Examples

Layer 1: Generate an image using the DALL-E provider

claude-studio provider test dalle -t image -p "A steaming cup of artisan coffee on a rustic wooden table, morning sunlight streaming through a window, warm golden highlights, professional food photography, shallow depth of field" --live

Testing DalleProvider...
Prompt/Text: A steaming cup of artisan coffee on a rustic wooden table, morning sunlight streaming through a 
window, warm golden highlights, professional food photography, shallow depth of field

╭────────────────────────────────────────────────── Test Result ──────────────────────────────────────────────────╮
│ ✓ Generation successful!                                                                                        │
│                                                                                                                 │
│ Result: ImageGenerationResult(success=True,                                                                     │
│ image_url='https://oaidalleapiprodscus.blob.core.windows.net/private/org-ihcXOEpn8fNNTnQOSctFD7wY/user-qPY9Ebd0 │
│ HzZUA2k2xqy3cvF0/img-5tPQpsir93TIsaJx5Iht2UsU.png?st=2026-01-31T00%3A25%3A30Z&se=2026-01-31T02%3A25%3A30Z&sp=r& │
│ sv=2024-08-04&sr=b&rscd=inline&rsct=image/png&skoid=7daae675-7b42-4e2e-ab4c-8d8419a28d99&sktid=a48cca56-e6da-48 │
│ 4e-a814-9c849652bcb3&skt=2026-01-31T01%3A25%3A30Z&ske=2026-02-01T01%3A25%3A30Z&sks=b&skv=2024-08-04&sig=9Hbdq%2 │
│ Bg6CeN/V8ostPi06u7xQtK5HUNXg3CCOwCdxGo%3D', image_path=None, width=1024, height=1024, format='png', cost=0.04,  │
│ error_message=None, provider_metadata={'model': 'dall-e-3', 'quality': 'standard', 'style': 'vivid',            │
│ 'revised_prompt': 'A cup of steaming, freshly brewed artisan coffee sits on a rustic wooden table. The soft     │
│ morning sunlight shines through an adjacent window, casting warm golden highlights upon the scene. The depth of │
│ field is shallow, a technique often used in professional food photography, infusing the image with a sense of   │
│ focused tranquility while the background gently blurs.', 'created': 1769822731})                                │
╰─────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
⠏ Complete!

image of a coffee cup

Layer 3: Generate video from image using Luma

claude-studio test-provider luma -p "Steam rises gently from the coffee cup, morning light shifts slowly across the wooden table, peaceful cozy atmosphere" -i "https://raw.githubusercontent.com/aaronmarkham/claude-studio-producer/main/docs/screenshots/coffee_layer1.png" -d 5

Result:

This example takes the DALL-E generated coffee image and brings it to life with Luma’s image-to-video generation, adding realistic steam effects and dynamic lighting.

Layer 4: Add narration using ElevenLabs TTS

claude-studio test-provider elevenlabs -t "A perfect morning begins with the gentle aroma of freshly brewed coffee. Watch as delicate wisps of steam rise and dance in the golden morning light, creating a peaceful moment of tranquility before the day begins." --voice lily --live

Result:

This adds a beautiful narration using ElevenLabs’ Lily voice, completing the multi-sensory coffee experience. The TTS system converts text to natural-sounding speech with emotional expression and proper pacing.

Layer 5: Combine video and audio

The render mix command provides three different fit modes for handling video/audio length mismatches. Here’s a comparison:

claude-studio render mix docs/videos/coffee_layer3.mp4 --audio docs/videos/coffee_narration_lily.mp3 -o docs/videos/coffee_final.mp4 --fit speed-match

Best for: This mode slows down video playback to match audio duration, keeping steam animation flowing smoothly throughout. Creates a meditative, cinematic feel.

