SesquiLSR: Open-Source Lightweight Latent Super-Resolution for Flux2/SDXL
LoganBooker · reddit · 2026-07-06
Developer LoganBooker has open-sourced SesquiLSR, a lightweight latent space super-resolution upscaler designed for mainstream image generation models like Flux2, Anima, and SDXL. It supports arbitrary 1-2x upscaling and includes ComfyUI nodes. Unlike traditional VAE encode-decode upscaling pipelines, SesquiLSR performs upsampling directly in the latent space. Built over roughly 10 months of research, it incorporates recent advancements in pixel-domain super-resolution (such as L1 loss and LR/HR pair training) while maintaining low hardware requirements.
More from Multimodal
- A fine-tuned Krea 2 raw model produced a rainy-night driving scene — darlens13 · 2026-07-27
- Users ask whether Video2X can load custom OpenModelDB models — Used-Profit2355 · 2026-07-27
- A builder wants AI to reverse-engineer viral video effects into ComfyUI workflows — stale2000 · 2026-07-27
- Midjourney V8.2 adds personalization and shows off stylized image outputs — Mr_AllenT · 2026-07-27
- Midjourney’s image variety draws a Krea 2 comparison and asks how to reproduce it — diffusion_throwaway · 2026-07-27
- AI short film sets a 1985 dystopia to music and leans into cinema — ProfessorKey98 · 2026-07-27