Yuri Noviello, Anastasiia Birillo, Gosia Migut (2026) — Delft University of Technology. arXiv:2605.16295 [cs.CY; cs.AI; cs.CL; cs.GR; cs.HC; cs.MM]
Noviello, Birillo, and Migut (2026) present ANVIL, an end-to-end multimodal generation pipeline for educational content — one of the first systems to automate the full journey from concept definition to rendered instructional animation. The four-stage pipeline (analogy generation, screenplay compilation, animation code generation with automated repair) represents a significant advance in AI-generated instructional-design materials.
ANVIL's evaluation approach is methodologically notable: rather than relying solely on automated metrics, the authors ground quality assessment in teacher evaluations and then use those findings to guide scalable automated screening. The LLM-based evaluator for analogy quality and fidelity-to-screenplay proxy for video assessment offer a replicable framework for evaluating generative-ai educational outputs at scale — addressing a key challenge identified in benchmark and efficacy-study literature.
The positive educator response to perceived value and usability suggests that AI-generated instructional content may be crossing a threshold of practical classroom utility. This connects to the teacher-role discussion: ANVIL positions AI as a content-generation assistant that amplifies rather than replaces instructor expertise. The focus on CS education also complements the cs-education literature on AI tools, though ANVIL's architecture is domain-agnostic and could generalize to stem-education broadly. For faculty-development, tools like ANVIL lower the production barrier for high-quality instructional media, potentially democratizing access to professional-grade educational animations.
Related Pages
- generative-ai — related
- llm — related
- instructional-design — related
- cs-education — related
- teacher-role — related
- stem-education — related
- faculty-development — related
- benchmark — related
- efficacy-study — related
- multimodal — related
Citation
APA: Yuri Noviello, Anastasiia Birillo, Gosia Migut (2026). ANVIL: Analogies and Videos for Lecturers. arXiv:2605.16295. arXiv:2605.16295 [cs.CY; cs.AI; cs.CL; cs.GR; cs.HC; cs.MM].