PowerPoint Agent

Creates professional PowerPoint presentations with:

•Full-slide AI images - Each slide is an AI-generated image that fills the entire slide
•16:9 widescreen format - Standard presentation dimensions (10" x 5.625")
•Brand consistency - Reference images (logos, colors) applied to all slides
•Rolling style references - Previous slides used as references for consistent styling
•Editable output - Standard .pptx file that can be opened in PowerPoint, Google Slides, etc.

Output Format

code

slide_01.png --> Full-slide image in PowerPoint
slide_02.png --> Full-slide image in PowerPoint
...

Final output: .pptx file with slide titles in speaker notes

Prerequisites

Before generating, verify:

•FAL_KEY in .env file (get from https://fal.ai/dashboard/keys)
•python-pptx installed: pip install python-pptx
•ANTHROPIC_API_KEY in .env (optional, for validation)

Check prerequisites:

python

from scripts import print_prerequisites_status
print_prerequisites_status()

Available Reference Images

Located in Reference Images/ folder:

•Abonmarche Primary Logo Full Color Transparent BG.png
•Abonmarche A Transparent BG.png
•COP Logo.png

Workflow

Step 1: Gather Information

Ask the user for:

•Topic/subject of the presentation
•Number of slides (default: 5)
•Reference images - which images from Reference Images folder to use for branding
•Rolling references - use previous slides to maintain style (default: yes)

Step 2: Generate Slide Content

For each slide, define:

•title: Heading text displayed on slide
•bullet_points: 3-4 brief points (3-6 words each)
•visual_description: Description of the icon/graphic

Example for 3 slides:

python

slides = [
    SlideContent(
        title="Welcome to Abonmarche",
        bullet_points=["Engineering Excellence", "Innovation in Design", "Trusted Partner"],
        visual_description="Professional title slide with Abonmarche logo prominently displayed",
        is_title_slide=True,
    ),
    SlideContent(
        title="Our Services",
        bullet_points=["Civil Engineering", "Land Surveying", "Environmental Consulting"],
        visual_description="Simple line-art icons representing engineering services",
    ),
    SlideContent(
        title="Contact Us",
        bullet_points=["Visit abonmarche.com", "Call for consultation", "Partner with us"],
        visual_description="Contact information with simple phone/web icons",
    ),
]

Step 3: Create Configuration

python

from scripts import PPTXConfig, SlideContent

config = PPTXConfig(
    slides=slides,
    reference_images=[
        "Reference Images/Abonmarche Primary Logo Full Color Transparent BG.png"
    ],
    use_rolling_references=True,
    rolling_reference_count=1,
    validate_slides=True,
    output_path="./output/presentation.pptx",
    presentation_title="Abonmarche Overview",
)

Step 4: Generate PowerPoint

python

from scripts import create_powerpoint

result = create_powerpoint(config)
print(f"PowerPoint saved: {result}")

Style Reference Feature

The skill supports two types of reference images:

Static References

•Logos, brand colors, style examples
•Applied to ALL slides
•Uploaded once to fal.ai

Rolling References

•Previously generated slides used as reference
•Slide 1: Uses static references only
•Slide 2+: Uses static references + previous slide(s)
•Helps maintain consistent styling across the deck

Configuration:

python

config = PPTXConfig(
    slides=slides,
    reference_images=["logo.png"],      # Static refs
    use_rolling_references=True,         # Enable rolling refs
    rolling_reference_count=1,           # Include last 1 slide
)

Cost Estimate

Item	Cost
Slides	$0.15/image (Nano Banana Pro)
Validation	~$0.001/slide (Claude Haiku 4.5, optional)

5-slide presentation: ~$0.75-0.80 (much cheaper than video!)

python

from scripts import estimate_cost
print(estimate_cost(num_slides=5))

