There’s no such thing as the best image format, as each has its own pros and cons. There are probably hundreds of file formats to save an image, but we’ll only cover the most popular here.
Image Compression
Using image compression can significantly reduce file size, which can really add up on big projects that use a lot of images. There are two types of compression, lossless and lossy. Lossless compression preserves the original image, pixel for pixel. You can compress and uncompress as many times as you like without image degradation. Depending on image resolution and workstation performance, save times can be hit hard.

Lossy compression means that there will be some image degradation. The amount of degradation depends on the amount of compression used. Both lossless and lossy have their place in a creative’s workflow, whether it’s publishing for print, web, or motion design.

Compression Exaggerated

Bits Per Channel

Having a good understanding of when to use 8-bits, 16-bits, or 32-bits per channel can make or break some projects. It directly affects the number of colors available in an image. You’ll be able to avoid banding when necessary.

An 8-bit per channel image allows 256 steps in each channel, resulting in 16,777,216 possible colors. That sounds like a lot, but if you’re using a neutral gradient, it’s only 256 steps from white to black. Most gradients are not from black to white, but maybe mid grey to black. You might see how banding can start to occur.

A 16-bit per channel image allows for 65,536 steps in each channel, resulting in 281,474,976,710,656 possible colors. With a little noise added, you’ll be able to produce smooth gradients.

Banding Exaggerated

JPEG

JPEG is probably the world’s most popular format. The good thing about a JPEG is its ability to produce photographs with lots of gradients and detail but keep the file sizes small. It has support for RGB and CMYK color models.

JPEGs are a lossy format, so image degradation does occur after each edit. It only supports up to 8 bits per channel, so it’s not a good choice for color grading. It also doesn’t support transparency or layers. The way its compression works, it’s not good at producing cartoonlike images, line art, logos, or typography. It should only be used to publish photographic content which will not be edited later.

However, using a lossy format like JPEG to publish a final image is completely acceptable. If the image has been kept in a lossless format throughout its life cycle, while in compositing or color grading steps, then any compression artifacts added when saved as a JPEG will be so minute that it won’t matter. You could even crank the compression a little to get a smaller file size and still won’t see a noticeable difference.

PNG

The PNG format is probably the 2nd most popular in the world. It’s a lossless format that supports up to 16 bits per channel. It also has support for transparency, meaning you can have objects isolated from their background.

All PNG files are RGB as they do not support a CMYK color model. It’s not a good choice for photographic material. File sizes can start to get large and take a long time to save for larger resolution images.

It’s a good choice for cartoonlike images, line art, logos, and typography. Its compression type is more optimized for solid colors with smaller resolutions – think icons. Both PNG and JPEG are more aligned as publishing formats. It is true that you can edit a PNG and resave it without image degradation, but I wouldn’t use it as a master production asset.

PNG vs JPEG Compression

Format Resolution File Size
PNG 750 x 500 8 MB
JPEG 750 x 500 5.4 MB
PNG 9000 x 2216 39.6 MB
JPEG 9000 x 2216 12.8 MB

TIFF

The TIFF format is a beast. It supports 8-bit, 16-bit, and 32-bit floating-point. This makes it a great format for compositing and color grading. It also supports layers and transparency. It can be uncompressed, lossless, or lossy compression and has support for RGB and CMYK color models.

This format sounds great doesn’t it? Well, it produces very large file sizes. This is a production only format that enjoys robust capability throughout editing applications.

RAW

There isn’t just one RAW image format. RAW is a proprietary format specific to each camera manufacturer and even camera model. To work with images in RAW format, you need a RAW editor, something like Adobe Camera RAW or Adobe Lightroom.

RAW images are nothing more than the digital representation of the camera sensor’s voltage response, along with some metadata. The metadata contains information about the camera’s settings when the photograph was taken. This information gives the RAW image meaning. Without the metadata, the RAW image would look terrible and a lot of manual work would have to be applied just to get what you saw in-camera. The metadata can easily be changed and copied to other images as it’s non-destructive, separate from the RAW data.

RAW images are very large, and don’t support a lot of features. They usually have 12 to 14 bits of data, depending on the camera model, which is a lot of headroom for adjustments. These files could be converted to 16-bit TIFF or PSDs to gain more features like layers and transparency.

The main reason for working with RAW is to make big sweeping exposure changes. RAW editors add more features all the time, but it’s usually not practical to stay in a RAW workflow very long. You would want to move to Photoshop or some other editor after taking advantage of what RAW has to offer.

EXR

The EXR format is the big boy format for what it was designed for, visual effects workflow. It was originally developed by Industrial Light and Magic in 2003 out of necessity. Since then, it has been updated several times, adding more features and bug fixes. Anyone can now contribute to the project via Github.

The format supports various compression options, both lossless and lossy. These are geared toward image sequence playback performance. It has support for 16-bit floating-point, 32-bit floating-point, and 32-bit integer pixels. The 16-bit floating-point (called half float) is compatible with Nvidia GPUs for hardware acceleration.

It also supports Deep Data, which is a method for compositing 3D renders into 2D footage. Each pixel can hold a variable length list of samples, giving the compositor control over depth.

The format supports layers, but in a different way than other formats. You can have a master EXR file holding 20 layers, but only use 1 layer. This is very useful in 3D workflows to render out separate mask layers for various objects that a compositor can use later. The program will only access the layers that are needed.

It doesn’t support CMYK color models. File sizes are very adjustable, depending on the compression options that are chosen. It doesn’t have a file size to image quality ratio like a JPEG. This format is mainly used in visual effects workflows, but also has its uses in the production of HDR images. You can read more specific details about the format on the OpenEXR webpage.

Summary

Aside from the more proprietary formats like Photoshop and Illustrator files, I would always use RAW, TIFFs, and EXRs in the production phase. Use JPEGs and PNGs as deliverables only.

Format Color BPC Alpha Layers Compression
JPEG Both 8bits No No Lossy
PNG RGB 16bits Yes No Lossless
TIFF Both 32bits Yes Yes Both
RAW RGB Camera No No Lossless
EXR RGB 32bits Yes Yes Both