It’s April 1st, and that means a special edition from the PetSignal Squad! Today we’re going beyond practical demos. We’re crafting GPTs (Generative Pet Tales): legendary, AI-generated pet backstories that turn your furry friends into superheroes.
Using Label Studio’s latest AI-powered tools, we’re going to transform our pet images into origin stories worthy of blockbuster drama. Want to follow along? You’ll be building an AI pet story generator in just a few steps.
Note: Label Studio’s AI Assistant and Prompts features are available in our Enterprise and Starter Cloud plans. You can try them for free with a Starter Cloud Trial or by contacting your customer success manager.
First things first, we need photos of our furry (or less-furry) friends. We’ve compiled a splendid gallery of pets from HumanSignal. Yes, they're real, and yes, they know they're cute. You can see them too, by dropping this URL into the data import field when creating your project.
Next, we configure our labeling setup. Label Studio's AI Assistant is good at generating labeling templates for serious business. It’s also great at generating templates for content to include in our Pet Tale! Here are the instructions we provided:
Given an image of a pet, I want to provide it with a name, a rating for mischievousness, a written backstory for the pet, and select any applicable super powers (including: Zoomie Speed, Mind Control, Sofa Melting, Chaos Field, Snack Radar). Please include headers for the sections.Feel free to customize your own story template and style! In seconds, we get a template to use directly for our Pet Tales generation capturing all the info we want. All that’s left is to click “Save & Open”.
Awesome! Now the project’s all set up, and you can already see a preview of the pet images 👀.
This is where some real magic happens. We head over to Label Studio Prompts, click “Create Prompt”, select our project, and click “Create”.
Now we can interact directly with an LLM and feed it instructions to generate some content! Here’s the prompt we used with a gpt-4o-mini model (and don’t forget to include the image variable itself).
Given the image provided:
- Come up with a name for the pet that reflects their appearance, the photo, and the backstory.
- Based on their actions in the photo, rate the pet's mischievousness (liklihood to cause a little chaos or break the rules).
- Create a short, creative, and funny backstory for the pet in the style of a hero or villain origin story. Use the image's context as inspiration.
- Assign fitting superpowers to the pet based on their image.
#Image
{image}
Make sure to save your Prompt.
Finally, we hit “Run” and our generative pet tales did not disappoint. In a matter of moments, every pet had a tailored tale worthy of a TV series.
Clicking into any of the images, you’ll get the full scoop on one pet’s profile. Here’s one example, and we’ve attached a few other Generated Pet Tales at the bottom of this post.
For a project as mission-critical as this one, you may want to evaluate and refine your prompt for more detailed backstories, or perhaps you need multiple reviewers to verify the accuracy of these tales. Either way, Label Studio is the most flexible platform for your data labelling and AI quality needs.
Happy labeling and may your pets' backstories be ever legendary!
Using LLMs to label data is fast—but is it accurate? This post shows two ways to evaluate model performance in Label Studio, with or without ground truth. Learn how to compare outputs, track accuracy and cost, and choose the right model for your workflow.
OpenAI’s new Structured Outputs feature allows you to ensure outputs conform to a defined JSON structure. In this blog, we’ll explore how to leverage this feature for various labeling tasks.
In this article, we want to demonstrate a method of curating large datasets to reduce but not remove the cost for curating a high quality medical Q&A dataset in Label Studio and fine-tuning Llama 3 on this data.