AI art bot

The app generates random images using AI Stable Diffusion and posts them in Telegram https://t.me/ai_art_random and Twitter https://twitter.com/ai_art_random

Configuring

Environment variables:

• COMMAND_ONLY_MODE - bot command mode only
• FACE_ENHANCER_ARCH - face enhancer architecture
• FACE_ENHANCER_MODEL_PATH - face enhancer model path for GFPGAN
• FP16 - Use half-precision model
• HUGGING_FACE_HUB_TOKEN - token for Hugging Face for downloading models
• LOW_VRAM - low video RAM mode
• PREMODERATION - premoderation mode (post only in turbo chat)
• PROMPT_MODEL_ID - Hugging Face model id for prompt
• PROMPT_MODEL_TOKENIZER - Hugging Face model tokenizer for prompt
• RANDOM_PROMPT_PROBABILITY - probability of generate full random prompt without ideas (default 0.5)
• REALESRGAN_MODEL_PATH - model path for RealESRGAN
• RESOLUTION - image resolution (default 512x512)
• SD_MODEL_ID - Hugging Face model id for Stable Diffusion
• SD_MODEL_VAE_ID - Hugging Face model id for Stable Diffusion VAE
• SD_REFINER_ID - Hugging Face model id for Stable Diffusion XL Refiner
• SLEEP_TIME - how many seconds to sleep between generations (default 600s)
• TELEGRAM_TOKEN - Telegram bot token
• TELEGRAM_ADMIN_ID - user ID to manage the bot
• TELEGRAM_CHAT_ID - chat where images will be sent
• TELEGRAM_TURBO_CHAT_ID - chat where images will be sent in turbo mode
• TURBO_SLEEP_TIME - how many seconds to sleep between generations in turbo mode (default 60s)
• TWITTER_CONSUMER_KEY - Twitter consumer key
• TWITTER_CONSUMER_SECRET - Twitter consumer secret
• TWITTER_ACCESS_TOKEN - Twitter access token
• TWITTER_ACCESS_TOKEN_SECRET - Twitter access token secret
• UPSCALING - up to 4x image resolution with Real-ESRGAN (default true)

Usage

Build and running in docker:

docker build -t ai-art-bot .

docker volume create ai-art-bot_huggingface

docker run --name ai-art-bot -d --env-file .env -v ai-art-bot_huggingface:/root/.cache/huggingface --gpus=all ai-art-bot

Running with docker-compose:

docker-compose up -d