The more context and details you add, the better your response because the model understands precisely what you expect. Do not miss our Prompt Engineering Guide to learn how to write great prompts.

The models can call another model in the background to generate images. The current image model in Langdock is Dall-E3 by OpenAI. Image generation uses the following steps:

  1. The model you selected chooses the image generation tool and writes a prompt to the image model in the background.
  2. The image model generates the image based on the prompt and returns it to the main model and you as the user.

You can select any language model for image generation. Each model can send prompts to the image model, but the language models prompt the image model differently. Feel free to try different models and see how the generated images differ.

Here are some known limitations we are working on:

  • Text in images has mistakes / is written in non-existing letters:
    Image models were trained on real images, which included text. Based on the training data, the model generates objects that look similar. The image model can not (yet) write full, correct sentences but tries to mimic letters from the alphabet. This leads to incorrect spelling or non-existing letters. This is a current limitation of image models and will be improved by the model providers in upcoming model versions.