Alternative: Freeze-Frame Mode

claude-studio render mix docs/videos/coffee_layer3.mp4 --audio docs/videos/coffee_narration_lily.mp3 -o docs/videos/coffee_final_longest.mp4 --fit longest

Best for: When you want the last frame as a static backdrop. Steam stops halfway, then holds on final frame while narration completes.

Alternative: Shortest Mode

claude-studio render mix docs/videos/coffee_layer3.mp4 --audio docs/videos/coffee_narration_lily.mp3 -o docs/videos/coffee_final_shortest.mp4 --fit shortest

Best for: When you want video and audio to end together. Cuts off narration mid-sentence at 5 seconds when video ends.

Comparison Summary:

Mode Video Length Audio Length Result File Size
speed-match ✓ Slowed to match audio Full narration Smooth, continuous animation 1.4 MB
longest Extended with freeze Full narration Animation stops, frame freezes 1.2 MB
shortest Original 5s Truncated at 5s Both end together, cuts narration 2.1 MB

Recommendation: Use --fit speed-match for this type of content where continuous motion enhances the viewing experience.

Full Production Pipeline with Automatic Mixing

The most powerful workflow: generate video, audio, and automatically mix them into a final output. No manual mixing required!

Audio-Led Production (Podcast Style)

# Audio-led production where narration drives the timeline
claude-studio produce "The history of espresso" --style podcast --budget 10 --live --provider luma

# What happens automatically:
# 1. ScriptWriter breaks down the concept into scenes
# 2. VideoGenerator creates videos for each scene (Luma)
# 3. AudioGenerator creates narration for each scene (ElevenLabs)
# 4. QAVerifier analyzes quality with Claude Vision
# 5. EditorAgent creates edit decision list with best takes
# 6. Automatic mixing: Each scene's video + audio → mixed scene
# 7. Concatenation: All mixed scenes → final_output.mp4

Output Structure

After a complete production run, your artifacts directory contains:

artifacts/run_20260131_143022/
├── video/
│   ├── scene_001_var_0.mp4       # Generated video clips
│   ├── scene_002_var_0.mp4
│   └── scene_003_var_0.mp4
├── audio/
│   ├── scene_001.mp3              # Generated narration
│   ├── scene_002.mp3
│   └── scene_003.mp3
├── mixed/                         # ← NEW: Individual mixed scenes
│   ├── scene_001_mixed.mp4        # Video + audio combined
│   ├── scene_002_mixed.mp4
│   └── scene_003_mixed.mp4
├── final_output.mp4               # ← NEW: Final concatenated output
├── edl.json                       # Edit decision list with audio URLs
├── metadata.json                  # Production metadata
└── qa_results.json                # Quality analysis scores

Production Modes

The system supports two production workflows:

Mode Timeline Driver When to Use Auto-Detected For Fit Mode
audio-led Audio duration sets timing Podcast, educational, documentary --style podcast, educational, documentary stretch
video-led Video duration sets timing Cinematic, visual-first --style visual_storyboard, default stretch

Examples with Different Modes

Audio-Led (Explicit):

claude-studio produce "Coffee brewing techniques" \
  --mode audio-led \
  --budget 15 \
  --live \
  --provider luma \
  --audio-provider elevenlabs

Video-Led (Default):

claude-studio produce "Cinematic product showcase" \
  --style visual_storyboard \
  --budget 10 \
  --live \
  --provider luma

Resuming with Audio Mixing

The resume command supports automatic mixing if it was skipped:

# Resume from editing stage and automatically mix audio+video
claude-studio resume run_20260131_143022 --from-step editor

If audio files exist in artifacts/<run_id>/audio/, the resume command will:

  1. Load scene audio files
  2. Pass them to the EditorAgent
  3. Perform automatic mixing pipeline
  4. Generate final_output.mp4

Fit Modes for Duration Mismatches

When video and audio durations don’t match exactly:

Manual Override

You can still use the render mix command for manual control (see Layer 5 above), but the automatic pipeline handles 99% of use cases with intelligent defaults based on your production mode.