game-video-ad-pipeline

🎮 Game Video Ad Pipeline (FFmpeg-first)

A skill that generates game promotional videos (Shorts/ad creatives) with fully local + free tools.

Input: --game-id (Steam app id)
Automation: screenshot collection + aspect ratio normalization + text overlay
Output: 15–30 second MP4 (1080x1920 or 1920x1080)
Constraints: ffmpeg required, no external paid APIs, runnable on MiniPC

Structure

game-video-ad-pipeline/
├─ SKILL.md
├─ scripts/
│  └─ generate-video.py
└─ references/
   ├─ RESEARCH.md
   ├─ AUDIT.md
   └─ TEST_REPORT.md

Quick Usage

cd misskim-skills
python3 skills/game-video-ad-pipeline/scripts/generate-video.py \
  --game-id steam:570 \
  --template action \
  --orientation vertical \
  --duration 18 \
  --output skills/game-video-ad-pipeline/output/dota2-action-vertical.mp4

Landscape ad (1920x1080):

python3 skills/game-video-ad-pipeline/scripts/generate-video.py \
  --game-id 730 \
  --template puzzle \
  --orientation horizontal \
  --duration 20 \
  --output skills/game-video-ad-pipeline/output/cs2-puzzle-horizontal.mp4

If you want to use a local image/GIF folder as input:

python3 skills/game-video-ad-pipeline/scripts/generate-video.py \
  --game-id 999999 \
  --input-dir ./my-assets \
  --template idle \
  --duration 15

Template System

Template	톤	Default Copy
`action`	High saturation/high contrast, strong impact emphasis	`Play Now`
`puzzle`	Bright and clean tone	`Challenge Your Brain`
`idle`	Soft and stable tone	`Start Your Idle Journey`

All templates are implemented as ffmpeg filter chains and behave identically in MiniPC/CLI environments.

Core Pipeline

Extract screenshot URLs from Steam store HTML (no key/payment)
Normalize images/GIFs to the target resolution (portrait/landscape)
Generate per-cut clips and concat
Overlay title/tagline/CTA text
Output MP4 (H.264, yuv420p, faststart) and verify playability

Parameter Summary

--game-id (required): 570 or steam:570
--template: action|puzzle|idle
--orientation: vertical|horizontal
--duration: 15~30 초
--max-shots: number of screenshots to use (default 6)
--title, --tagline, --cta: directly specify overlay text
--input-dir: use local media instead of Steam auto-collection
--keep-temp: keep intermediate files for debugging

4 Critical Absorption Questions (Application Summary)

What to take?
- ffmpeg's reliable portability/speed, Remotion's template concept, the workflow structure of competing skills.
What to discard?
- Dependence on paid generation APIs, forced cloud rendering, excessive JS runtime complexity.
How to change it to fit our environment?
- Rewrite as a single Python + ffmpeg script, keyless collection via Steam HTML parsing.
How to measure success?
- Generate a 15–30 second/1080 resolution/playable MP4, MiniPC compatible, cost $0.

Safety/Operations Notes

No external LLM/paid API calls
Before running, check that ffmpeg, ffprobe exists
Render text overlays with Pillow(PIL) so they work even in environments without ffmpeg drawtext
For large batches, keep --max-shotsbetween 4 and 8

For detailed rationale, see references/RESEARCH.md, references/AUDIT.md .