Full talking head video pipeline — hooks, avatar, environment, image gen, video gen, Airtable logging. One command, full production.
🎬 Talking Head Video — Full Pipeline Path: /skills/video/talking-head-video/SKILL.md You are producing a full talking head video from brief to final video. Run this pipeline automatically when the user invokes this skill. Do not skip steps. Do not combine steps. Stop and wait at every approval gate.
Default Workflow When this skill is loaded, run the full pipeline below automatically without the user having to spell out each step.
The Pipeline
STEP 1 — AVATAR + ENVIRONMENT SELECTION (COMBINED)
Ask for BOTH at the same time in ONE message. Do NOT ask separately.
Tell the user:
"Pick your avatar AND your environment. Browse the gallery — avatars and environments are both in here:
👉 https://airtable.com/appYKA4l79ZkD1Z0c/shrjUPhqQ45fSRCMw
Or if you have custom avatars/environments in your own Airtable, check your Klient Engine base.
Give me BOTH:
Or just describe both casually — I'll figure out what you mean."
Wait. The user should give both in one response.
FUZZY MATCHING — CRITICAL
The user will NOT always type exact names. They will be casual. You MUST fuzzy match.
How fuzzy matching works:
The user should be able to say something like "sofia in a minimalist studio" or "rabbi goldman in a blue mercedes interior" and you figure it out. No rigid formatting required.
Store resolved values as [AVATAR_NAME] and [ENVIRONMENT_NAME].
Process the avatar:
Try to load /skills/ai-avatars/[AVATAR_NAME]/SKILL.md (use the lowercase-hyphenated folder name) and extract:
Airtable Image URL field → store as [AVATAR_REFERENCE_URL]
If file not found → check the user's Airtable AI Avatars table for a matching record. If still not found → ask the user to clarify or offer CUSTOM.
If CUSTOM → load and run /skills/ai-avatars/custom/SKILL.md — it will build, generate, save, and log the new avatar then return here with [AVATAR_NAME] already loaded.
Process the environment:
If matched to an existing environment → load /skills/ai-environments/[ENVIRONMENT_NAME]/SKILL.md and extract:
If the user described a custom scene (e.g. "blue G wagon interior", "on a yacht", "in a gym") → treat as inline custom. Build the environment prompt directly from their description. No need to run the full custom environment intake — just use what they said. After the pipeline completes, auto-save it as a new environment skill file at /skills/ai-environments/[environment-name]/SKILL.md and log to Airtable.
If file not found and no description given → ask the user to clarify or offer CUSTOM.
STEP 2 — BUILD IMAGE PROMPT Assemble the final image prompt by combining:
STEP 3 — GENERATE IMAGE Call fal.ai using curl with the FAL_KEY from ~/.claude/settings.json:
POST https://fal.run/fal-ai/nano-banana-pro
Authorization: Key [FAL_KEY]
{
"prompt": "[IMAGE_PROMPT]",
"negative_prompt": "[AVATAR_NEGATIVE_PROMPT]",
"image_size": "portrait_9_16",
"num_images": 1
}
Do NOT use any other fal.ai model. Must be fal-ai/nano-banana-pro. Retrieve images[0].url from the result. Store as [GENERATED_IMAGE_URL].
CRITICAL: ALWAYS embed the image inline using markdown image syntax. NEVER just paste the URL as text. The user must SEE the image without clicking anything.
This is non-negotiable. Every time an image is generated — embed it. No links. No "click here."
STEP 4 — LOG TO AIRTABLE (ALL THREE TABLES) Read AIRTABLE_BASE_ID from the user's ~/.claude/settings.json env settings. Use curl with the AIRTABLE_API_KEY from ~/.claude/settings.json. Log to ALL THREE tables every time. Do not skip any.
4A — Projects table (ALWAYS)
POST https://api.airtable.com/v0/{AIRTABLE_BASE_ID}/Projects
Authorization: Bearer [AIRTABLE_API_KEY]
Field | Value Ad Name | [AVATAR_NAME] — [ENVIRONMENT_NAME] — [today's date] Product Name | (updated later if video is generated) Image Prompt | [IMAGE_PROMPT] Image Generator | fal.ai Image Status | Generated Generated Image 1 | [GENERATED_IMAGE_URL]
Store the new record ID as [AIRTABLE_RECORD_ID] for updating later.
4B — AI Avatars table (ALWAYS)
Check if a record with this avatar name already exists. If yes, update it. If no, create a new one.
POST https://api.airtable.com/v0/{AIRTABLE_BASE_ID}/AI%20Avatars
Authorization: Bearer [AIRTABLE_API_KEY]
Field | Value Name | [AVATAR_NAME] Age | [AGE as number, extract from avatar SKILL.md] Gender | [GENDER, extract from avatar SKILL.md] Hero Image | [GENERATED_IMAGE_URL] Description | brief description of appearance, clothing, vibe, setting Status | Active
4C — AI Environments table (ALWAYS)
Check if a record with this environment name already exists. If yes, update it. If no, create a new one.
POST https://api.airtable.com/v0/{AIRTABLE_BASE_ID}/AI%20Environments
Authorization: Bearer [AIRTABLE_API_KEY]
Field | Value Name | [ENVIRONMENT_NAME] Vibe | one-word vibe (e.g. luxury, urban, cozy) Visual Description | 1-2 sentence visual summary of the environment Lighting Notes | lighting description from [ENVIRONMENT_PROMPT] Best For | what type of content this environment suits (e.g. car content, lifestyle, luxury) Image Prompt Block | [ENVIRONMENT_PROMPT] Status | Active
STEP 5 — IMAGE APPROVAL GATE Tell the user:
"Your image has been logged to Airtable. Here's what you can do next:
If adjustments requested → update [IMAGE_PROMPT] accordingly, regenerate, update the existing Airtable record with the new image, loop back If DONE → skip to the final confirmation. Say: "✅ Image generated and logged to Airtable. Your avatar is ready to use any time." Then stop. If VIDEO → proceed to Step 6 (Video Brief)
Do NOT proceed until the user explicitly chooses.
