Dify 1.11.0 (2025-12-11)
公式リリースノート: https://github.com/langgenius/dify/releases/tag/1.11.0
v1.11.0 - Your knowledge base just went from mono to full HD
このバージョンの記事
| # | テーマ | 状態 |
|---|---|---|
| 01-multimodal-knowledge-base | マルチモーダルナレッジベース (1.11.0) | draft |
| 02-whats-actually-happening | 実装の内側で起きていること (1.11.0) | draft |
| 03-other-new-features | その他の新機能 (1.11.0) | draft |
| 04-performance-infrastructure | パフォーマンスとインフラ (1.11.0) | draft |
| 05-key-fixes | 主要な修正 (1.11.0) | draft |
リリースノート抜粋
🧠 Multimodal Knowledge Base
Dify can now understand both text and images in your knowledge base.
Before, embeddings only cared about words. But real docs? They’re full of screenshots, diagrams, memes (don’t lie), and markdown with embedded images. Now those visuals count too.
What’s Actually Happening
- Auto‑Image Extraction
Drop a doc with Markdown images — like— and Dify grabs them automatically (JPG, PNG, GIF ≤ 2 MB). - Smart Attachment Handling
Each image is linked to its matching text chunk, so context stays tight. - Embedding Behavior
- Using a multimodal embedding model? Then both text + image get vectorized. You can:- search text ↔ image,
- image ↔ image,
- or image ↔ text.
Basically, everything ↔ everything. - Sticking with a regular text‑only model? Cool — the images still go along for the ride and will appear in prompts when you use a Vision‑enabled LLM.
- Developer Candy
Knowledge Pipeline’sKnowledgeBasenode speaks two new dialects —
-multimodal‑Parent‑Child
-multimodal‑General
Perfect for plugin devs who want to feed Dify structured mixed‑modality data without headaches. - Supported Multimodal Models （tips: update below plugin to the latest version）
- AWS Bedrock:nova‑2‑multimodal‑embeddings‑v1:0
- Google Vertex AI:multimodalembedding@001
- Jina:jina‑embedding‑v4,jina‑clip‑v1,jina‑clip‑v2,jina‑reranker‑m0
- Tongyi (Qwen):multimodal‑embedding‑v1
Once everything’s embedded, you’ll see a shiny new Multimodal tag in your knowledge base — because now your data’s two‑dimensional (and twice as cool).
Huge thanks to @JohnJyong (#29115) and @WTW0313 (#27793) for bringing visual intelligence to Dify. Absolute legends.
🧩 Other New Features
- Zen Mode — disappear into your flow; fewer buttons, more focus. use
cmd + kto enter.
thanks @ZeroZ‑lab (#28794) - “Open Workflow” shortcut for WorkflowTool — hop straight into editing lik
…(以下省略、公式リリースノートを参照)
本記事は非公式まとめです。 正式な機能仕様、互換性、移行手順については Dify 公式ドキュメント および 公式リリースノート を必ずご確認ください。