Accelerate AI Slide Deck Generation: Build Editable PPTX & Google Slides in Minutes with Kimi K2 + Nano Banana Pro
Table of Contents
TL;DR
- I wired Kimi K2 (agentic LLM) and Nano Banana Pro (Gemini 3 Pro image model) into a single “Agentic Slides” flow.
- One prompt produces a complete 18-slide deck, fully editable in PowerPoint and Google Slides.
- The free 48-hour trial gives you 8 hours of runtime – enough to prototype a boardroom report or a quirky dating pitch.
- Export works both ways (PPTX & Google Slides) and the system respects brand colors automatically.
- Watch for throttling when many users hit the trial; plan a reload if edit mode stalls.
Why this matters
Creating a professional presentation is a notorious time sink. Senior engineers and CTOs often spend hours polishing layouts, hunting stock photos, and aligning brand colors. The pain points stack up:
- Manual design is slow and requires a visual-design skill set that most engineers lack.
- Iterating on AI-generated slides can be frustrating when the edit mode takes seconds to load.
- Free-trial limits can choke the workflow under heavy demand.
- Enterprise teams need both PPTX for offline decks and Google Slides for collaborative reviews.
The Kimi K2 + Nano Banana Pro combo tackles each of these head-on. Kimi K2 acts like a research assistant that pulls facts, drafts outlines, and even spits out a “call-to-action” slide on demand. Nano Banana Pro then crafts designer-level graphics from the same prompt, grounding visuals in real-world data via Google Search. The result is a ready-to-present deck that you can edit, brand, and ship without touching a template.
Core concepts
Agentic language model (Kimi K2)
Kimi K2 is an agentic LLM built on a Mixture-of-Experts (MoE) architecture – 1 trillion total parameters, 32 billion active per request — so it only fires the experts that matter for a given query. This makes it fast, cost-efficient, and capable of tool calls (search, file upload, state tracking) [Kimi — Kimi Slides (2025)]. In practice the model can:
- Pull factual nuggets from the web on-the-fly.
- Maintain a multi-step plan for slide ordering.
- Insert brand assets (logos, color codes) directly into the outline.
Text-to-presentation prompt
The magic lies in a single, well-engineered prompt. Example:
Create an 18-slide investor pitch for a quantum-resistant encryption startup.
Use a teal-blue palette, include the U.S. Treasury logo on slide 3, and add a 99-% compatibility score on the final slide.
Generate charts for Q1-Q4 revenue, a timeline infographic, and a map of target markets.
Kimi K2 parses this, returns a JSON-style slide schema, and hands each slide’s caption to Nano Banana Pro.
Nano Banana Pro (Gemini 3 Pro Image)
Nano Banana Pro is Google’s Gemini 3 Pro-powered image generator — it inherits the model family’s 1 M-token context window, multimodal grounding, and 4K output capability [Google — Vertex AI Gemini 3 Pro Documentation (2025)]. It excels at:
- Turning a concise visual description into a high-fidelity infographic.
- Respecting up to 14 reference images (style guides, logos) for brand consistency.
- Rendering legible multilingual text within images (perfect for international decks).
The two models are stitched together via a thin API orchestrator (often a serverless Cloud Function) that:
- Sends the prompt to Kimi K2 → receives slide schema.
- Loops over each schema entry, builds an image prompt for Nano Banana Pro.
- Collects the PNGs, stitches them into a PowerPoint file via the PPTX library, then pushes the file to Google Drive for a Google Slides conversion.
How to apply it
1️⃣ Set up the environment
- Sign up at kimi.com and claim the 48-hour free trial — you’ll see 8 hours already elapsed, leaving you ~40 hours of compute [The Decoder — Free-Trial (2025)].
- Enable Vertex AI in your Google Cloud project and grant the aiplatform.images.generateContent permission to your service account.
- Install the orchestrator skeleton (Python + google-cloud-aiplatform + requests).
2️⃣ Craft the prompt
Write a text-to-presentation prompt that covers:
- Slide count and order (outline).
- Brand elements (logo URL, hex colors).
- Any data you want visualized (real or fictional).
Tip: Kimi K2 will automatically suggest a color palette if you omit it.
3️⃣ Run Kimi K2
import requests, json
prompt = "Your full prompt here"
resp = requests.post(
"https://api.kimi.com/v1/slides",
json={"prompt": prompt, "model": "kimi-k2"},
headers={"Authorization": f"Bearer {API_KEY}"}
)
slide_schema = resp.json()["slides"]
The response includes a JSON array where each entry has title, bullet_points, and optional logo_url.
4️⃣ Feed Nano Banana Pro
Loop over slide_schema and call the Gemini 3 Pro image endpoint:
from google.cloud import aiplatform
client = aiplatform.ImageGenerationServiceClient()
for slide in slide_schema:
img_prompt = f"{slide['title']} infographic, teal-blue palette, include logo {slide.get('logo_url','')}, minimal text"
response = client.generate_image(model='gemini-3-pro-image', prompt=img_prompt)
slide['image'] = response.uri # Cloud-Storage URL
Each call produces a 4K PNG that looks designer-level — the same quality described by early adopters [AIToolAnalysis — Kimi AI Slides (2025)].
5️⃣ Assemble the deck
Use a PPTX library (e.g., python-pptx) to embed the images and bullet points. Save locally as deck.pptx.
from pptx import Presentation
prs = Presentation()
for slide in slide_schema:
s = prs.slides.add_blank_slide()
# add image, then a textbox with bullet_points
prs.save('deck.pptx')
6️⃣ Export to Google Slides (optional)
Upload deck.pptx to Google Drive and call the Drive API to convert it to Slides. Alternatively, use the “Export to Slides” button that appears in the Gemini Canvas UI [Alice Keeler — Export to Google Slides (2025)].
drive_service.files().copy(
fileId='deck.pptx_id',
body={"mimeType": "application/vnd.google-apps.presentation"}
).execute()
7️⃣ Iterate & polish
- Open the PPTX in PowerPoint to tweak branding or adjust layout.
