Skip to content

Latest commit

 

History

History

Approach#1-Text-to-Text

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 

Approach #1 | Text-to-Text

The system will help users to generate text for one or multiple placeholders depending on the element context.

Technology readiness Risks Complexity
🟢 Ready for implementation
🟡 Moderate risk
🟢 Light complexity

Technologies

Open-source LLMs are already capable of working with this task, like open-llms [Github]. A solution like privateGPT [Github] gives the ability to work offline with local files, increasing the security of the solution.

Pipeline

Step #1) User selects elements or pages to populate with text

Step #2) System extracts information about the number and context of the module

Step #3) Based on input parameters, the Language model generates the output

Step #4) User verifies the output and approves the insertion

Relevant works

Figma plugins presented here mostly demonstrate the possible interface of the implemented solution.

Pros and Cons

🟢 Pros

  • This approach would significantly streamline the process of generating text for placeholders, saving time and effort for designers.
  • The system can adapt to different element contexts, ensuring the generated text is relevant and fits the design's purpose.
  • With technologies like privateGPT, the system can work offline with local files, providing an added layer of data security.

🔴 Cons

  • The success of this feature heavily relies on the system's ability to accurately determine the context of the module. That context should be pre-processed for the system, like count and type of placeholders, name, the functionality of the element, etc.
  • AI can generate text based on context, but understanding the designer's precise intent or the emotion they wish to convey might be challenging. Especially in the visual domain.