사용 방법
- 설치 명령을 복사하고 자신의 에이전트 환경에 설치하기 전에 소스 패키지를 확인하세요.
- 필요한 도구, 인증 정보, 지원 환경은 원본 문서에서 확인하세요.
- 입력과 원하는 산출물을 설명하고 출처와 결과를 대조하세요.
npx skills add https://github.com/jpeetz/agent-skills --skill betterlife-image-generation시작 전 확인
- 설치에는 모델 구독, API 한도, 파일 접근 권한이 포함되지 않습니다.
- 별점 수와 설치 수는 활동 지표이며 독립적인 품질·보안 검증이 아닙니다.
원본 저장소의 제목, 예시, 코드, 표, 링크, 이미지를 유지해 표시합니다.
BetterLife Daily Social Image Generation (kie.ai image-to-image)
Generate the daily BetterLife social media image using kie.ai's grok-imagine/image-to-image model, conditioned on Mira reference photos from Google Drive. This is image-to-image, not text-to-image — the model generates from reference images, not from a prompt alone.
Architecture
Compose Prompt (scene + attire + mood + angle)
→ Select 8 Mira reference images (5 for grok) from 20 GDrive photos
→ Send to kie.ai grok-imagine/image-to-image (strength 0.85)
→ Fallback: flux-2/flex-image-to-image on failure
→ Save to ~/workspace/tool-image-generation/Credentials
- KIE_API_KEY — for the kie.ai API
- Read from credential source by label — never hardcode.
Reference images (Supabase S3 storage)
Reference images are stored in Supabase storage at: https://dbbppsbtpsahdkmevxgm.supabase.co/storage/v1/object/public/instagram_images/
File naming: {character}_reference_{nn}_{pose}.jpg
Reference selection
- Face anchors (MANDATORY): front + 3/4 angle references. Always in the first 2 slots.
- Pose refs (3-6 remaining): scored by tag overlap with scene keywords, randomly sampled
from top 14. Grok caps at 5 total refs, flux at 8.
Character anchor
Mira, a wellness coach with natural beauty, shoulder-length dark hair,
warm olive skin, kind eyes, minimal makeupPrompt composition
Format: {anchor}, {scene}. {attire}. {detail}. {mood} {accent}. Premium editorial wellness photography, {angle}, no text, no watermarks.
Components picked combinatorially to avoid repeats: 36 scenes, 8 attires, 10 details, 6 moods, 8 angles, 4 accents.
kie.ai API call
POST https://api.kie.ai/api/v1/jobs/createTask
{
"model": "grok-imagine/image-to-image",
"input": {
"prompt": "<composed prompt>",
"image_urls": ["<5 ref URLs>"],
"strength": 0.85,
"aspect_ratio": "1:1", "resolution": "1K", "output_format": "jpg"
}
}Poll GET /api/v1/jobs/recordInfo?taskId={id} every 2s until state=="success".
Fallback: flux-2/flex-image-to-image (uses input_urls, allows 8 refs).
Key parameters
- strength 0.85 — higher = more Mira-likeness, lower = more prompt-driven scene
- Known limitation: at strength 0.85 with the same 20 refs, output is structurally dominated
by the refs. Scene words shift background but not composition. For genuine variety, lower strength or add shot-size/framing contrast per variant.
Post-generation
- Image saved to
~/workspace/tool-image-generation/betterlife-{theme}-{date}-v{n}.jpg - Review JSON at
/tmp/bl-social-review.json - Posted via
bl_social_post.pyon approval
