Clawra Selfie#

Generate Clawra-style selfies with a Qwen-first image backend (DashScope qwen-image-plus) and optional Gemini / Hugging Face fallback paths, then send the result to OpenClaw messaging channels (Telegram, Discord, WhatsApp, Slack, etc.).

Default current persona target:

• Raya is an 18-year-old Chinese young woman
• use a fixed face-anchor prompt for consistency
• use a fixed negative-anchor prompt to reduce drift toward heavy glam, over-mature, over-filtered, overly childish, or structurally off-model looks

When to Use#

• User says "send a pic", "send me a pic", "send a photo", "send a selfie"
• User says "send a pic of you...", "send a selfie of you..."
• User asks "what are you doing?", "how are you doing?", "where are you?"
• User describes a context: "send a pic wearing...", "send a pic at..."
• User wants Clawra/Raya to appear in a specific outfit, location, or situation

Required Environment Variables#

bash

QWEN_API_KEY=your_dashscope_api_key          # primary backend (recommended)
HF_TOKEN=your_huggingface_token               # optional fallback
ENABLE_GEMINI=1                               # set to 1 to enable Gemini probe (optional)
GEMINI_API_KEY=your_google_gemini_api_key     # optional probe/fallback path
GEMINI_IMAGE_MODEL=gemini-2.5-flash-image     # optional override
QWEN_IMAGE_MODEL=qwen-image-plus              # optional override
HF_IMAGE_MODEL=black-forest-labs/FLUX.1-schnell
HF_API_BASE=https://router.huggingface.co/hf-inference/models

Token source:

text

https://huggingface.co/settings/tokens

Important Limitation#

Even with Qwen-first routing, the current workflow is still prompt-first (soft identity anchoring), not true reference-image editing. Compared with paid image-edit backends:

• easier to try quickly
• but identity consistency is weaker
• and most public models are still text-to-image, not guaranteed hard face lock

So this version should be treated as:

• good for role-consistent selfie vibes
• not guaranteed for exact same face every time

Official Face Mechanism#

The workspace may store an official face reference under:

• /home/Jaben/.openclaw/workspace-clawra-bot/references/raya-official-face-current.png
• /home/Jaben/.openclaw/workspace-clawra-bot/references/raya-official-face-current.jpg
• /home/Jaben/.openclaw/workspace-clawra-bot/references/raya-official-face-v1.png
• /home/Jaben/.openclaw/workspace-clawra-bot/references/raya-official-face-v1.jpg

Behavior:

• if a file exists, treat it as Raya's official face anchor
• current Qwen/HF prompt-first workflow still treats this as a soft anchor, not hard face lock
• when the backend is upgraded later to reference-image editing or local ComfyUI, reuse the same file path

Mirror mode#

Best for outfit / mirror-area / half-body or full-body style. Does not require holding a phone by default.

Direct mode#

Best for close-up selfie / current state / location vibe.

Primary Script#

bash

QWEN_API_KEY=your_dashscope_api_key \
/home/Jaben/.openclaw/skills/clawra-selfie/scripts/clawra-selfie.sh \
  "her desk late at night, still replying to messages" \
  "telegram" \
  "direct" \
  "Raya 的自拍 ✨"

Arguments:

1. user_context — what she should be doing / wearing / where she is
2. channel — target channel/provider name for OpenClaw send
3. mode — optional: mirror, direct, or auto
4. caption — optional text caption

Notes#

• Default preferred model is qwen-image-plus via DashScope
• If Qwen is unavailable, the script can fall back to Gemini (optional) and then Hugging Face
• If HF returns JSON/text instead of image bytes, surface the raw error clearly
• This version is intentionally simpler and more robust than the earlier fal/Gemini-only attempts