AI-generated materials have two fatal weaknesses: "they cannot be edited" and "the layout shifts when converted to PPT." Actually, both can be solved with a single prompt.
Today, I'm leaving the entire procedure here in a copy-pasteable state. (Saving this is a must!)
Why is this necessary?
First of all, why can't AI materials or PDFs be edited?
It's because the content is just "a single image." Even if it looks like a slide, the reality is the same as a photo. That's why you can't grab text or shapes no matter how hard you try. Even if you use a conversion tool to make it a PPT, the layout usually falls apart.
This slowly takes a toll. Reference materials from other companies, AI-generated drafts, past PDF documents... you want to reuse them, but you end up recreating them from scratch every time. It's a subtle time-sink.
What we are doing this time is rebuilding that single image into a "PPT where the appearance remains the same but the content is editable."
Below, the top is the original image/PDF, and the bottom is after PPT conversion. In the bottom one, text and shapes can be selected individually—meaning they are in an editable state.
When you can do this, the world changes.
Any image or PDF slide becomes your own "editing material." You'll be able to prepare a draft for a proposal in just 15 minutes. (Actual manual work time is about 5 minutes.)
Tutorial: Converting Image/PDF to PowerPoint
It might seem difficult, but this is all you have to do.
Leave it to the AI; you won't write a single line of code.
- Preparation: Get what you need
- 1. Give the material to the AI
- 2. Paste the conversion prompt
- 3. Generate the .pptx
- 4. Open and check if it's editable
- 5. Replace numbers and proper nouns
I will go through this flow one by one with screenshots.
Preparation: What you need
Only three things.
- The material you want to convert (Image or PDF)
- An AI that can execute code, such as ChatGPT or Claude. *I used ChatGPT.
- PowerPoint (Google Slides is also fine)
The key is point number 2. Choose an "AI that can write and run code" rather than just an "AI that writes text." Since this conversion involves writing code in the background to assemble the .pptx, you need the type that can generate and provide files.
Step 1: Give the material to the AI
Prepare the slide you want to convert as an image or PDF and attach it to the AI.
Tip: 1 slide = 1 image yields the cleanest results. For multi-page PDFs, it's safer to try one page at a time first.
Step 2: Paste the conversion prompt
This is the main event. Copy the prompt below in its entirety, insert your material at the <input> position, and send it.
It looks long, but there are almost no parts you need to touch. Just insert the material, and only touch the Color Palette if you want to change colors. That's it.
Why this prompt works: These 3 points
- Preventing text from being turned into images This allows you to retype text character by character later. This is the most common cause of conversion failure, so we prioritize banning it.
- Drawing dotted/dashed frames as "dashed outlines" This is the core trick this time. AI often tries to replicate dotted lines by "arranging small dots at equal intervals," which leads to shaky spacing or crushed corners. Instead, by using a method that draws a dashed line on a shape with no fill (specifying prstDash), it doesn't break.
- The AI "looks and fixes it itself" Once output, the AI renders it once, compares it with the original, and finds and fixes any misalignments itself. By running this cycle a few times, it finishes while "checking the answers" rather than just a one-shot output. This makes layout shifts less likely.
More in-depth details (verification methods, how to suppress shifts) are summarized in the "For those who want even higher accuracy" section in the latter half.
Step 3: .pptx is generated
The AI assembles the code in the background and exports an editable pptx. Download the resulting file.
By the way, the PPT generation time was 5 minutes.
Step 4: Open and confirm it's "editable"
Once opened in PowerPoint, try selecting all with Ctrl+A (⌘+A on Mac). If selection boxes appear for each text and shape, it's a success.
It has become a "collection of touchable parts, not an image" ↓↓
Step 5: Replace numbers and proper nouns
The rest is normal PowerPoint work. Change company names and numbers to your own and you're done. For complex materials, some minor shifts may occur, so make slight manual corrections.
For those who want even higher accuracy (Troubleshooting, Finishing, Support)
How to handle things when they don't go well, and for those who want to go the extra mile. Fix one thing at a time.
Problem → Solution
- Text is not editable / is an image → Add "Banning image conversion of text, make all text editable" and regenerate.
- Strange gaps between Japanese and alphanumeric characters → Say "Specify the same font for both eastAsia and latin."
- Dotted frames are shaky / corners are crushed → "Specify with prstDash, ban dot arrangement." *Note: PNGs might look broken; make the final judgment in PowerPoint.
- Elements are shifted / periods are cut off → Regenerate with "Normalize coordinates by original image ratio" and "Wider right margin."
Finishing Tips
- Stop re-prompting after 1-2 cycles (after that, you're just chasing errors).
- Apply Bold only to headings, English labels, and accent colors (don't let the AI randomly bold words in the body text).
Extra: Auxiliary Prompts
Only after the main part is successful. Two things: unifying color schemes and changing ratios.
(A) Unify color schemes
(B) Change ratios (4:3 ⇔ 16:9)
Closing
Thank you for reading.
From now on, I will continue to post about:
- AI x Documents
- Tips for creating materials
- AI x Business utilization (automatic email replies, automatic quote creation, automatic Salesforce entry) using Claude/ChatGPT/Copilot, etc.
Thank you for your support.
