Quality assurance · Production

Leading genAI text-to-image app accelerates human preference training data creation with Labelbox RLHF tooling

The problem

A leading generative AI text-to-image app needed to rapidly generate high-quality human preference training data to improve its models, but creating and labeling this data internally would divert valuable resources from core product development and required dedicated tooling and labeling teams to meet tight deadlines.

Workflow diagram · grounded in source
1
Training data need identified
trigger
“looking for rapid ways to generate high-quality human preference training data”
2
RLHF preference labeling
human_review
“LLM human preference editor to perform reinforcement learning from human feedback (RLHF), which enabled them to easily create preference data for training a reward model based on multiple outputs from a single model”
3
Multimodal model output comparison
validation
“comparing model outputs on text-to-image prompts versus other leading models”
4
Weekly data engine benchmarking
validation
“compare multiple models against each other in live, multi-turn conversations and rank outputs for a more data-driven evaluation of models”
5
Targeted data creation and re-verification
feedback_loop
“target training data creation towards the areas which needed the most improvement and which would have the most impact on model performance, and iteratively fed back and verified their results through another human evaluation run”
Reported outcome

Using Labelbox, the company significantly improved the speed at which it could create high-quality human preference training data and sped up product development from months to weeks.

Reported metrics
Training data creation speed (headline)2x
Training data creation speed (body)over 50%
Product development timefrom months to weeks
Reported stack
LabelboxLabelbox Labeling ServicesLLM human preference editormultimodal chat solution
Source
https://labelbox.com/customers/text-to-image-app-customer-story/
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Using Labelbox, the company significantly improved the speed at which it could create high-quality human preference training data and sped up product development from months to weeks.

What tools did this team use?

Labelbox, Labelbox Labeling Services, LLM human preference editor, multimodal chat solution.

What results were reported?

Training data creation speed (headline): 2x; Training data creation speed (body): over 50%; Product development time: from months to weeks (source-reported, not independently verified).

How is this quality assurance AI workflow structured?

Training data need identified → RLHF preference labeling → Multimodal model output comparison → Weekly data engine benchmarking → Targeted data creation and re-verification.

Related quality assurance cases