PixelstoLayers
Turning Generated Infographics into Editable Assets
Pick the right granularity. Edit the text that matters. Keep everything else.
CVPR Demos 2026
by Abhay Bhandarkar · Vineeth N Balasubramanian
PixelstoLayers
Workspace

Drop your infographic

Any format. Any number of pages. We'll figure out the rest.

Preparing your slides

Rendering pages. No GPU work yet.

  1. Reading the file
  2. Pages rendered
  3. Generating thumbnails
  4. Ready
PixelstoLayers
Library

Your slides

Click any slide to extract its visual parts and editable text. Pages you don't open stay untouched.

PixelstoLayers
Workspace

Save your work

PPTX gives you a layered, editable presentation. SVG gives you real text where you asked for it, on a single page.

PixelstoLayers
Help

How Pixels to Layers works

A six-step flow that turns a flat infographic into editable parts.

  1. 1 · Upload

    Drop a PPTX, PDF, or image. Multi-page files appear in the Library where you pick one to work on.

  2. 2 · Decompose

    SAM finds visual parts. Drag the slider to control granularity — the ★ shows our recommended setting. Hover any part to preview it.

  3. 3 · Merge similar parts optional

    Flip the toggle to collapse overlapping detections into one cleaner set, and to let CLIP label them (Logo, Chart, Icon, Photo…).

  4. 4 · Confirm

    Locks the layout. Unlocks Save and the text editor. You can still re-decompose later — confirming isn't final.

  5. 5 · Make text editable optional

    EasyOCR finds every text region. Tick the ones you want as real PowerPoint text frames — the rest stay in the background.

  6. 6 · Save

    PPTX — every part as its own movable layer. SVG — promoted text as real <text>, visuals as image patches.

Controls at a glance

Granularity slider

How fine-grained the detection should be. Ultra-Coarse finds bigger blocks; Ultra-Fine finds tiny icons. The starred tick is the RL agent's recommendation.

Merge similar parts

Off by default. When on, we collapse overlapping SAM detections and label what survives. Use this when you have too many redundant boxes.

Show layers

Reveals the detection boxes on the canvas. Hover any to preview the actual extracted part. The smallest part wins the hover.

Pick editable text

After confirming, click Make text editable. A picker lists every detected text region with its OCR'd content — edit the text inline before committing.

Confirm slide

The "lock it in" button. Materializes the parts you see as actual PowerPoint shapes. Required before Save.

Save

Two formats. PPTX keeps every part as a separate movable layer plus your editable text frames. SVG is a single web-ready file with the same hybrid structure.

Tips