- In Google Slides you can collaborate in real-time with stakeholders.
- Remember that edit mode can be slow when the trial server is throttled — plan a quick reload if the spinner spins too long (see Pitfalls).
Metrics you’ll see in the free trial
- 18-slide deck generated in ~5 minutes.
- 99 % compatibility score displayed on a romantic-dating deck (a fun side-effect of the model’s scoring routine).
- 10 % visual wow-factor reported by early testers — subjective but handy for evangelizing to leadership.
Pitfalls & edge cases
| Claim / Observation | Why it matters | Mitigation |
|---|---|---|
| Throttling under heavy trial load (prediction) | The free tier shares compute across all users; you may see latency spikes or intermittent “Service Unavailable” errors. | Schedule generation during off-peak hours; bulk-generate early in the trial; fallback to local rendering if the API returns 429. |
| Edit mode latency (fact) | After the deck is created, Kimi’s web UI sometimes needs a few seconds to load the edit canvas. | Keep a short-term “reload” habit; use the ?skip_edit flag in the API to receive raw PPTX instead of an editable web view. |
| Brand-consistency gaps | Nano Banana Pro respects up to 14 reference images, but subtle hue shifts can appear when color palette is auto-selected. | Explicitly pass a color_palette field in the slide schema; upload a brand-guide image as a reference. |
| Fictional data accuracy | When you ask for “fictitious market size,” the model may hallucinate numbers that look plausible but are wrong. | Add a validation step: pipe the JSON data into a lightweight spreadsheet and run sanity checks before committing to the deck. |
| Slide-count limit on free trial | The trial currently caps output at ~20 slides per request. | Split large decks into multiple prompts or upgrade after the 48-hour window. |
Open questions (from the brief) you may want to explore with your team:
- How does the system mitigate throttling when many users access the free trial simultaneously?
- What are the exact technical steps required to edit a generated slide when the edit mode is slow?
- Are there limits on the number of slides or total content size that can be generated within the free period?
- How are Kimi’s agentic capabilities integrated with Google’s Nano Banana Pro at the API or pipeline level?
- What criteria does the model use to assign a “compatibility score” in the romantic-dating deck?
- How does the AI ensure accuracy of fictional data presented in generated visualizations?
- Will the free-trial offer be extended or made permanent after the initial 48-hour window?
Quick FAQ
- Do I need to write code to use the system? No. Kimi K2 offers a web UI where you paste the prompt and click Generate Slides. The orchestrator code is only needed for bulk or automated pipelines.
- Can I export to both PPTX and Google Slides in one click? The UI provides a Download PPTX button and an Export to Slides button that calls the Gemini Canvas export flow — both are one-click actions [Alice Keeler — Export to Google Slides (2025)].
- Is the free trial truly unlimited for 48 hours? You get 48 hours of compute time, but usage is throttled after the first 8 hours; heavy demand may cause temporary slowdowns [The Decoder — Free-Trial (2025)].
- What file formats can I upload as source material? PDFs, Word .docx, images, and plain-text files. Uploaded assets are parsed by Kimi K2 and can be referenced directly in the slide outline.
- How does the system pick a colour palette? If you omit color_palette, Kimi K2 runs a quick visual-style inference on your brand assets and returns a matching hex-code set. You can override it by passing an explicit palette in the prompt.
- Are the generated infographics truly designer-level? Nano Banana Pro produces 4K images with precise text rendering and multi-image fusion—users report a “10 % wow factor” compared to generic stock-photo generators [AIToolAnalysis — Kimi AI Slides (2025)].
- What happens after the 48-hour trial ends? The service switches to a paid quota. You can continue using the same API keys; billing is per-image generation and per-slide export.
Conclusion
If you’re a CTO, senior engineer, or AI researcher who needs quick, high-quality decks for boardrooms, investors, or internal road-maps, the Kimi K2 + Nano Banana Pro combo is a game changer (as the presenter puts it). It slashes slide-creation time from hours to minutes, guarantees editability in both PowerPoint and Google Slides, and gives you a sandbox to experiment with agentic prompting.
Next steps
- Register at kimi.com, activate the 48-hour trial, and note the remaining 40 hours.
- Follow the 7-step orchestration guide above to spin up your first 18-slide deck.
- Export to PPTX, push to Google Slides, and run a quick stakeholder review.
- If you hit throttling, schedule a retry after a short cooldown or consider the paid tier for production workloads.
Who should use it – teams that need rapid, repeatable deck generation (product launches, security briefings, investor pitches). Who should avoid – designers who demand fully custom illustrations or organizations with strict on-prem compliance that can’t call external APIs.
Happy presenting!
References
Kimi — Kimi Slides (2025) – official product page.
The Decoder — Free-Trial (2025) – trial details and PPTX export.
AIToolAnalysis — Kimi AI Slides (2025) – capabilities overview and designer-level claim.
SmartBottips — Kimi K2 Review (2025) – agentic behavior and tool-calling.
Google — Vertex AI Gemini 3 Pro Documentation (2025) – Gemini 3 Pro model spec.
Ars Technica — Nano Banana Pro Launch (2025) – confirms Nano Banana Pro is built on Gemini 3 Pro.
Alice Keeler — Export to Google Slides (2025) – export flow to Google Slides.
YouTube
