Content Platform · AI Tooling
Transforming fragmented educational content production into a scalable AI-assisted workflow for images, animations, and videos.
TAL’s content platform supports curriculum teams across course creation, asset management, and production. Educational images, animations, and videos still relied heavily on specialized designers, making content creation slow and hard to scale.
As the product designer for AI content production, I led the 0→1 design of image, animation, and video workflows. I partnered with product, engineering, QA, curriculum designers, and instructors.
I designed a human-in-the-loop workflow that accelerated repetitive work while preserving creative control. Creators could review, adjust, and regenerate outputs within their existing production process.
The AI workbench helped curriculum teams shift from manual production queues to scalable, AI-assisted creation across images, animations, and videos. In the first six months, teams produced 40K+ assets through the workflow.
in the first 6 months
using the workflow
2 workdays → 3 hrs
AI could speed up generation—but production was still hard to scale.
Before designing the AI tools, I mapped how curriculum teams produced images, batch assets, animations, and videos. Across workflows, the same three challenges kept appearing.
A common production pattern across different content types.
AI needed more than faster generation—it needed consistency, human control, and a workflow that could scale. These three principles guided the 0→1 design of the workbench.
Optimize for control—not one-click automation.
I reframed the workflow around upfront constraints and review checkpoints, so teams could catch inconsistencies early and correct them before they became costly downstream.
Early concept
Fast in theory, but quality issues surfaced only after generation was complete.
Final direction
Upfront constraints improved consistency, while checkpoints let teams correct issues before moving downstream.
AIgenerate · vary · repeat
Humandefine · review · approve
One workbench, built for different production needs.
I translated the same design strategy into four AI-assisted workflows, each designed for consistency, control, and production at scale.
I unified image, batch, animation, and video tasks in one shared queue so teams could track progress, resume work, and hand off tasks without switching tools.
A shared production model across different AI workflows.
I designed image, batch, animation, and video around the same core production model: define inputs, generate, review, correct selectively, then continue.
01 Image Workflow
The image workflow follows the shared model: define inputs, generate options, review the result, and refine only what needs to change.
01Define inputs
Define shared styles, formats, and references upfront to reduce variation and keep related outputs visually consistent.
02Generate & refine
Creators could compare multiple outputs, choose the strongest direction, and refine the selected result with built-in tools—correcting only what needed to change without restarting the workflow.
02 Batch Images Workflow
The batch workflow turns one script into a structured content set: define shared conditions once, review prompts before generation, then correct only the outputs that need attention.
01Shared setup
Shared model, style, format, and references give every output the same starting conditions.
02Review before generation
The script is structured into scene, character, and element prompts. Each prompt stays editable, so creators can refine only the inputs that need attention before generation.
Prompt-level controls
Edit, upload, or regenerate one prompt without rebuilding the batch.
03Review by exception
Results return in one structured view, so creators can approve what works and adjust only the outputs that drift.
Output-level controls
Edit or replace one generated asset without restarting the batch.
03 Video Workflow
Video was the most complex workflow in the workbench. I moved key decisions earlier—locking content, voice, and character direction before rendering, then using storyboard checkpoints to catch issues before costly video generation.
01Define the production inputs
Script, voice, and character references are confirmed early so later scenes share the same content, tone, and visual direction.
02Storyboard checkpoint
Storyboard frames let creators review scene composition, characters, and continuity before committing to video generation.
03Generate & edit
Once the storyboard is approved, the workbench generates video clips. Creators can send selected clips to CapCut for detailed editing while keeping AI generation inside the workbench.
From AI generation to a repeatable production system.
The workbench brought image, batch, animation, and video production into one shared workflow, turning AI generation into a repeatable production system for curriculum teams.
Moving constraints, review, and correction upstream made production more predictable and reduced late-stage rework. In the video workflow, production time dropped from 2 workdays to 3 hours.