Case study · ByteDance Intelligent Creation · 2022–2024
One generation platform, from Doubao to CapCut
As product owner of ByteDance's multimodal generative AI platform, I put text, image, and video generation behind one set of APIs and SDKs, and into products used by billions.
LayersResearchModelSystemProductGrowthRevenue
At a glance
50M / dayimages generated by Doubao's text-to-image modelSource: 财新, 2024-09-24 ↗
Billionsusers reached by the editing and effects capabilities across TikTok, Douyin, CapCut, and Jianying
FiledDoubao text-to-image model, filing no. Beijing-FuLuGua-20231205Source: 豆包 ↗
01context
Context
Generative models were improving faster than product teams could adopt them. TikTok, Douyin, CapCut, Jianying, and Doubao each needed generation and editing capabilities at the same quality, safety, and compliance bar, without each rebuilding the stack.
02system
How it works
One platform, many consumers. GAN and diffusion models for text, image, and video sit behind the MaaS APIs and the VE / Effect AI SDK, and each business line calls them for its own effects and intelligent-editing features.
- Models
- GAN and diffusion models
- Text, image, and video generation and conversion
- Platform
- MaaS APIs
- VE / Effect AI SDK
- Model filing and compliance
- Business lines
- Doubao: text-to-image
- TikTok: effects and intelligent editing
- Douyin: effects and intelligent editing
- CapCut and Jianying: intelligent creation and editing
03role
What I owned
- Product owner for the multimodal generative AI MaaS platform: API services and SDKs for text, image, and video generation and conversion.
- Project owner for the model filing of Doubao's text-to-image model (filing no. Beijing-FuLuGua-20231205).
- Served ByteDance business lines through the VE / Effect AI SDK: effects and intelligent editing for TikTok and Douyin; intelligent creation and editing for CapCut and Jianying.
04more
Coverage & related
Coverage
Related