Separate pixels from file bytes
Pixel dimensions describe the sampling grid: width multiplied by height. File size describes the bytes needed to store the encoded image, often including metadata. A 1600-pixel-wide image can have many different file sizes depending on its subject, format and quality setting.
Start with a written delivery requirement. For example, “1600 × 1200 pixels, 4:3, below the receiving system's stated byte limit” tells you three separate checks. A bare instruction such as “make it 1 MB” does not establish the required dimensions or acceptable visual quality.
Resize proportionally, or crop deliberately
To preserve the aspect ratio, multiply both dimensions by the same factor. For a 4000 × 3000 source reduced to 1600 pixels wide, the factor is 1600 ÷ 4000 = 0.4. The height becomes 3000 × 0.4 = 1200 pixels.
A square placement is a different decision. Stretching the 4:3 image into a square changes the product proportions. Crop to a square if losing the edges is acceptable, or fit the complete image inside a square canvas if the whole product must remain visible. Check the receiving layout before choosing.
Compress a delivery copy
Keep the original master so you can prepare another size later. Resize a delivery copy, then compare the formats and quality controls supported by your destination. A photographic render and a sharp diagram may need different choices.
Lossy compression discards some information to reduce file size. Inspect gradients, small type, fine edges and textured surfaces before approving it. Lossless encoding preserves the encoded pixel data, but converting an already damaged source to a lossless format cannot restore the missing detail.
Check quality where the image will be used
Review the full composition at its intended display size, then zoom into the most demanding areas. Look for ringing around lettering, blocks in smooth backgrounds, damaged transparency and altered edges. A tiny preview can hide all of these.
Reopen the downloaded export and verify its dimensions, file type, transparency and bytes. If the target system still rejects it, compare its actual requirements with the exported properties. Do not keep lowering quality when the real problem is a disallowed format or incorrect aspect ratio.
Texture files need an additional check
A compressed download is not a reliable measure of decoded texture memory. For an uncompressed RGBA8 base level, width × height × 4 bytes gives the storage before mipmaps and other allocations. A 4096 × 4096 texture is 64 MiB at that level even if the source download is much smaller.
For normal, roughness and other material-data maps, compression damage can change shading. Check the material in the destination renderer, not only the flat thumbnail. Choose a format and colour interpretation supported by that pipeline.
