Solutions: Generative AI Platforms

Generative AI doesn’t generate people in the same way that advertisers, agencies, and film and TV studios cast them, leading to pledges to not use Generative AI for content with people in it. 

Alltold solutions help Generative AI platforms produce images and videos of people that match brand, agency, and studio expectations.

Step 1

You pick a set of ads you want to analyze (e.g., a campaign, Q2 ads for a brand, etc), and send them to us

Step 2

We analyze your ads and send you a report

Step 3

We’ll meet once to review the results and gather your feedback

Alltold solutions for Generative AI providers fall into two categories:

Step 1

You pick a set of ads you want to analyze (e.g., a campaign, Q2 ads for a brand, etc), and send them to us

Step 2

We analyze your ads and send you a report

Step 3

We’ll meet once to review the results and gather your feedback

I. Solutions for model training

  • Training dataset evaluations
  • Model baseline evaluations

These solutions give a model trainer insight into the distribution of people in the dataset and in the trained models, highlighting areas for further dataset acquisition and model fine-tuning.

asian family having breakfast together
asian family having breakfast together

Training dataset evaluations

The dataset you use in pre-training influences the distribution of people generation of the trained model. Knowing what’s in your dataset is a necessary first step.

Alltold evaluations deliver analytics on the distribution of people across your training data from the individual person detected, to image or video level analytics, to insight into groups of people.

Model baseline evaluations

Distributions of people present in training data don’t map directly to distributions of people generated from a trained model or candidate model. It’s important to measure distributions of a model as well.

Alltold model baselining generates a statistically significant set of images of people following a standard methodology, and measures the distribution. If training set evaluations exist, it compares the model baseline to the training set evaluation.

II. Solutions for trained models

  • Alltold prompt iterations
  • Alltold prompt / response insights API

These solutions allow a Generative AI platform to offer users functionality to optimize their use of a trained model.

asian family having breakfast together
asian family having breakfast together

Alltold prompt iterations

Enterprises need to ensure Generative AI use is adhering to brand or organization guidelines. Training can help, but integrated tooling gives enterprises the confidence they need for broader organizational rollouts.

Alltold prompt iterations allow an organization to set a desired distribution across a set of identity attributes, and suggests prompt modifications to encourage people generation in accordance with the guideline distributions.

Alltold prompt / response insights

In addition to setting guidelines for desired distributions, enterprises need a reporting mechanism to measure performance over time.

Alltold prompt / response reporting shows how closely organizational people generations have aligned to brand guidelines and organizational guardrails.