This repository creates images which are afterwards posted to an instagram account. Basically what happens is that text is created and applied on an image. Afterwards some fitting hashtags are created on basis of that text, those are then used as the caption.
The text that is applied on the images and the hashtags are created using a LLM.
One example of such an image would be the following. The text was created through a LLM and the text and the author are then applied on the white canvas.
Afterwards the text and author name are fed again into the LLM to generate trending hashtags that are relevant to that post.
The LLM is also fed the information of what posts the LLM has already created in the past. That helps to avoid creating text and therefore images which were already created in the past.
The codebase is using kedro for managing the pipelines. It is also making us
of dynamic pipelines.
For all pipelines it is needed to define a topic that the post is generally trying to
do, for example the generation of quotes. The second step is then to define
the subtopics such as life, love, breakup. Those are then defined in the config
files.
To run the quote pipeline for example, one needs to write:
kedro run --pipeline quote_pipelineThe above will create images for all subtopics that were defined.
## Further Notes
The project was described and explained in more detail on my blog.