yhf20060726 
Success metric 01
New additions per creative
How many new users a single video material can drive.
2026
AIGC Video Automation
Exploring how AIGC can turn product and trend signals into better-performing e-commerce advertisement videos. Creating a reusable workflow for generating trend-led video ads.
TIMELINE
2026 Jun - 2026 Aug
TEAM
2 Designers, 1 Product Manager, 1 Developer
ROLE
AI Product Design Intern @TikTok
SKILLS
AI Workflow, AIGC, Prompt Engineering, Video Strategy
OVERVIEW
At my internship at TikTok, my team collaborated with the TikTok User Growth team to explore how AIGC can turn product and trend signals into better-performing e-commerce advertisement videos.
01
AI Workflow Building
Exploring how to create a workflow that generates high-quality videos with minimal human intervention.
02
Prompt Engineering
Editing prompts for plot, script, character, and scene descriptions to support a reusable AIGC workflow with strong results.
03
Video Strategy
Exploring different video formats and styles, then evaluating how effectively they can be produced through the AIGC workflow.
GOAL & SUCCESS METRIC
Goal: Create advertisement videos that deliver strong performance results, and build a workflow that can later support automatic generation at scale.
Success metrics
Success metric 02
Spend per creative
How strongly a material performs in ad delivery, measured by spend per creative.
SOLUTION
A creative automation workflow that turns one product brief into several ad-video directions.
CREATIVE GENERATION MINDMAP
From an idea to a completed ad video
- 01
Plot Generation
Add supporting image - 02
Script Generation
Add supporting image - 03
Visual Generation
- Scene Images
- Character Images
Add supporting images - 04
Video Generation
Add supporting image
IMPACT
The 投放数据复盘 delivery review found that several generated materials performed better than the agency-made and other production-path baselines.
2 CIDs
Agency benchmark exceeded
On Douyin, a single material generated 400+ new additions and about 7k in spend — exceeding the agency-video baseline.
3 CIDs
Clear production-path advantage
A single material generated 100+ new additions and 1k+ in spend, performing clearly better than the other production paths in the review.
NOW
Moving into automation
TikTok is putting the workflow into automation to create similar videos at scale.
NEXT
Expanding the system
A new system is being developed to capture hot topics and high-performing videos, turn them into human-made templates, and unlock more automated video types.
DETAIL
Each step turns a clear input into a tangible output that can be reviewed before the next step begins.
01 · Build the product brief
The workflow starts by translating raw commerce data into an instruction the creative system can use.
- Input
- Product link, title, product images, price, selling points, and target audience
- Output
- A structured product brief: key message, visual references, product assets, and the selected creative route
02 · Generate the idea and script
The product brief is paired with a proven trend or video structure — such as a drama twist or rapid remix — to create a story people will recognize in the feed.
- Input
- Structured product brief, selected format, trend reference, and conversion goal
- Output
- A hook, plot beats, product-insertion moment, script, and shot-level content direction
03 · Create the visual world
The script becomes a coherent set of characters, scenes, and product placements that the video model can use as visual anchors.
- Input
- Script, character descriptions, scene descriptions, product image, and unified visual direction
- Output
- Character reference images, scene images, and product-placement visuals ready for generation
04 · Generate video segments
Instead of asking for one long, unreliable generation, the system produces the story as controllable video clips that can be checked independently.
- Input
- Script beats, product visual, character and scene references, and individual video prompts
- Output
- Short vertical video segments that match the approved story and visual direction
05 · Assemble, review, and publish
The final stage turns the generated clips into a delivery-ready ad and creates a feedback loop for the next round of automation.
- Input
- Approved video segments, captions, voiceover, fixed elements, end card, and campaign requirements
- Output
- A final 9:16 advertisement video, ready for AI and human review, platform delivery, and performance analysis
PROCESS & REFLECTION
Turning an exploration into a reusable creative system.
Explore formats before automating them
I tested short-drama insertion, rapid remix, AI short drama, animation, and IP-led routes to understand where each structure could create attention for a product.
Make the creative choices reviewable
Staging the process creates useful checkpoints: the hook, script, visual assets, edit, and delivery strategy can each be assessed and improved.
Let performance shape the next template
The most valuable output is not one video; it is a system that can learn from strong results and create the next automated template faster.
Share the system, not only the result
Sharing the exploration helped make the workflow actionable for the team and supported its transition into a broader automation effort.