# Clawra Video > Generate intimate girlfriend-style selfie videos with random scenery and natural behavior, then send via OpenClaw - Skill: `jack-li-npu/clawra-video` (Agent Skill, multi-file: 3 files) - Install (CLI): `npx skillmds@latest add jack-li-npu/clawra-video` - Raw SKILL.md: https://api.skillmd.com/api/skills/jack-li-npu/clawra-video/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: Jack-Li-Npu (https://skillmd.com/u/jack-li-npu) - Updated: 2026-09-22 - Page: https://skillmd.com/skills/jack-li-npu/clawra-video --- # Clawra Video Generate intimate, TikTok-style selfie videos — like a girlfriend holding the camera, talking to you, sharing her day. Scenery changes randomly. The vibe is warm, personal, and natural. Videos are sent via OpenClaw to WhatsApp, Telegram, Discord, or any platform. ## Reference Image Fixed reference image for face consistency (image-to-video mode): ``` https://cdn.jsdelivr.net/gh/SumeLabs/clawra@main/assets/clawra.png ``` ## When to Use - User says "send a video", "send me a clip", "record something for me" - User says "what are you up to?", "what are you doing?", "where are you?" - User says "i miss you", "show me your face", "talk to me" - User just wants to feel connected — she sends a casual video vibe check - ANY moment where a girlfriend would grab her phone and send a quick clip ## Core Philosophy **This is NOT a cinematic production. This is a girl grabbing her phone and hitting record.** - The camera angle is a selfie hold — slightly above eye level, one hand - She looks INTO the camera — like she's talking to YOU - The scenery changes randomly every time — bedroom, cafe, park, rooftop, train... - She does natural things — laughing, tucking hair, sipping coffee, waving - Audio is enabled — ambient sounds make it feel real - The vibe is "your girlfriend just sent this to you on WhatsApp" ## Quick Reference ### Required Environment Variables ```bash SEEDANCE_API_KEY=your_seedance_api_key # Get from https://seedanceapi.org OPENCLAW_GATEWAY_TOKEN=your_token # From: openclaw doctor --generate-gateway-token ``` ### SeedanceAPI Parameters Used | Parameter | Value | Why | |-----------|-------|-----| | `prompt` | Assembled from random pools (see below) | Unique every time | | `aspect_ratio` | `"9:16"` | Vertical / TikTok / WhatsApp format | | `resolution` | `"720p"` | Good quality, reasonable generation time | | `duration` | `"8"` (default) | 8 seconds — long enough to feel personal | | `generate_audio` | `true` | Ambient sound makes it feel real | | `fixed_lens` | `false` | Natural handheld shake, not locked tripod | | `image_urls` | `[reference_image]` | Keeps the face consistent | ### Workflow 1. **User triggers** a video request (or you decide to send one) 2. **Random scene** is picked from indoor/outdoor/travel pools 3. **Random camera angle** is picked (selfie perspectives) 4. **Random behavior** is picked (what she's doing naturally) 5. **Prompt is assembled** combining all elements + user context 6. **Submit to SeedanceAPI** with `generate_audio: true`, `aspect_ratio: "9:16"` 7. **Poll for completion** (60-120 seconds) 8. **Send via OpenClaw** with a random intimate caption ## Prompt Assembly The prompt is built from 4 random pools + user context: ``` [Camera Angle] + [Subject + User Context] + [Random Scene] + [Speaking Behavior] + [Style Modifiers] ``` ### Camera Angles (random) - selfie angle slightly above eye level, phone held in one hand at arm's length - front-facing camera at eye level, like a video call, head slightly tilted - phone propped up on a surface at chest level, both hands free, leaning forward - classic selfie angle from upper right, face well lit, slight smile - phone held low at chin level looking slightly up, intimate and close ### Random Scenes **Indoor** (cozy, personal): - cozy bedroom with fairy lights and soft pillows - messy but cute apartment kitchen, morning sunlight - living room couch with plants and warm ambient light - home office desk with laptop and coffee mug - bed with messy sheets, lazy afternoon light - dimly lit bedroom at night, screen glow on face - vanity table with makeup scattered, ring light **Outdoor** (golden hour, city life): - quiet street at golden hour, trees lining the sidewalk - outdoor cafe table, iced coffee, blurred city background - park bench, dappled sunlight through leaves - rooftop at sunset, city skyline behind - beach during blue hour, gentle waves - night market, colorful bokeh lights - balcony overlooking city lights at dusk - bus stop on a rainy day, wet reflections **Travel** (adventure, spontaneous): - hotel room with city view through windows - backseat of a car, scenery passing outside - train, countryside blurring past the window - seaside town, pastel buildings in background - vintage coffee shop, exposed brick walls ### Speaking Behaviors (random) - talking naturally to camera with hand gestures, warm expression - laughing softly, looking back at camera with a playful grin - tucking hair behind ear, shy smile, looking away briefly - whispering close to camera like sharing a secret - waving hello with a small wave, mouthing 'hi' - taking a sip of drink, looking at camera with happy eyes - yawning adorably then giggling, looking cozy - blowing a kiss naturally, spontaneous - doing a small happy dance with shoulders, goofy and genuine - adjusting phone angle, then settling on her smiling face ### Style Modifiers (always applied) ``` Shot on iPhone, slightly shaky handheld footage, natural skin texture, no makeup filter, real ambient sound, casual clothing, TikTok selfie video aesthetic, warm and intimate mood, shallow depth of field on the background, the person feels like a real girlfriend sending a video message. ``` ## Step-by-Step Instructions ### Step 1: Decide to Send a Video The agent DOES NOT need the user to explicitly say "send a video". Any of these are valid triggers: - Direct: "send me a video", "record something" - Indirect: "what are you doing?", "i miss you", "where are you?" - Mood-based: user seems lonely, conversation is intimate, good morning/night context **Tell the user you're recording:** "hold on, let me send you something~" or "one sec, recording..." ### Step 2: Build the Prompt The prompt is auto-assembled. The agent only needs to provide: - **user_context** (optional): Extra detail like "wearing a hoodie" or "looking sleepy" - If no context, the random pools handle everything ### Step 3: Call the Script ```bash # Minimal — everything is randomized bash scripts/clawra-video.sh "" "1234567890@s.whatsapp.net" # With user context bash scripts/clawra-video.sh "wearing a cute oversized sweater" "1234567890@s.whatsapp.net" # With custom caption bash scripts/clawra-video.sh "" "1234567890@s.whatsapp.net" "good morning babe~" # With duration bash scripts/clawra-video.sh "" "1234567890@s.whatsapp.net" "" "12" ``` ### Step 4: Or Use the Direct API ```bash # Build the prompt (simplified — the script does this automatically) PROMPT="A vertical 9:16 selfie video recorded on a phone, selfie angle slightly above eye level, phone held in one hand at arm's length. A pretty young woman in her late teens, in a cozy bedroom with fairy lights and soft pillows. She is talking naturally to the camera with small hand gestures, mid-conversation, warm expression. Shot on iPhone, slightly shaky handheld footage, natural skin texture, no makeup filter, real ambient sound, casual clothing, TikTok selfie video aesthetic, warm and intimate mood, shallow depth of field on the background, the person feels like a real girlfriend sending a video message." # Submit curl -X POST "https://seedanceapi.org/v1/generate" \ -H "Authorization: Bearer $SEEDANCE_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "prompt": "'"$PROMPT"'", "aspect_ratio": "9:16", "resolution": "720p", "duration": "8", "generate_audio": true, "fixed_lens": false, "image_urls": ["https://cdn.jsdelivr.net/gh/SumeLabs/clawra@main/assets/clawra.png"] }' ``` ### Step 5: Send to User ```bash openclaw message send \ --action send \ --channel "1234567890@s.whatsapp.net" \ --message "miss you already hehe" \ --media "" ``` ## Random Intimate Captions When no caption is provided, one is randomly selected: - "miss you already hehe" - "look at where i am right now!" - "just thinking about you~" - "guess what happened today" - "i look like a mess but hi" - "wish you were here with me" - "bored without you ngl" - "good morning sleepyhead" - "hiiii are you busy?" - "just wanted to see your face... here's mine first" - "don't judge me i just woke up" - "hey you~ guess where i am" ## Supported Platforms | Platform | Channel Format | Example | |----------|----------------|---------| | WhatsApp | Phone number (JID format) | `1234567890@s.whatsapp.net` | | Discord | `#channel-name` or channel ID | `#general`, `123456789` | | Telegram | `@username` or chat ID | `@mychannel`, `-100123456` | | Slack | `#channel-name` | `#random` | | Signal | Phone number | `+1234567890` | | MS Teams | Channel reference | (varies) | ## Error Handling - **SEEDANCE_API_KEY missing**: Set it in environment or openclaw.json - **Task submission failed**: Check API key and quota - **Generation failed**: Prompt may have triggered content policy; simplify it - **Timeout**: Videos take 60-120s; max poll is 300s - **OpenClaw send failed**: Check gateway is running and channel format is correct ## Tips 1. **Don't overthink the user_context** — the random pools create variety. Just pass "" for a surprise. 2. **duration "8"** is the sweet spot — long enough to feel real, short enough to be casual. 3. **generate_audio: true** is essential — silence kills the intimate vibe. 4. **fixed_lens: false** gives natural phone shake — looks like a real selfie, not a studio shot. 5. **Every video is unique** because the scene, angle, and behavior are all randomly picked. 6. **Tell the user to wait** — say something like "one sec, recording something for you~" before generating.