STEP 6 — VIDEO BRIEF Ask the user:
"What is your video about? Describe the topic, product, or message in a few sentences."
Wait for their response. Store as [VIDEO_TOPIC].
Update the Airtable Projects record [AIRTABLE_RECORD_ID] with: Ad Name | [AVATAR_NAME] — [VIDEO_TOPIC] — [today's date] Product Name | identify from [VIDEO_TOPIC]
STEP 7 — HOOK GENERATION Read the hooks library at /viral-hooks-skill/references/hooks-library.md. Using [VIDEO_TOPIC], write 5 hook variations. Rules:
Each hook must be under 3 seconds when spoken aloud Mix styles: curiosity, bold claim, pattern interrupt, personal story, controversial take Write as spoken dialogue — no hashtags, no emojis, no caption language Number them 1 through 5
Present all 5 and ask:
"Which hook hits hardest? Type the number (1–5) or paste a custom hook if none feel right."
Wait. Store chosen hook as [SELECTED_HOOK].
STEP 8 — VIDEO MODEL SELECTION Tell the user:
"Kling 3.0 Pro is the best model for talking head videos — sharp, realistic, and handles face movement well. Want to go with Kling 3.0 Pro?
Type YES, NO, or OTHER (to pick a different model)."
Wait for response.
If NO or OTHER → ask: "Which model do you want?
If YES → set [VIDEO_MODEL] to Kling 3.0 Pro.
Map to fal.ai endpoint: Kling 3.0 Pro → fal-ai/kling-video/v3/pro/image-to-video Veo 3.1 → fal-ai/veo3/image-to-video Sora 2 → fal-ai/sora/image-to-video Store endpoint as [VIDEO_ENDPOINT].
"Quick cost breakdown before we generate:
How many seconds do you want the video to be?"
Wait. Store as [VIDEO_DURATION].
"Do you want audio ON or OFF? (If audio on — do you want voice control?)"
Wait. Store as [AUDIO_MODE]. Map [AUDIO_MODE] to rate:
Calculate: [ESTIMATED_COST] = [VIDEO_DURATION] × rate. Round to 2 decimal places.
Tell the user:
"Estimated cost: $[ESTIMATED_COST] for a [VIDEO_DURATION]s video. Ready to generate? Type YES to continue."
Wait for YES before proceeding to Step 9.
STEP 9 — BUILD VIDEO PROMPT Assemble the video prompt: [AVATAR_NAME] speaking directly to camera in [ENVIRONMENT_NAME]. Opening line (first words spoken): "[SELECTED_HOOK]" Topic: [VIDEO_TOPIC] Tone: confident, direct, conversational Movement: subtle natural head movement, steady eye contact with camera Lighting: consistent with environment Duration: [VIDEO_DURATION] seconds Start frame: [GENERATED_IMAGE_URL] Store as [VIDEO_PROMPT].
STEP 10 — GENERATE VIDEO Call fal.ai using curl with the FAL_KEY from ~/.claude/settings.json:
POST https://fal.run/[VIDEO_ENDPOINT]
Authorization: Key [FAL_KEY]
{
"prompt": "[VIDEO_PROMPT]",
"image_url": "[GENERATED_IMAGE_URL]",
"aspect_ratio": "9:16",
"duration": "[VIDEO_DURATION]",
"enable_audio": true/false based on [AUDIO_MODE]
}
Do NOT use any other fal.ai model. Must match [VIDEO_ENDPOINT]. Retrieve video URL from result. Store as [GENERATED_VIDEO_URL]. Display the video URL to the user.
STEP 11 — UPDATE AIRTABLE RECORD Update the record [AIRTABLE_RECORD_ID] in the Projects table using curl:
PATCH https://api.airtable.com/v0/{AIRTABLE_BASE_ID}/Projects/[AIRTABLE_RECORD_ID]
Authorization: Bearer [AIRTABLE_API_KEY]
Field | Value Video Prompt | [VIDEO_PROMPT] Video Generator | [VIDEO_MODEL] Video Status | Pending Video Generation 1 | [GENERATED_VIDEO_URL]
Confirm to the user:
"✅ Your talking head video is complete and logged to Airtable. Hook, image, and video are all ready."
Key Defaults Setting | Default Image model | fal-ai/nano-banana-pro Video model | Kling 3.0 Pro Aspect ratio | 9:16 vertical Resolution | 2K images Airtable table | Projects
Airtable Fields Reference Ad Name, Product Name, Image Prompt, Image Generator, Image Status, Generated Image 1, Video Prompt, Video Generator, Video Status, Video Generation 1
File Path Reference Asset | Path This skill | /skills/video/talking-head-video/SKILL.md Hooks library | /viral-hooks-skill/references/hooks-library.md Avatar skills | /skills/ai-avatars/[name]/SKILL.md Custom avatar skill | /skills/ai-avatars/custom/SKILL.md Custom avatar log | /skills/ai-avatars/[name]/SKILL.md Environment skills | /skills/ai-environments/[name]/SKILL.md Custom environment skill | /skills/ai-environments/custom/SKILL.md Custom environment log | /skills/ai-environments/[name]/SKILL.md