Recorded proof run · July 2026
When translating no image is the safer choice
A 12-slide language-and-cognition deck tests whether a visual system can decline unsuitable edits instead of forcing them.
Case study / 04Times, model token counts and estimated provider costs are from this captured proof run—not a price promise for a future job.
12slides
31translation units
15visuals checked
0forced edits
01 · What happened
Outcome
The text run translated 31 units in 3 calls and 23 seconds. The visual run evaluated 15 candidates (13 unique), translated none, and skipped all 15 as unsuitable. It recorded about $0.0094 image-analysis cost and $0.0058 text cost.
→ The visual output is intentionally almost the same size as the text-only file. Zero forced image edits is the evidence: the pipeline did not invent a win where none passed its safety threshold.
02 · Download the evidence
Download the evidence
Times, model token counts and estimated provider costs are from this captured proof run—not a price promise for a future job.
Proof cases
Translate your own deck
Run the same editable-PPTX workflow on your presentation. The first eligible text translation is free after Google sign-in.