abra

The goal: take a vid at an event, throw it you your abra agent, it will cut it in shape, enhance it, add branding, music and even gifs and finally schedule it for publication. A fully automatized brand agent.

Technical goal: make use of local resources (GPU particularly for video editing and generation):

• preset that can be installed on top of any openclaw-like agent
• brand skill gather brand information and assets + post-scheduler facilitate posting using buffer
• 16+ skills handling media transformation and generation
• skills build as isolated services for easy expansion and customization
• high level workflow scripts enable controlled processing to reduce AI indeterminism

Skill	Input	What it does	Min VRAM	CPU ok?
photo-picker	images	Score and pick the best photos	~1 GB	✅ fast
bokeh-effect	images	Synthetic bokeh / portrait mode	~1.5 GB	✅ acceptable
background-remover	images	Remove background	~0.5 GB	✅ acceptable
image-captioner	images	Auto-describe and caption	~4 GB	✅ slow
frame-interpolator	videos	Frame interpolation	~2 GB	⚠️ impractical
video-matte	videos	Remove video background	~3 GB	❌ no
audio-splitter	video/audio	Separate vocals from music	~2 GB	✅ slow
music-generator	prompt/video	Generate music	~3 GB	✅ very slow
animate-image	images	Image → video clip	~8 GB	❌ no
video-editor	video	Edit/inpaint video	~8 GB	❌ no
giphy	videos	Animated GIF stickers from GIPHY + optional SFX	0 GB	✅ instant
freesound	videos	Freesound SFX into video + optional GIF overlays	0 GB	✅ instant
pixabay	videos	Royalty-free image/video overlays + optional SFX	0 GB	✅ instant
video-enhancer	videos	Sharpen, colour grade, normalise audio	0 GB	✅ fast
video-captioner	videos	Whisper transcription + animated captions	0 GB	✅ ~30s/clip

Based on claw-parade repo (similar setup, ~10 skills, no workflows, no automated installation) developed during the first OpenClaw hackathon with AI tinkerers in Paris.