Go to file
2023-05-06 21:21:49 +03:00
.gitignore fix negative prompt 2023-04-27 13:57:48 +03:00
Google Colab Run.ipynb colab is valid, but untested and old 2023-03-18 15:12:50 +02:00
LICENSE Initial commit 2023-01-12 09:52:08 +05:30
main.py lowercase all payalod keys 2023-05-06 21:18:30 +03:00
README.md additions 2023-05-06 21:21:49 +03:00
requirements.txt Create requirements.txt 2023-01-12 10:09:08 +05:30

AI Powered Art in a Telegram Bot!

this is a txt2img bot to converse with SDweb bot API running on tami telegram channel

How to

supported invocation:
/draw <text> - send prompt text to the bot and it will draw an image
you can add negative_prompt using ng: <text>
you can add denoised intermediate steps using steps: <text>

basicly anything the /controlnet/txt2img API payload supports

  "prompt": "",
  "negative_prompt": "",
  "denoising_strength": 0.5,
  "seed": -1,
  "n_iter": 1,
  "steps": 20,
  "cfg_scale": 7

/draw a city street
and without people
/draw a city street ng: people
with more steps
/draw a city street ng: people steps: 50

to change the model use:
/getmodels - to get a list of models and then click to set it.

  • note1: Anything after ng will be considered as nergative prompt. a.k.a things you do not want to see in your diffusion!
  • note2: on negative_prompt (aka ng):
    thia is a bit of a black art. i took the recommended defaults for the Deliberate model from this fun alt-model spreadsheet.
    and you (currntly) can only ADD to it, not replace.
  • note3: on steps - step of 1 will generate only the first "step" of bot hallucinations. the default is 40. higher will take longer and will give "better" image. range is hardcoded 1-70.
    see video


Install requirements

conda create -n sdw python=3.8
conda activate sdw
pip install -r requirements.txt

(note: conda is not strictly necessary, but it is recommended)

Original readme

My Bot uses Automatic1111's WebUI as the backend. Follow the directions on their repo for setup instructions.

Once you have WebUI set up, run webui.sh with the --api argument. You can also add other arguments such as --xformers to use xformers memory efficient attention.

You can use the web ui interface that Automatic1111 provides to select the model and VAE to use. Their repo has documentation on how to do so. I also recommend doing a test generation

Create a file called .env in the same folder as main.py. Inside the .env file, create a line TOKEN = xxxx, where xxxx is your telegram bot token. create a line API_ID = xxxx, where xxxx is your telegram id api id. create a line API_HASH = xxxx, where xxxx is your telegram id api hash. create a line SD_URL = xxxx, where xxxx is your sd api url.

Now, you can run the bot

python main.py