Experiment: Can AI Replace Character Animation for Soft Skills Training?

What question did we test?

We tested whether AI character animation could replace traditional animation for soft skills training video without losing quality, consistency, or instructional value. Soft skills content relies on believable people and subtle reactions, so it is a hard test for AI. The short answer was yes, partly, but not in the way most people expect.

The pressure behind the question is real. Character animation is one of the most time-consuming parts of training video production, and soft skills topics often need several characters across several scenes. If AI could handle that, smaller teams could produce far more scenario-based content.

We did not test on a toy example. We used a real project.

How was the experiment set up?

We rebuilt a real harassment prevention training video originally developed for the Cornerstone Foundation, using AI-powered character animation in place of the original animation workflow. Keeping the same script and learning goals let us compare the two approaches directly. Tested in July 2026.

Our team includes instructional designers and animators, so the rebuild was reviewed by the same people who make these videos every week.

What did we build?

We produced an AI-animated version of the harassment prevention video using modular asset generation: characters, poses, backgrounds, and expressions generated as separate pieces, then assembled. The video above walks through the process, the problems we hit, and the side-by-side comparison with the traditionally animated version.

8:44

Generating assets in modules, rather than whole scenes at once, was the change that made the results usable. It let us reuse a stable character across scenes and fix one element without regenerating everything.

What happened when we tested it?

AI sped up asset creation and gave a small team more visual range, but it did not remove the work. Character consistency and lip sync were the weakest points, prompt iteration mattered more than tool choice, and human animators still added the most value in the moments that carry the learning.

What worked

• Modular generation made characters more consistent across scenes.
• Strong storyboards turned prompting into a predictable production step.
• Iterating on prompts produced better results than switching tools in search of a perfect one.
• A smaller team could produce a higher-quality piece than its size would normally allow.

What didn’t

• Characters drifted between shots without careful control.
• Lip sync was unreliable and needed human correction.
• Subtle reactions, the core of a harassment prevention scene, were the hardest to get right.
• Savings in asset creation were partly offset by time spent reviewing and fixing [VERIFY: net time difference].

Can AI replace character animation? The verdict

AI can replace parts of character animation for soft skills training, not the whole job. Use it inside a hybrid workflow where instructional design and animators set the story and fix the details. Avoid relying on it alone when subtle human reactions are the learning point or when characters must stay identical across many scenes.

For realistic live-action-style characters instead of animation, compare our AI video for customer service training experiment, which used Google Veo.

How can you try AI character animation yourself?

Start with a short scene from an existing video, so you have a baseline to compare against. Storyboard it in detail, generate assets in modules, and keep a human animator or designer in the loop. Judge the result on consistency and instructional clarity, not on how impressive the first frame looks.

1. Choose one scene you’ve already produced. A known baseline makes the comparison honest.
2. Storyboard every shot. Fix characters, expressions, settings, and timing before you prompt.
3. Generate in modules. Create characters, poses, and backgrounds separately so you can reuse and fix pieces.
4. Reuse prompts for each character. Keep a prompt sheet so the same person looks the same.
5. Plan for lip sync fixes. Budget human time for dialogue scenes.
6. Review with an instructional designer. Ask whether each scene still teaches the intended behavior.
7. Compare honestly. Track hours and quality against the original [VERIFY: suggested comparison template].

Our guide on how to create microlearning videos covers the scripting and storyboarding steps in more depth.

What’s next for this experiment?

The next question is consistency at scale: can a stable AI character carry a whole series of soft skills scenarios, not just one video? We already saw a related technique work in our Google Veo branching scenario, where reusing visual prompts kept a character named Julia consistent across takes.

We also want to measure how learners react to AI-animated characters compared with traditional animation [VERIFY: planned learner comparison]. Results will appear in the Lab. In the meantime, see finished pieces in our AI-generated training video examples.