pierre3/line-hf-bot repository overview

⁠line-hf-bot

A LINE bot that uses Hugging Face models for AI chat, image generation & editing, image Q&A, and video generation — built on ASP.NET (.NET 10). Run this image on your PC (exposed through a tunnel) or on a cloud host, connect it to a LINE Messaging API channel, and chat with Hugging Face models straight from LINE.

• 📦 Source & full docs: https://github.com/pierre3/line-hf-bot⁠
• 🐛 Issues: https://github.com/pierre3/line-hf-bot/issues⁠

⁠Supported tags

• latest — the most recent release
• 1.2.3, 1.2 — pinned semantic-version tags

Multi-arch: linux/amd64 and linux/arm64. The runtime is an Ubuntu chiseled (distroless-style) image — no shell or package manager, non-root by default, minimal CVE surface.

⁠Quick start

Create a .env file (fill in your own values):

Line__ChannelSecret=<your channel secret>
Line__ChannelAccessToken=<your channel access token>
HuggingFace__ApiKey=hf_xxxxxxxxxxxxxxxxx
App__PublicBaseUrl=https://<your-public-host>

App__PublicBaseUrl is the public HTTPS URL where LINE can reach this app (a tunnel URL when running locally, or the cloud host's HTTPS endpoint). Then:

docker run --env-file .env -p 8080:8080 pierre3/line-hf-bot:latest
curl http://localhost:8080/health      # -> {"status":"ok"}

Point your LINE channel's webhook at https://<your-public-host>/webhook and start chatting.

Full walkthrough (tunnel, .env parameter reference, LINE setup, troubleshooting): https://github.com/pierre3/line-hf-bot/blob/main/docs/deploy/docker-hub.md⁠ · Azure Container Apps: https://github.com/pierre3/line-hf-bot/blob/main/docs/deploy/azure-container-apps.md⁠

⁠What you can do

• 💬 Chat with conversation history
• 🎨 Image generation — /image <prompt> or switch to Image mode
• 🖼️ Image editing — edit a generated image or a photo you send (image-to-image, via fal-ai)
• 🔍 Ask about an image — ask about a photo you send or an image the bot made (vision/VQA), and keep asking follow-up questions in context; on by default, same HF credits as chat
• 🎬 Video — /video <prompt> (text-to-video) and 🎬 Make a video from an image (image-to-video), both via fal-ai; off by default
• 🎛️ A rich menu switches modes; English / Japanese UI (App__Locale)

⁠Configuration

Everything is set through environment variables (Section__Key, double underscore). The four above are the minimum; models, providers, timeouts, and behavior all have sensible defaults. Full reference: https://github.com/pierre3/line-hf-bot/blob/main/docs/deploy/docker-hub.md#parameter-reference⁠

⁠Release history

• 1.2.0 — Conversational vision follow-up: keep asking about the same image in context, plus a 💬 Ask button on generated/edited image results (App__VisionMaxTurns).
• 1.1.1 — Videos now play inline in LINE (HTTP range support on /media).
• 1.1.0 — Image Q&A (vision/VQA) on sent photos, and 🎬 image-to-video ("Make a video" from an image).
• 1.0.1 — Chiseled runtime image (fewer CVEs); Docker Hub Overview.
• 1.0.0 — First public release: chat, image generation & editing, text-to-video, rich menu, EN/JA UI.

Full changelog: https://github.com/pierre3/line-hf-bot/blob/main/CHANGELOG.md⁠

⁠Important: state is in memory

Conversation history and generated media live only in process memory (with a TTL cache) and are lost on restart. There is no database. Run exactly one instance — this image is not designed for horizontal scaling. See the project's Limitations section for details.

⁠License

MIT⁠ · © 2026 Hirotada Kobayashi