Tekken's Harada on AI Story Videos: Output Without Process Misses the Point
Former Tekken director Katsuhiro Harada responded to a viral AI-generated video summarizing the series, and his message hits a nerve for anyone building with AI. Judging the final output without considering how it was made leads to weak conclusions. For dev teams, that's a reminder: process, data, and intent matter as much as the result.
Why "better output" isn't a complete metric
Harada pushed back on claims that AI videos present Tekken's story better than the games. He argued that quality can't be measured by surface polish alone. If you ignore the pipeline-what the model was trained on, the constraints, and the creative intent-you're grading a screenshot, not a system.
Context always frames the work
He compared the reaction to asking why ancient battles didn't use modern weaponry, or laughing at pre-1980s films for lacking drones and CGI. It's an easy mistake: judging older work against tools that didn't exist. Evaluation needs context: time, tools, constraints, intent.
What AI summaries get wrong about character and story
According to Harada, the AI video exists because it's trained on the original team's visuals and narrative. Yet the interpretations drift from the intended characters, the visuals are inconsistent, and the language feels off-more like exaggerated localization errors than clear storytelling. Impressive at first glance doesn't mean faithful, coherent, or meaningful.
What's worth respecting in AI
Harada still credits AI's pace and cost efficiency. Those are real advantages in production. But speed and cheap output don't automatically equal better creative execution-or a legitimate claim that AI is "doing it better" than the original games.
Tech changes, craft adapts
Game development is glued to technology, so AI will be part of the mix. Harada recalled thinking Photoshop would end hand drawing; it didn't. Tools shift the workflow, not the need for taste, judgment, and a strong pipeline.
Practical takeaways for dev and AI teams
- Evaluate outputs against intent: fidelity to character, world rules, and narrative beats.
- Score models on more than polish: coherence, consistency, authorial voice, and explainability of the process.
- Track data provenance: know what your models learned from, and where creative IP boundaries sit.
- Use AI where it compounds: previsualization, iteration, testing, localization drafts, asset tagging-then route through human review.
- Document style guides and narrative bibles so AI outputs have a reference for tone and canon.
- Optimize for the trio: cost, speed, and quality. Don't let one silently throttle the others.
- Be explicit about limits: what AI can suggest vs. what creators must decide.
The bigger lens
Harada's stance isn't anti-AI. It's pro-context and pro-craft. Focus on the systems that produce the work, not just the superficial result.
If you're new to Tekken, start with the official series hub for context on the characters and canon: tekken.com.
Building AI skills with a focus on practical workflows? Browse current courses and compare paths by role: Latest AI courses.
Your membership also unlocks:
