adaptive-memory-multi-model-router 2.14.46 → 2.14.48

package/demo/VEO3_PROMPTS.md

@@ -1,269 +0,0 @@
-# Frame-by-Frame Veo 3 Prompts for A3M Router Product Video
-# Copy each prompt into Gemini (Pro account) with Veo 3 enabled
-# Generate 8-12 second clips, then stitch with ffmpeg
-
----
-
-## FRAME 1: "Routing Decision" (8-12 seconds)
-
-### Step 1: Generate still image with Nano Banana Pro
-
-Open Gemini, paste:
-
-```
-Generate a photorealistic image of a developer laptop screen in a dark room.
-The screen shows a terminal window with dark theme (#0d1117 background).
-The terminal is running an LLM router called "A3M Router".
-Text appears showing:
-- Input: "What is 2+2?" in white
-- Routing: "→ complexity: 8/100 (TRIVIAL)" in cyan
-- Decision: "→ routed to: groq/llama-3.3-70b" in green (#3fb950)
-- Cost: "→ cost: $0.000009" in green
-A glowing blue (#58a6ff) routing graph appears beside the terminal showing
-query → [complexity analysis] → [provider selection] → response.
-The laptop sits on a wooden desk with warm desk lamp lighting.
-Shallow depth of field. 16:9 aspect ratio. Photorealistic.
-JetBrains Mono font on the terminal.
-```
-
-### Step 2: Animate with Veo 3
-
-Upload the generated image as reference, then paste:
-
-```json
-{
-  "shot": {
-    "composition": "Close-up of laptop screen, 50mm lens, shallow depth of field, warm desk lamp glow on right, cool monitor light",
-    "camera_motion": "Slow dolly in from showing the full laptop to tight on the terminal text, then subtle rack focus from the terminal to the routing graph and back",
-    "frame_rate": "24fps",
-    "film_grain": "very subtle, Kodak 5219 emulation"
-  },
-  "subject": {
-    "description": "A developer laptop showing an AI routing tool. Terminal text types in line by line. The routing path lights up with cyan and green glows as each decision is made. The routing graph on the side pulses with blue light along the decision path.",
-    "motion": "Text appears line by line as if typed. Routing path illuminates sequentially. Green checkmark appears beside the selected provider. Cost number counts up briefly then locks in."
-  },
-  "environment": {
-    "ambient_lighting": "Warm desk lamp from upper right casting soft shadows. Cool blue monitor glow illuminating the keyboard. Subtle plant shadow in background.",
-    "atmosphere": "Late night productive coding session. Dark room. Focused."
-  },
-  "audio": {
-    "type": "Subtle keyboard typing sounds as text appears. Soft ambient lo-fi beat. Gentle confirmation chime when provider is selected.",
-    "mood": "calm, focused, productive, efficient"
-  }
-}
-```
-
----
-
-## FRAME 2: "Cost Savings" (8-12 seconds)
-
-### Step 1: Still image with Nano Banana Pro
-
-```
-Generate a photorealistic image of a laptop screen showing a dark-themed (#0d1117) cost comparison infographic.
-Title at top: "Cost Per Query" in white text.
-Two vertical bar charts side by side:
-- Left bar: Tall RED bar labeled "GPT-4 (No A3M)" showing "$0.03" at the top
-- Right bar: Tiny GREEN bar labeled "A3M Router" showing "$0.0004" at the top
-Below the bars in large bold green (#3fb950) text: "213× cheaper"
-The laptop is on a dark desk. Warm desk lamp.
-Photorealistic. 16:9. Clean, minimal, Apple-style presentation.
-```
-
-### Step 2: Animate with Veo 3
-
-```json
-{
-  "shot": {
-    "composition": "Medium shot of laptop screen, slight angle, 65mm lens, shallow depth of field",
-    "camera_motion": "Static for 2 seconds, then subtle tilt down to reveal the '213× cheaper' text, then hold",
-    "frame_rate": "24fps",
-    "film_grain": "minimal"
-  },
-  "subject": {
-    "description": "A cost comparison chart on a dark screen. The red bar animates growing upward to full height with a rising red glow. Then the green bar grows to its tiny height with a satisfying green flash. The '213× cheaper' text fades in with a subtle scale animation.",
-    "motion": "Red bar rises first with ascending motion. Brief pause. Then green bar rises quickly with a satisfying pop. '213× cheaper' text scales from 0 to 100% with a subtle bounce. Numbers count up as each bar grows."
-  },
-  "environment": {
-    "ambient_lighting": "Cool monitor glow on dark desk. No warm lamp this time - pure data visualization focus.",
-    "atmosphere": "Clean, data-driven, confident. Like a financial report reveal."
-  },
-  "audio": {
-    "type": "Subtle bass hit when red bar reaches top. Higher, satisfying chime when green bar arrives. Swoosh sound as '213×' text appears.",
-    "mood": "impactful, surprising, satisfying"
-  }
-}
-```
-
----
-
-## FRAME 3: "Parallel Execution" (8-12 seconds)
-
-### Step 1: Still image with Nano Banana Pro
-
-```
-Generate a dark-themed (#0d1117) technical diagram on a laptop screen showing parallel LLM execution.
-Left side: "Your Query" in a cyan (#58a6ff) circle.
-Center: 5 arrows branching out rightward simultaneously to 5 provider boxes:
-- "Groq" with green checkmark, showing "325ms, FREE"
-- "Cerebras" with green checkmark, showing "180ms, FREE"
-- "DeepSeek" with green checkmark, showing "800ms, $0.14/1M"
-- "Mistral" with green checkmark, showing "1.2s, $2.00/1M"
-- "OpenAI" with yellow clock, showing "2.1s, $2.50/1M"
-Right side: "Best Response ✓" in green (#3fb950) highlighting the Groq result.
-Below: "Latency: 180ms (fastest provider)" in green.
-All 5 arrows animate simultaneously (not sequentially).
-Photorealistic laptop on desk. 16:9. Dark room.
-```
-
-### Step 2: Animate with Veo 3
-
-```json
-{
-  "shot": {
-    "composition": "Overhead angle looking down at laptop screen, 35mm lens, showing the full diagram clearly",
-    "camera_motion": "Subtle slow zoom in on the diagram as it progresses. Camera stays relatively static to not distract from the data.",
-    "frame_rate": "24fps",
-    "film_grain": "minimal"
-  },
-  "subject": {
-    "description": "A routing diagram on a dark screen. The 'Your Query' circle pulses cyan. All 5 arrows extend outward simultaneously (not one-by-one). Each provider box lights up as the arrow reaches it with a brief flash. Then the Groq box pulses brighter with a green glow and the 'Best Response' label appears on the right.",
-    "motion": "Query circle pulses. 5 arrows extend simultaneously left-to-right. Provider boxes illuminate as arrows arrive (Groq and Cerebras arrive first). Green highlight travels down to the selected response. '180ms' counter appears and locks."
-  },
-  "environment": {
-    "ambient_lighting": "Pure monitor glow. Dark desk. Technical presentation feel.",
-    "atmosphere": "Efficient, fast, decisive"
-  },
-  "audio": {
-    "type": "Subtle electronic/tech whoosh as arrows extend. Light ping as each provider lights up. Satisfying confirmation tone when Best Response appears.",
-    "mood": "fast, efficient, smart"
-  }
-}
-```
-
----
-
-## FRAME 4: "40 Providers" (8-12 seconds)
-
-### Step 1: Still image with Nano Banana Pro
-
-```
-Generate a dark-themed (#0d1117) terminal window on a laptop showing a provider status table.
-The table has columns: Provider | Status | Cost | Latency
-Rows visible:
-✓ groq/llama-3.3-70b      FREE        325ms
-✓ cerebras/llama-3.3-70b  FREE        180ms
-✓ deepseek/chat            $0.14/1M    800ms
-✓ mistral/mistral-large    $2.00/1M    1.2s
-✓ openai/gpt-4o            $2.50/1M    2.1s
-... (35 more rows visible but dimmed)
-All green checkmarks. Title: "40 Providers — All Online ✓"
-Dark room, desk lamp, MacBook. Photorealistic. 16:9.
-```
-
-### Step 2: Animate with Veo 3
-
-```json
-{
-  "shot": {
-    "composition": "Slight angle on laptop screen, 50mm lens, showing the provider table",
-    "camera_motion": "Camera slowly tilts down as the table scrolls, revealing more providers. Settles on the bottom showing '... 35 more' with the '40 Providers' header.",
-    "frame_rate": "24fps",
-    "film_grain": "subtle"
-  },
-  "subject": {
-    "description": "A terminal table listing AI providers. The table appears line by line with each row. Green checkmarks appear with a brief flash next to each provider name as it loads. The status shows 'All Online'. The table scrolls down smoothly to reveal more entries.",
-    "motion": "Table rows load one by one from top to bottom. Checkmarks flash green as each appears. Subtle pulse on the 'All Online ✓' badge. Smooth scroll down revealing more providers."
-  },
-  "environment": {
-    "ambient_lighting": "Warm desk lamp from side. Cool monitor glow.",
-    "atmosphere": "Impressive scale. Professional. Technical."
-  },
-  "audio": {
-    "type": "Subtle ticking/click sound as each row loads. Soft ambient hum. Final satisfying chord when full table is visible.",
-    "mood": "impressive, professional, comprehensive"
-  }
-}
-```
-
----
-
-## FRAME 5: "End Card" (8-12 seconds)
-
-### Step 1: Still image with Nano Banana Pro
-
-```
-Generate a dark-themed (#0d1117) product card on a laptop screen.
-Center: "A3M Router" in large white bold text with a subtle blue (#58a6ff) glow.
-Below in smaller text: "#1 on RouterArena · 213× cheaper than GPT-5 · 40 providers"
-Below that in a terminal-style box: "$ npm install adaptive-memory-multi-model-router"
-Bottom: GitHub logo and npm logo side by side.
-Very clean, minimal, dark. Apple-style presentation.
-Photorealistic laptop on dark desk. 16:9.
-```
-
-### Step 2: Animate with Veo 3
-
-```json
-{
-  "shot": {
-    "composition": "Centered view of laptop screen, 50mm lens, clean and symmetrical",
-    "camera_motion": "Camera slowly pushes in on the center text, creating a subtle zoom that focuses attention on the npm install command at the end",
-    "frame_rate": "24fps",
-    "film_grain": "very subtle"
-  },
-  "subject": {
-    "description": "A clean dark product card. The 'A3M Router' title fades in first with a subtle glow. Then the tagline '#1 on RouterArena · 213× cheaper · 40 providers' appears below with a brief typewriter effect. Finally the npm install command types out character by character with a blinking cursor. GitHub and npm logos fade in at bottom.",
-    "motion": "Title fades in with glow. Tagline types line by line. npm command types character by character. Cursor blinks at the end. Subtle continuous blue glow on the title."
-  },
-  "environment": {
-    "ambient_lighting": "Pure monitor glow on a dark desk. Cinematic. Clean.",
-    "atmosphere": "Confident, final, memorable"
-  },
-  "audio": {
-    "type": "Soft ambient synth pad. Subtle typing sounds as npm command appears. Final soft confirmation tone.",
-    "mood": "confident, complete, memorable"
-  }
-}
-```
-
----
-
-## STITCH ALL 5 CLIPS (run after generating all 5 videos)
-
-```bash
-# Download all 5 clips from Gemini/Veo to the demo/ folder
-# Naming: frame1-routing.mp4, frame2-cost.mp4, frame3-parallel.mp4, frame4-providers.mp4, frame5-endcard.mp4
-
-# Create concat file
-cat > demo/clips.txt << EOF
-file 'frame1-routing.mp4'
-file 'frame2-cost.mp4'
-file 'frame3-parallel.mp4'
-file 'frame4-providers.mp4'
-file 'frame5-endcard.mp4'
-EOF
-
-# Stitch all clips
-ffmpeg -f concat -safe 0 -i demo/clips.txt \
-  -c:v libx264 -preset slow -crf 18 \
-  -pix_fmt yuv420p -movflags +faststart \
-  assets/a3m-product-demo.mp4
-
-# Generate GIF version (for Twitter/GitHub)
-ffmpeg -i assets/a3m-product-demo.mp4 \
-  -vf "fps=15,scale=900:-1:flags=lanczos,split[s0][s1];[s0]palette=max_colors=128:stats_mode=diff[p];[s1][p]histogram=th=0.001" \
-  -loop 0 \
-  assets/a3m-product-demo.gif
-
-# Generate 30-second clip for ProductHunt
-ffmpeg -i assets/a3m-product-demo.mp4 -t 30 \
-  -c:v libx264 -preset slow -crf 20 \
-  -pix_fmt yuv420p -movflags +faststart \
-  assets/a3m-ph-30s.mp4
-
-echo "✅ All videos generated!"
-echo "  Full demo: assets/a3m-product-demo.mp4"
-echo "  GIF:       assets/a3m-product-demo.gif"
-echo "  PH clip:   assets/a3m-ph-30s.mp4"
-```

package/demo/VIDEO_PRODUCTION_GUIDE.md

@@ -1,333 +0,0 @@
-# A3M Router — Video Production Guide
-# Using: Nano Banana Pro (Google) + Veo 3 (Google Pro) + Asciinema + FFmpeg
-
-## AVAILABLE TOOLS (you have Google Pro access)
-
-| Tool | What it does | Access |
-|------|-------------|--------|
-| **Nano Banana Pro** | Google's image gen (in Gemini) | ✅ Google Pro |
-| **Veo 3 / Veo 3.1** | Google's video gen (in Gemini) | ✅ Google Pro |
-| **Kling 2.1 / 3** | Video gen with start/end frames | Free tier |
-| **Hailuo 2** | Video generation | Free tier |
-| **asciinema** | Terminal recording | `brew install asciinema` |
-| **VHS** | Terminal recording → GIF | `brew install vhs` |
-| **ffmpeg** | Video stitching, conversion | Already installed ✅ |
-| **ElevenLabs** | Voiceover / music | Free tier |
-
----
-
-## VIDEO 1: Terminal Demo (60s, for HN/GitHub README)
-
-### METHOD A: Asciinema (recommended for HN)
-
-```bash
-# 1. Install asciinema
-brew install asciinema
-
-# 2. Record the demo
-asciinema rec -c "bash demo/asciinema-demo.sh" demo/recording.cast
-
-# 3. Upload to asciinema.org (auto-embeds in HN comments)
-asciinema upload demo/recording.cast
-# → Returns URL like https://asciinema.org/a/ABC123
-
-# 4. Convert to GIF (for README)
-pip3 install agg
-agg demo/recording.cast assets/demo-hn.gif
-
-# OR install agg via cargo:
-cargo install agg
-agg --theme demo/recording.cast assets/demo-hn.gif
-```
-
-### METHOD B: VHS (GIF only, no asciinema needed)
-
-```bash
-# 1. Install VHS
-brew install vhs
-
-# 2. Record from tape file
-vhs demo/demo-hn.tape
-# → outputs to assets/demo-hn.gif
-```
-
-### Embed in HN post:
-```
-Asciinema: https://asciinema.org/a/YOUR_CAST_ID
-GIF for README: ![demo](assets/demo-hn.gif)
-```
-
----
-
-## VIDEO 2: Product Demo Video (60s, for ProductHunt + Twitter)
-
-### Pipeline: Nano Banana Pro → Veo 3 → FFmpeg
-
-### STEP 1: Generate key frames with Nano Banana Pro
-
-Open Gemini (you have Pro access) and use this system prompt:
-
-```
-You are generating product showcase frames for a developer tool called A3M Router.
-
-For each scene, create a photorealistic image using these specifications:
-- Dark theme (#0d1117 background)
-- Terminal/IDE aesthetic
-- Screen-recording style, like a modern developer setup
-- JetBrains Mono font, macOS-style window chrome
-- Subtle green (#3fb950) and blue (#58a6ff) accents
-- Clean, minimal, Apple-style product photography feel
-
-Generate these 5 frames:
-```
-
-Then generate each frame:
-
-**Frame 1: "Routing Decision" (0-12s)**
-```
-Nano Banana Pro prompt:
-A developer laptop screen showing a terminal. The terminal shows:
-Lines of code being routed to different AI models.
-Left side: input queries in white text.
-Right side: provider names lighting up in green (Groq, DeepSeek, OpenAI).
-Center: a routing decision tree glowing with blue nodes.
-Dark background (#0d1117), JetBrains Mono font.
-Photorealistic screen recording, 16:9, shallow depth of field.
-```
-
-**Frame 2: "Cost Savings" (12-24s)**
-```
-Nano Banana Pro prompt:
-A clean dark-themed infographic on a laptop screen.
-Title: "Cost Comparison" in white text.
-Two bar charts side by side:
-Left: "Without A3M" showing a tall red bar ($0.03/query)
-Right: "With A3M" showing a tiny green bar ($0.0004/query)
-Below: "213× cheaper" in large bold green text.
-Background: #0d1117. Font: JetBrains Mono. Photorealistic, 16:9.
-```
-
-**Frame 3: "Parallel Execution" (24-36s)**
-```
-Nano Banana Pro prompt:
-A diagram on a dark terminal screen showing parallel LLM execution.
-Left: "Your Query" in a blue circle.
-Center: 5 arrows splitting out simultaneously to 5 provider boxes (Groq, Cerebras, DeepSeek, Mistral, OpenAI) — all lighting up at once.
-Right: "Best Response" selected with a green checkmark.
-Below: "Latency: 138ms" in green text.
-Background #0d1117, JetBrains Mono, photorealistic, 16:9.
-```
-
-**Frame 4: "40 Providers" (36-48s)**
-```
-Nano Banana Pro prompt:
-A terminal window on a dark themed screen showing a table of providers.
-Rows: "groq/llama-3.3-70b  ✓ FREE  325ms", "cerebras/llama-3.3-70b  ✓ FREE  180ms", "deepseek/chat  ✓ $0.14/1M  800ms", "mistral/mistral-large  ✓ $2.00/1M  1.2s", "openai/gpt-4o  ✓ $2.50/1M  2.1s"
-... 35 more rows dimmed below.
-Green checkmarks next to each. Status: all online.
-Background #0d1117, photorealistic, 16:9.
-```
-
-**Frame 5: "End Card" (48-60s)**
-```
-Nano Banana Pro prompt:
-A dark product card on a #0d1117 background.
-Center: "A3M Router" in large white bold text.
-Below: "#1 on RouterArena · 213× cheaper than GPT-5 · 40 providers"
-Bottom: "npm install adaptive-memory-multi-model-router"
-GitHub logo and npm logo at the bottom.
-Clean, minimal, photorealistic, 16:9.
-```
-
-### STEP 2: Animate frames into video with Veo 3
-
-For each frame, open Gemini and use Veo 3 to generate 8-12 second video clips:
-
-```
-Veo 3 prompt (for each frame):
-{
-  "shot": {
-    "composition": "Close-up of laptop screen, 85mm lens, shallow depth of field, cinematic product showcase",
-    "camera_motion": "Slow dolly in from wide to tight on the terminal output, then subtle pan across the data",
-    "frame_rate": "24fps",
-    "film_grain": "very subtle, Kodak 5219 emulation"
-  },
-  "subject": {
-    "description": "A developer's laptop screen displaying an AI routing tool. The terminal shows queries being routed to different providers with green confirmation checkmarks appearing. Clean dark theme UI.",
-    "motion": "Text appears line by line as if being typed. Data highlights pulse briefly. Provider names light up sequentially with green glow."
-  },
-  "environment": {
-    "ambient_lighting": "Warm desk lamp from the right, cool monitor glow from the screen. Subtle desk plant shadow in background.",
-    "atmosphere": "Late night coding session. Dark room with focused monitor light."
-  },
-  "audio": {
-    "type": "subtle keyboard typing sounds, gentle ambient lo-fi beat",
-    "mood": "calm, focused, productive"
-  }
-}
-```
-
-**Specific Veo 3 prompts per frame:**
-
-Frame 1 → Veo 3: Camera slowly zooms in on the terminal as routing decisions light up
-Frame 2 → Veo 3: The red bar animates growing tall, then the green bar grows to show savings, "213×" text zooms in
-Frame 3 → Veo 3: Arrows animate out from center simultaneously, providers light up at the same time, best response pulses green
-Frame 4 → Veo 3: Table scrolls down showing all providers, checkmarks appear one by one with subtle sound
-Frame 5 → Veo 3: Text fades in line by line, minimal and clean, npm command types out character by character
-
-### STEP 3: Stitch with FFmpeg
-
-```bash
-# Concatenate all clips
-ffmpeg -i frame1.mp4 -i frame2.mp4 -i frame3.mp4 -i frame4.mp4 -i frame5.mp4 \
-  -filter_complex "[0:v][1:v][2:v][3:v][4:v]concat=n=5:v=1:a=0[outv]" \
-  -map "[outv]" \
-  -c:v libx264 -preset slow -crf 18 \
-  -pix_fmt yuv420p \
-  assets/a3m-product-demo.mp4
-
-# Generate GIF version for Twitter
-ffmpeg -i assets/a3m-product-demo.mp4 \
-  -vf "fps=15,scale=900:-1:flags=lanczos,split[s0][s1];[s0]palette[p];[s1][p]histogram=th=0.001" \
-  assets/a3m-product-demo.gif
-
-# Generate short clip for PH (30s)
-ffmpeg -i assets/a3m-product-demo.mp4 -t 30 \
-  -c:v libx264 -preset slow -crf 20 \
-  assets/a3m-ph-30s.mp4
-```
-
-### STEP 4: (Optional) Add voiceover with ElevenLabs
-
-```bash
-# Script for voiceover (60 seconds):
-# "Every LLM router does the same thing. Try provider A. If it fails, try B.
-#  That's sequential. And slow. A3M Router is different. It fires all providers
-#  at the same time. Scores the responses. Returns the best one.
-#  The result? 213 times cheaper than GPT-5. Number one on RouterArena.
-#  40 providers. Zero ML. Three megabyte install.
-#  npm install adaptive-memory-multi-model-router."
-```
-
----
-
-## VIDEO 3: 30-Second Social Clip (Twitter/X, LinkedIn)
-
-### Quick pipeline using existing assets
-
-```bash
-# Take the animated SVG and convert to MP4
-ffmpeg -loop 1 -i assets/hero-diagram.svg \
-  -vf "scale=1200:675,format=yuv420p" \
-  -t 30 -r 30 \
-  -c:v libx264 -preset fast -crf 23 \
-  assets/social-30s.mp4
-```
-
-Or use Veo 3 with a single frame:
-```
-Veo 3 prompt:
-Single dark-themed product card showing "A3M Router - #1 on RouterArena".
-Camera slowly pushes in. Text animates in.
