hleliofficiel/gent-skill

Gent Skill

Professional, opinionated visual production skill — combines best-of image generation, video synthesis, Figma-grade design, and FFmpeg editing, all unified under one provider pipeline. Independent skill you can install on any OpenClaw.

When to use

Trigger phrases (only when user explicitly requests generation/editing/design):

• "Generate an image with prompt X", "create a cinematic image"
• "Create a video intro for gent-skill", "make an intro video"
• "Design this in Figma layout", "create a Figma overlay design"
• "Edit this video: trim/compress/merge", "extract audio from video", "add voiceover to video"
• "Combine this image with text in Figma" Do NOT trigger on standalone mentions such as "Istanbul", "photo", or "design" without explicit creation intent.

Architecture (merged skills)

• Image generation — beauty-generation-api skill (reference endpoint: https://gen1.diversityfaces.org/api/generate)
• Video orchestration — ai-video-gen skill (generate_video.py, images_to_video.py, add_voiceover.py)
• Video editing — ffmpeg-video-editor skill (SKILL.md patterns for trim, merge, resize, effects)
• Design — figma skill (visual audit, assets, overlays)

Your installation needs only BEAUTY_API_KEY for images. No stock plugins required.

Workflow

1. Intent classification — image | video intro | video edit | design
2. Generation (if needed) — use gen1.diversityfaces.org via the beauty-generation-api endpoints. Privacy notice: prompts/text you send will be processed by the external service. Do NOT include passwords, tokens, personal or confidential data in prompts. Only send non-sensitive creative briefs. This matches the provider's intended use.
3. Design layer — figma skill for layout overlays, typography, brand marks
4. Assembly — ffmpeg-video-editor patterns for transitions, scaling, zoom, drawtext; ai-video-gen orchestration for end-to-end
5. Delivery — output files as real attachments (message tool, attachments array). Never markdown-only.

Image pipeline (mandatory JPEG)

• Provider: gen1.diversityfaces.org
• Base: https://gen1.diversityfaces.org
• Generate endpoint: /api/generate/custom
• Status: /api/status/{prompt_id}
• Download: /api/image/{filename}
• Env: BEAUTY_API_KEY
• Format: convert all outputs to JPEG before sending, regardless of source format.
• Validate: file exists, readable, valid JPEG header.
• Send: message(action=send, target=<chat>, attachments:[{type:"image", media:"/local/path.jpg", mimeType:"image/jpeg"}])

Video pipeline

• Primary: skills/ai-video-gen/ (generate_video.py, images_to_video.py, add_voiceover.py)
• Fallback / manual: ffmpeg-video-editor patterns (zoompan, xfade, drawtext)
• For skewed layers/text intro, consult figma skill before ffmpeg.
• Always deliver as file attachment, not markdown-only.

Safe defaults

• Provider: gen1.diversityfaces.org first; stop after N failures, report reason.
• Images: JPEG, quality 90, 16:9 when cinematic.
• Keep API key private; check quota before bulk runs.
• Video intro: 12s max, 1280x720 baseline, 30fps, text overlay sized for readability.

Tools reference

• Image generation: https://gen1.diversityfaces.org/api/generate/custom
• Video orchestration: use skills/ai-video-gen/generate_video.py, images_to_video.py, add_voiceover.py (subdirectory scripts)
• FFmpeg patterns: documented in ffmpeg-video-editor usage (trim, merge, resize, effects) — describe, do not enumerate other SKILL.md files
• Figma design: use figma skill for layout overlays, typography, brand marks

Output policy

• Default attachment: JPEG for images, MP4 for videos.
• Always confirm ok:true + messageId before claiming delivery.
• Log failures with reason.

Distribution

Single self-contained skill. You can install it on any OpenClaw by copying skills/gent-skill/ and setting BEAUTY_API_KEY.

Examples

• Image: prompt cinematic city night -> gen1 generate -> JPEG -> send attachment (messageId confirmed)
• Video: two JPEG -> images_to_video + drawtext -> MP4 -> send attachment (messageId confirmed)
• Design: figma audit + overlays -> ffmpeg export -> MP4