跳到正文
DEV Community· Kushalka·· 3 天前AI 評分39

我開發了一個生成動漫圖片的 Telegram 機械人

I built a Telegram bot that generates anime images

AI 導讀

作者開發了 Stable Diffusion Bot,讓用戶透過 Telegram 提交提示詞並取得動漫風格圖片。機械人提供多種動漫 checkpoint、可疊加的 LoRA、負面提示詞及採樣器和步數設定,也可在一則訊息中傳送同一提示詞生成的多張圖片。服務可免費開始,另設付費模式以解鎖更多內容類別。

正文

Kushalka

I've been generating anime-style images with Stable Diffusion locally for a while. At some point I noticed I was repeating the same loop every day: open the web UI, type a prompt, wait, download, send it to a friend. So I did what any reasonable person does at 2am — I wrapped the whole thing into a Telegram bot.

Stable Diffusion Bot is simple: you send a prompt, you get an image back. A few things it can do:

  • several anime checkpoints to pick from, each with a different vibe
  • styles and LoRAs you can stack on top of a model
  • negative prompts, sampler and step settings if you like to tinker
  • albums — multiple variations of one prompt in a single message
  • a free tier for everyone (a premium mode unlocks additional content categories)

The surprise for me was that the model is maybe 20% of the work. The rest is plumbing: queueing requests so a popular checkpoint doesn't melt the GPU, deduplicating button taps, sending albums as multipart without Telegram complaining about it.

Here are a couple of images straight from the bot:

a glowing cheeseburger with anime-style energy effects
an anime girl with horns

Try it here: Stable Diffusion Bot — the site with docs is kushalka.dpdns.org.

It's free to start. Feedback is welcome, especially prompt ideas I should test.

來源:DEV Community · dev.to