Unique Research · Interview · August 21, 2024 · Guest: Jiajun Lü (吕家俊), Founder & CEO, AKOOL · Interviewer/editor: Qiuping
AI video is among 2024’s hottest tracks. AKOOL is a full-stack player — avatars, video translation, generation, face swap and film-grade effects — for enterprise marketing. Lü holds a UIUC AI PhD, Stanford visiting researcher time, HBS executive education; early ops at two B+ tech cos; 10+ years on person generation/processing; Apple and Google alum. Tagline motivation: ads bombard without immersion — build a “cool” AI marketing platform. ~50 people across 3 countries, 4 time zones, 8 cities.
Interview highlights below.
Q: Core advantages?
Lü: Leading face AIGC edit/generation; custom solutions for premium brand clients (plus consumer SKUs). Full-stack ownership — algorithms, data, deploy — priced above peers; aim Top-1 quality or cut features. Deep lip-sync and face know-how; own architecture for clearer, higher-quality models; data from self-collect, purchase, partners (incl. film/set data). Algorithm cadence every 2–6 weeks; two new models soon.
Q: How do you position AKOOL?
Lü: Build the “Adobe of AI” — an AI content platform for enterprises.
Q: Team geography / customers?
Lü: HQ Silicon Valley; Shanghai office; Seoul studio; staff elsewhere — distributed. Acquisition via flagship clients and case proof; product strength ultimately wins. Products are mainly standardized; brand marketers co-create; APIs for “Lego” customization. Vertical solutions (e.g. personalized video email for bulk send) shipping; China still early (MMS-like trials hit video-size limits).
Q: Data / fine-tune needs?
Lü: Pretrained models usable out of box; fine-tunes need little data — minutes of video → good fine-tune in 1–2 hours.
Q: Homogenized AI creative / Cannes CMO point on missing quality standards?
Lü: Creative + attention are the point; hard to unify “good ad” taste — use ROI, brand memory, purchase intent. Differentiate vs free C-end tools by going premium B2B — high-value problems competitors don’t solve. Mature winners rarely win on one feature; chase shifting creative trends with a comprehensive product aimed at enterprise.
Q: Copyright / misuse?
Lü: Tools assist creation; copyright and liability sit with creators (Adobe/Word analogy — vendor isn’t liable for user content). Users must own rights to uploads.
Q: U.S. brand adoption vs China; marketer misconceptions?
Lü: Big brands still early — adoption maybe <0.1% yet market already large; at 10%+ scale expands sharply. Brands cautious on IP/PR; smaller brands move faster. Those who try see traffic and clear ROI. Marketers fear replacement — tools still need human oversight at high standards; urgent task is learning AI to raise throughput; lowers skill floor, raises industry competition (search-auction dynamic).
Q: Dependent on underlying foundation models?
Lü: Better speech/text/video/image models enrich UX, but the biggest gap remains excellent effect — brands care about ROI. Many products sit on the usable/unusable edge until models mature.
Q: Other hard parts / market education?
Lü: Pick real market needs; keep fast iteration and stickiness or education spend trains customers for someone else. Video outputs are intuitive — education cost lower than many categories; start simple UI, then unlock advanced features. >95% customers overseas; overseas wants less customization than China — transplanting China custom know-how could be an edge. Country buying styles differ (some decide before sales call; others never close).
On China vs U.S. AI entrepreneurship: paces roughly parallel; broad participation and comparable product/tech levels from his overseas-focused vantage.
— END —