Every AI chat product has an image ceiling and a token bill, and most people only find out about the second one after the first screenshot fails to upload. ChatGPT caps images at 20MB per image and lets you upload up to 80 files every three hours, with free accounts limited to 3 file uploads a day — so a failed multi-megabyte upload is not just slow, it burns quota. Claude accepts JPEG, PNG, GIF and WebP up to 10MB per image on claude.ai, with dimensions capped at 8000x8000 and up to 20 images per chat, and its own help center suggests using images of 1000x1000 pixels or larger. Gemini Apps takes up to 10 files per prompt with non-video files up to 100MB each. Underneath all three sits a token cost that scales with pixels: Claude counts an image as 28x28 pixel patches, Gemini tiles anything bigger than 384px into 768px tiles at 258 tokens a piece. Uploading a 4032x3024 phone photo to ask one question about a receipt wastes both time and tokens. Skwish compresses and resizes images locally in your browser with a live before/after readout per file, so you can drop a screenshot set to a sane pixel box and a fraction of the bytes before attaching them. Batch the set, export a ZIP. Nothing is uploaded to a server.