Skip to content

Albedolizer v1.6.1-beta

Choose a tag to compare

@invisiblelevel invisiblelevel released this 19 Sep 16:01
· 82 commits to main since this release
6791969

Albedolizer v1.6.1-beta

🔧 Changes

  • CLIP vision quantized — 167 MB → 84 MB (INT8), faster inference
  • CLIP tokenizer bundled — no longer downloads from HuggingFace on first launch
  • FilePicker — single instance across the entire app (fixes memory leaks)
  • UI — wider right panel (320px), log panel moved under the preview
  • Math button stretched to full width
  • Window transparency removed
  • Tiling — new 🔲 Seamless button (radial mask + scatter)
  • Viewer — fully reworked lighting (Cook-Torrance BRDF + IBL)
  • lut_model — fixed LAB→RGB conversion (clip before astype)
  • Slimmer exe — CUDA/TensorRT/ffmpeg binaries stripped

📦 Installation

Download Albedolizer_Setup_v1.6.1-beta.exe and run it.

🔧 Изменения

  • CLIP vision квантизован — 167 МБ → 84 МБ (INT8), инференс быстрее
  • CLIP tokenizer забандлен — больше не тянет с HuggingFace при первом запуске
  • FilePicker — один инстанс на всё приложение (фикс утечек)
  • UI — правая панель шире (320px), лог-панель сдвинута под превью
  • Math-кнопка растянута на всю ширину
  • Прозрачность окна убрана
  • Тайлинг — новая кнопка 🔲 Seamless (radial mask + scatter)
  • Вьюер — полностью переработано освещение (Cook-Torrance BRDF + IBL)
  • lut_model — фикс конвертации LAB→RGB (клип перед astype)
  • exe похудел — вырезаны CUDA/TensorRT/ffmpeg бинари

📦 Установка

Скачай Albedolizer_Setup_v1.6.1-beta.exe и запусти.