Multimodal safety for children's video — transcript, frames, and family values aligned in one score families can trust.
Lemu Kids is a family safety product from Lemu Lab. It scores children's video content using multimodal AI — combining what is said, what is shown, and what each family considers acceptable — so parents get clear signals without monitoring every minute.
Screen time is not the enemy. Uninformed screen time is. Lemu Kids helps families make intentional choices about what children watch, grounded in measurement rather than marketing promises.
Platform age ratings are blunt instruments. A video can be technically "for kids" while clashing with a family's values around language, themes, or tone. Parents either micromanage every viewing session or trust black-box algorithms they cannot interrogate.
Single-modality filters — transcript-only or thumbnail-only — miss context. A harmless frame can accompany risky dialogue; a scary visual might appear in an otherwise educational clip.
Lemu Kids fuses transcript analysis, frame-level visual signals, and configurable family profiles into a unified safety score. Parents see why a score was assigned, not just a red or green label.
Child safety cannot ship on vibes. SafeWatch-Bench measures cross-modal agreement, false negatives on subtle risks, and alignment drift when family profiles change. Every content score in Lemu Kids is regression-tested there before release.
The eval is documented on the lab domain; the live product runs at kids.lemu-lab.com.
Families who want intentional screen time — parents balancing autonomy for children with guardrails they can explain and adjust. Educators and child-development professionals evaluating content pipelines may also find the measurement approach relevant.
Related: SafeWatch-Bench eval · All cases · Lab evals