-"npm install adaptive-memory-multi-model-router" types out at bottom.
-30 seconds. Dark theme. Clean. Minimal.
-```
-
----
-
-## VIDEO 4: YouTube Walkthrough (3 min, for ProductHunt + README)
-
-### Record with asciinema + screen recording
-
-```bash
-# Option 1: Full asciinema recording (3 min)
-asciinema rec demo/youtube-walkthrough.cast
-
-# In the recording:
-# 0:00-0:20 - Install and intro
-# 0:20-1:00 - Route queries (simple, code, complex)
-# 1:00-1:40 - Show cost comparison, benchmark data
-# 1:40-2:20 - Start proxy, show OpenAI compatibility
-# 2:20-3:00 - Show providers, circuit breaker, cache
-
-# Option 2: Use QuickTime screen recording
-# Open QuickTime → File → New Screen Recording → Select terminal area
-# Run: bash demo/asciinema-demo.sh
-# Record for 3 minutes
-# Export as demo/youtube-walkthrough.mov
-# Convert: ffmpeg -i youtube-walkthrough.mov -c:v libx264 -crf 18 assets/youtube-walkthrough.mp4
-```
-
----
-
-## POSTING GUIDE
-
-### Hacker News
-- Embed asciinema link directly in Show HN text
-- Also add GIF to GitHub README
-- HN doesn't support embedded video — asciinema is the best format
-
-### ProductHunt
-- Upload the 60s product demo video (MP4, under 100MB)
-- Also add the 30s clip as a secondary gallery item
-- Thumbnail: Frame 5 (end card) as 1200×675 PNG
-
-### Twitter/X
-- Upload the 30s clip as native video
-- Or post the 60s demo as a thread with GIF
-- Quote tweet yourself with the YouTube walkthrough
-
-### README.md
-- Add demo GIF at the very top (before the install command)
-- Add asciinema embed below
-- Add YouTube link in the badges section
-
----
-
-## QUICK START (fastest path to video)
-
-### 60-second terminal demo (5 minutes):
-
-```bash
-# 1. Install asciinema
-brew install asciinema
-
-# 2. Record
-asciinema rec -c "bash demo/asciinema-demo.sh" demo/recording.cast
-
-# 3. Upload (for HN embed)
-asciinema upload demo/recording.cast
-
-# 4. Convert to GIF (for README)
-pip3 install agg
-agg demo/recording.cast assets/demo-hn.gif
-```
-
-### 60-second product video (30 minutes):
-
-1. Open Gemini with Pro access
-2. Generate 5 frames using the Nano Banana Pro prompts above
-3. Use Veo 3 to animate each frame (8-12s each)
-4. Download all 5 clips
-5. Stitch with ffmpeg:
-   ```bash
-   ffmpeg -i clip1.mp4 -i clip2.mp4 -i clip3.mp4 -i clip4.mp4 -i clip5.mp4 \
-     -filter_complex "[0:v][1:v][2:v][3:v][4:v]concat=n=5:v=1:a=0" \
-     -c:v libx264 -crf 18 assets/a3m-product-demo.mp4
-   ```
-
----
-
-## CHECKLIST
-
-- [ ] Terminal demo: `asciinema rec -c "bash demo/asciinema-demo.sh" demo/recording.cast`
-- [ ] Upload asciinema: `asciinema upload demo/recording.cast`
-- [ ] Convert to GIF: `agg demo/recording.cast assets/demo-hn.gif`
-- [ ] Generate 5 product frames in Nano Banana Pro (Gemini)
-- [ ] Animate 5 clips in Veo 3 (Gemini Pro)
-- [ ] Stitch clips with ffmpeg
-- [ ] Add GIF to README.md hero section
-- [ ] Add asciinema link to HN post
-- [ ] Upload product video to ProductHunt
-- [ ] Upload 30s clip to Twitter
-- [ ] Record 3-min YouTube walkthrough