Skip to content

Releases: Pescorr/kusa_commenter

kusa_commenter v1.1.0 (llama.cpp 同梱版)

Choose a tag to compare

@Pescorr Pescorr released this 25 Apr 13:42

ローカル LLM 同梱版(LM Studio 不要)

llama.cpp + Qwen3.5-9B Vision を同梱。3 ステップで起動できます。

セットアップ手順

  1. kusa_commenter_v1.1.0_win64.zip をダウンロードして解凍
  2. setup_model.bat を実行(HuggingFace から約 6.2 GB のモデル DL、初回のみ)
  3. ScreenCommentator.exe を起動

動作環境

  • Windows 10/11 (64bit)
  • メモリ 16 GB 以上推奨
  • GPU: NVIDIA + CUDA 12 推奨(VRAM 8 GB 以上)。CPU only でも動作可
  • 空きストレージ 8 GB 以上

同梱モデル

  • LLM: Qwen3.5-9B-Q4_K_M.gguf (5.3 GB)
  • Vision projection: mmproj-BF16.gguf (879 MB) ※BF16 推奨版
  • ライセンス: モデル Apache 2.0、llama.cpp MIT。商用利用可。

同梱推論エンジン

llama.cpp のビルド済み Windows 版

  • CUDA 12 backend(GPU 自動検出)
  • CPU フォールバック対応(Haswell〜Zen4 等の専用 DLL 同梱)

thinking 制御

config.ini の [llama_server] disable_thinking = true(既定)で
Qwen3 の <think>...</think> 生成を完全停止。コメント生成が高速化します。
有効にしたい場合は false に変更してください。

v1.0.0 との違い

  • v1.0.0: LM Studio 別途インストール必要
  • v1.1.0: 全部同梱、ZIP 解凍 → モデル DL → 起動 の 3 ステップ + thinking 抑制

詳細は README を参照。

kusa_commenter_LM_Studio v1.0.0

Choose a tag to compare

@Pescorr Pescorr released this 25 Apr 11:47

ローカル LLM で動く、ニコニコ風 AI コメントオーバーレイ

AI が画面を見て、5 ペルソナがリアルタイムにコメントを流します。完全ローカル動作・API 課金ゼロ。

使い方

  1. ZIP を解凍
  2. LM Studio を起動して Vision モデルをロード(localhost:1234)
  3. ScreenCommentator.exe を実行

動作環境

  • Windows 10/11
  • メモリ 8GB 以上推奨
  • LM Studio + Vision モデル(qwen2.5-vl など)

詳細は README を参照。