CORE GUIDE
Multimodal iteration
Text, image, audio and other modalities need iterative review against cross-modal meaning and production constraints, not independent generation followed by assembly.
Mental model
A multimodal artifact is one communication system. Each modality has its own generation constraints, but quality depends on whether the pieces agree on meaning, hierarchy, timing and audience experience.
Why it matters
A good script paired with a misleading image is still a bad lesson. Generated captions can contradict visuals, diagrams can introduce labels the text never explains, and aesthetically strong media can overwhelm the instructional point. Multimodal work therefore needs a shared brief and review loop that checks relationships between modalities rather than approving each asset in isolation.
01
Iterate from a shared communication contract
Define the learning or product objective, key claims, audience and hierarchy before generating individual assets. Produce a first pass across modalities, then review alignment: does the image support the claim, does narration match visible state, is critical information duplicated or missing, and can the artifact still work with accessibility alternatives? Revise the weakest relationship rather than regenerating everything at once.
02
Example: an animated explanation of a context window
The narration says old messages are being compacted, but the animation merely fades random tokens. A cross-modal review catches that the visual implies deletion rather than state-preserving compression. The animation is revised to show selected constraints and evidence surviving into a smaller representation while the script remains unchanged.
Common failure modes
- Reviewing text, image and audio quality independently without checking semantic alignment.
- Using decorative media that competes with the information hierarchy.
- Regenerating all modalities after one mismatch and losing already-correct work.
Engineering heuristics
- Give every modality the same objective, claims and audience constraints.
- Review cross-modal contradictions explicitly before polishing style.
- Change the smallest failing relationship during iteration.
Takeaways
- 01Multimodal quality is relational, not the sum of separate asset scores.
- 02Shared briefs preserve meaning across media.
- 03Iteration should target mismatches between modalities as well as local defects.
Used in
This Concept is reused across these canonical learning paths.