Albeit there are many image input file formats that could be fed to the renderer, and many that could be used to output images from the renderer, only few should really be used for performance, efficiency and reliability. This mini article helps you make the proper choice on a type-basis.

Format	Bits	Type	Has alpha/channels?	Default Color Space	Compression (bold: suggested)	Suggested use
TGA	8 → 24	int	no	sRGB	RLE	avoid usage (bad compression). Use EXR or TIF instead
TGA	32	int	yes	sRGB	RLE	avoid usage (bad compression). Use EXR or TIF instead
EXR	16	half	yes	linear	ZIP, ZIPS, PIZ	color, bump/normal/anisotropy, displacement, hdri environment, matte painting, output images (including deep)
EXR	32	float	yes	linear	ZIP, ZIPS, PIZ	displacement, hdri environment (use EXR HALF rather than HDR FLOAT when possible), output images (including deep)
HDR	32	float	no	linear	uncompressed	hdri_environment (use EXR when possible)
TIF	8	int	yes	sRGB	LZW	color, bump/normal/anisotropy, matte paintings (use EXR when possible)
TIF	16	int	yes	sRGB	ZIP	color, bump/normal/anisotropy, matte paintings (use EXR when possible)
TIF	32	int	yes	sRGB	ZIP	avoid usage (file size too big: use EXR instead)
TIF	32	float	yes	linear	ZIP	avoid usage (file size too big: use EXR instead)
PNG	8	int	yes	sRGB	LZW	color, bump/normal/anisotropy (use EXR when possible)
PNG	16	int	yes	sRGB	LZW	avoid usage  (use EXR instead)
JPG	8	int	no	sRGB	DCT	avoid usage  (compression artifacts)
PSD	8 → 16	int	yes	sRGB	Adobe	avoid usage (proprietary, layer evaluation cost)
PSD	32	float	yes	linear	Adobe	avoid usage (proprietary, layer evaluation costs, float data out of PhotoShop should never be trusted)

Important Notes

• integer textures are always sRGB. TIF and PNG texture painted/created in Photoshop are typically integer sRGB images.
• float and half textures are always linear. They are typically created in packages which can handle linear/float data such as Krita and Nuke.
• For a proper linear workflow, color textures must always be linearized before rendering, see:
  • Color management in 3Delight for Maya 2015 and older
  • Color management in 3Delight for Maya 2015 Extension and newer
  • Preparing textures for rendering
• "Non-color" data, such as float displacement maps, vector displacement maps, bump maps, normal maps, anisotropy maps should always be tagged as linear. Note that 3delight for Maya automatically interprets displacement maps as linear data when color management is active, however bump maps, normal maps, anisotropy maps, environment maps are not, and should be manually tagged as linear.
• tga (targa) files should never be used because they offer bad compression (RLE). Use tif with LZW compression instead of TGAs.
• 16 and 32 bit images are larger than 8 bit images, and should be used only when higher precision is actually really necessary.
• jpg textures always have compression artifacts due to DCT compression and should be absolutely never used as input textures.
• psd textures should not be used in Maya (nor in Nuke) due to the additional cost in reading through multiple layers. Instead of using psd save out layers separately as tif files (with LZW compression).
• Outputting multi-layer exr files from the renderer is possible, simply use the same file name in Default Image Fle Name of the render pass, for instance:
  • 3delight/<scene>/image/<scene>_<pass>_#.<ext>
• Outputting multi-part exr files (a mix of deep and non-deep image layers) into a single file is currently not supported, for the time being please output deep exr image files separately from non-deep multi-layer exr image files. This can be done easily in the render pass by duplicating the RGBA image layer and set an override to deepexr: