:2611.21890v1 Announce Type: new Abstract: Physically based facial assets require 4K albedo and displacement maps for cinematic close-ups, yet generative pipelines commonly produce them at only 1K. Naively applying photographic super-resolution is a poor fit: facial UV maps have domain-specific statistics, and displacement is a scalar height field whose errors directly distort geometry. We present a domain-tuned diffusion upscaler for 4x (1024 to 4096) facial asset super-resolution. Starting from a bilinearly upsampled map, we adapt a frozen SDXL backbone per modality with lightweight LoRA modules, while a Tile ControlNet preserves input structure as the diffusion prior restores pore- and wrinkle-scale detail. On held-out maps, our method produces sharper, more faithful detail than bilinear and off-the-shelf GAN and diffusion super-resolvers, reducing albedo LPIPS from 0.21 to 0.08 while avoiding misplaced or excessive displacement relief. These results show that lightweight domain adaptation can produce render-ready 4K facial asset maps without a custom super-resolution architecture.

── more in #generative-ai 4 stories · sorted by recency
── more on @sdxl 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/domain-tuned-diffusi…] indexed:0 read:1min 2026-11-18 · —