Example Prompts That Trigger This Skill

•"Create a PowerPoint about renewable energy"
•"Make me an editable presentation for our team meeting"
•"Generate a PPTX file about civil engineering"
•"Build a slide deck I can edit later"
•"Create a presentation about workplace safety that I can modify"
•"Make a 5-slide PowerPoint on our company services"

JSON Configuration

Alternatively, use JSON config file:

json

{
    "title": "Company Overview",
    "author": "John Smith",
    "reference_images": [
        "Reference Images/Abonmarche Primary Logo Full Color Transparent BG.png"
    ],
    "output_path": "./output/presentation.pptx",
    "validate_slides": true,
    "use_rolling_references": true,
    "rolling_reference_count": 1,
    "slides": [
        {
            "title": "Welcome",
            "bullet_points": ["Point 1", "Point 2", "Point 3"],
            "visual_description": "Title slide with logo",
            "is_title_slide": true
        },
        {
            "title": "Our Services",
            "bullet_points": ["Service A", "Service B", "Service C"],
            "visual_description": "Service icons"
        }
    ]
}

Run with:

python

from scripts import run_from_json
result = run_from_json("presentation_config.json")

Output Structure

code

output/
  <presentation_name>/
    slides/
      slide_01.png
      slide_02.png
      ...
    presentation.pptx

Editing Existing PowerPoints

The skill can edit existing presentations by regenerating specific slides and replacing them.

Replacing a Single Slide

When a user asks to "fix slide X" or "regenerate slide X":

•
Generate a new slide image using an adjacent slide as reference (for style consistency)
- •If fixing slide 1, use slide 2 as the style reference
- •If fixing slide N, use slide N-1 or N+1 as the style reference
•Replace the slide in the existing PowerPoint

python

from scripts import generate_slide, replace_slide_image, upload_reference_images

# Step 1: Upload the adjacent slide as style reference
ref_urls = upload_reference_images(["output/presentation/slides/slide_02.png"])

# Step 2: Generate new slide image
new_image = generate_slide(
    title="What is RAG?",
    bullet_points=["Retrieval Augmented Generation", "AI searches your docs", "Answers with YOUR data"],
    visual_description="Corporate memphis style intro slide with AI/search icon",
    slide_number=1,
    output_dir="./output/presentation/slides",
    reference_urls=ref_urls,
    is_title_slide=True,
)

# Step 3: Replace in existing PowerPoint
replace_slide_image(
    pptx_path="./output/presentation/presentation.pptx",
    slide_number=1,
    new_image_path=new_image,
    new_title="What is RAG?",
)

Replacing Multiple Slides

python

from scripts import replace_multiple_slides

replace_multiple_slides(
    pptx_path="./output/presentation.pptx",
    replacements=[
        {"slide_number": 1, "image_path": "new_slide_01.png", "title": "New Title 1"},
        {"slide_number": 3, "image_path": "new_slide_03.png", "title": "New Title 3"},
    ],
    output_path="./output/presentation_updated.pptx",  # Optional: save to new file
)

Key Principle: Style Reference from Adjacent Slides

When regenerating a slide, use an adjacent slide (not the original reference image) to maintain consistency with the rest of the deck:

•Slide 1: Use slide 2 as reference
•Slide N (middle): Use slide N-1 as reference (previous slide)
•Last slide: Use second-to-last slide as reference

This ensures the regenerated slide matches the style already established in the presentation.

Example: Fix Slide That Copied Reference Too Literally

If a slide looks too much like the reference image (copied graphics instead of creating new ones):

•Use an adjacent slide that has the correct style
•Give detailed visual_description for what NEW graphic to create
•Regenerate and replace

Comparison: PowerPoint vs Video Agent

Feature	PowerPoint Agent	Video Agent
Output	.pptx (editable)	.mp4 (video)
Cost/5 slides	~$0.75	~$2.65
Voiceover	No	Yes
Transitions	No	Yes (animated)
Editable	Yes	No
Use case	Edit/customize	Share/present