Turn a messy folder of gameplay captures into a clean, diverse, store-ready screenshot set — entirely in your browser.
PromoSift helps game teams review a large capture folder without sending it anywhere. It finds technical review signals and keeps editorial choices entirely in your hands.
PromoSift never uploads, modifies, or deletes your screenshots.
Quality scores are technical heuristics, not aesthetic judgments.
A dark, low-contrast, or visually similar screenshot may still be the correct creative choice. Always review results manually before publishing.
Load the built-in synthetic sample pack in the app. It includes clear, blurry, dark, bright, blank, portrait, wrong-ratio, duplicate, pixel-art, and night-scene examples. Live demo
- Local PNG, JPEG, and WebP import: file picker, folder picker, drag-and-drop, and clipboard paste.
- Steam Store Screenshot, generic 16:9/4:3, and configurable browser-local analysis primitives.
- Resolution, aspect-ratio, blur-like variance, brightness, clipping, contrast, and near-blank review suggestions.
- SHA-256 exact-file detection plus 64-bit dHash visual duplicate clustering.
- Explainable color-histogram diversity hints and a manually overridable Suggested shortlist.
- Keep / Maybe / Reject workflow with draggable shortlist ordering.
- Original-byte ZIP export, CSV report, JSON project record (no image data), and Canvas contact sheet.
npm ci
npm run fixtures
npm run devOpen the shown localhost URL. Nothing is uploaded; the worker decodes and analyzes each selected file inside your browser.
PNG, JPEG, and WebP. PromoSift accepts at most 500 files, up to 50 MB per file, and warns when the import totals more than 2 GB.
The app uses a resized grayscale Laplacian variance as a relative blur signal; luminance distribution for dark/bright/clipping/contrast signals; and color variance for near-blank frames. It says “Review suggested,” not “bad image.” See algorithm limitations.
Exact duplicate files use SHA-256. Perceptual dHash groups visually near or similar frames by Hamming distance. It is deliberately a prompt to compare—not an automatic deletion tool.
The Steam Store Screenshot preset expects 16:9 at 1920×1080 or larger. Correct size and ratio passes; close ratios warn; small, portrait, unreadable, or incorrect-ratio images fail.
Review cards, mark Keep / Maybe / Reject, optionally use Suggested shortlist, then reorder the Keep list with drag-and-drop. Suggested shortlist never overrides manual review.
promosift-selection.zip contains selected/ originals, contact-sheet.png, selection.json, report.csv, and README.txt. The project JSON stores only hashes, metrics, choices, ordering, and configuration—never image Base64 or device paths.
No accounts, analytics, API calls, remote fonts, uploads, or cloud persistence. Read privacy details.
Pixel art, intentionally dark scenes, loading frames, very minimal art, and UI-heavy captures can trigger useful but incorrect review suggestions. PromoSift does not inspect game content and does not make aesthetic judgments.
npm run dev
npm run build
npm run previewnpm run check
npm run test:e2eGenerate and run the optional large-batch smoke fixture locally with npm run fixtures:performance. It creates 300 synthetic files (including 50 near-duplicate pairs, blur, dark, and wrong-ratio cases) under an ignored test directory, so the repository never ships a large binary benchmark pack.
- Import custom screenshot requirement presets.
- Add itch.io and social-media presets.
- Compare two shortlisted screenshot sets.
See CONTRIBUTING.md. Bug reports for false positives are especially useful.
MIT.