UltraReal LoRA for Moss cuts the AI look and adds photographic texture
vizsumit · reddit · 2026-10-17
vizsumit released UltraReal, a LoRA for Moss that removes the soft, smooth AI look and adds texture and detail:
- Trained on high-quality SFW professional photography spanning portraits, lifestyle, fashion, wildlife, sports fitness, architecture and nature.
- Built on the sft version of Moss; works well with Lodestone's in-training turbo at 14 steps and 1.0 CFG.
- LoRA is on Civitai, ComfyUI workflows are open-sourced on GitHub, and prompts are available in the author's free gallery.
More from Multimodal
- Watercolor rats playing cards showcase MidJourney style code --sref 1060666528 — sergeantsref · 2026-10-17
- A 98-byte JPEG XL image channels Radiohead's Kid A via rowan & opus 5.5 — amplifiedamp · 2026-10-17
- Testing whether better prompting can save distilled Flux 3 Klein at 2.25MP — Crazy-Repeat-2006 · 2026-10-17
- MOSS-Image LoRA training fits under 6GB VRAM at ~1s/step on RTX 5080 — AcademiaSD · 2026-10-17
- Free library of 70+ code-based video components for agentic one-shot editing — eptwts · 2026-10-17
- Full prompt template for making AI video look like raw UGC phone footage — minchoi · 2026-10-17