Depth Model Comparison

Oku3D turns flat video into 3D by estimating how far away everything is. Which model does that estimating decides how much detail you get and how much of your graphics card it takes. Pick a scene, put any two settings side by side, and drag the divider.

Scene

Comparison

Left of the divider

Right of the divider

farnear

Speed and memory

Click any row to make it the reference - every other row then shows how it compares to your choice. Starts on the model Oku3D uses out of the box.

ModelResolutionSpeed*vs. referenceMemory*VideoDownload
Depth Anything V2 SmallFP16The default. Fast enough to follow video on most graphics cards.252×2523.4 ms-0.08 GBKeeps up with video47 MB
⋮364×3646.6 ms-0.11 GBKeeps up with video
⋮518×51811.7 ms-0.18 GBKeeps up with video
Depth Anything V2 SmallQ4F16Same model, compressed weights - a third of the download for a slight loss of detail.252×2523.9 ms-0.05 GBKeeps up with video18 MB
⋮364×3647.1 ms-0.08 GBKeeps up with video
⋮518×51812.9 ms-0.15 GBKeeps up with video
Depth Anything V2 BaseFP16Noticeably cleaner edges than Small, still comfortable for video.252×2526 ms-0.24 GBKeeps up with video185 MB
⋮364×36411.2 ms-0.31 GBKeeps up with video
⋮518×51820.4 ms-0.44 GBKeeps up with video
Depth Anything V2 BaseQ4F16Base quality at a third of the download.252×2525.7 ms-0.13 GBKeeps up with video69 MB
⋮364×36411.1 ms-0.20 GBKeeps up with video
⋮518×51820.6 ms-0.32 GBKeeps up with video
Depth Anything V2 LargeQ4F16The largest V2. Finer separation between objects at similar distance.252×25215.1 ms-0.34 GBKeeps up with video223 MB
⋮364×36428.3 ms-0.47 GBKeeps up with video
⋮518×51856.2 ms-0.73 GBToo slow for video
Depth Anything V3 Mono LargeFP16Best overall detail - thin structures like hair and railings survive.252×25212.2 ms-0.74 GBKeeps up with video655 MB
⋮364×36418.3 ms-0.91 GBKeeps up with video
⋮518×51832.2 ms-1.09 GBKeeps up with video
⋮728×72861.8 ms-1.64 GBToo slow for video
⋮1036×1036148 ms-1.88 GBToo slow for video
Depth Anything V3 Mono LargeQ4F16V3 detail at half the download and a noticeably smaller memory footprint.252×25211.1 ms-0.44 GBKeeps up with video336 MB
⋮364×36417.2 ms-0.60 GBKeeps up with video
⋮518×51830.8 ms-0.79 GBKeeps up with video
⋮728×72863.1 ms-1.34 GBToo slow for video
⋮1036×1036151.2 ms-1.57 GBToo slow for video
Lotus-DFP16Exceptional on thin detail such as wires and branches. Built for photos - only its smallest setting keeps up with video.256×25635.1 ms-2.00 GBKeeps up with video1815 MB
⋮384×38444.2 ms-2.48 GBToo slow for video
⋮512×51277 ms-3.01 GBToo slow for video
⋮768×768231.8 ms-6.46 GBToo slow for video
⋮1024×1024564.1 ms-12.6 GBToo slow for video
Depth Anything V1 SmallQ4F16The first generation, kept for comparison. V2 Small is better at the same cost.252×2524.2 ms-0.07 GBKeeps up with video19 MB
⋮364×3647.8 ms-0.12 GBKeeps up with video
⋮518×51812.8 ms-0.23 GBKeeps up with video
Depth Anything V1 BaseQ4F16First generation, superseded by V2 Base.252×2526.2 ms-0.13 GBKeeps up with video71 MB
⋮364×36411 ms-0.24 GBKeeps up with video
⋮518×51820.1 ms-0.41 GBKeeps up with video
MiDaS v2.1 SmallFP32The fastest option on the list, and the coarsest - depth reads as a smooth gradient.256×2562.2 ms-0.13 GBKeeps up with video63 MB
MiDaS DPT-Swin2-TinyFP32Older architecture, fixed at one resolution.256×2565 ms-0.30 GBKeeps up with video157 MB
MiDaS DPT-HybridQ4F16Older architecture with more detail than the other MiDaS builds.384×38423.1 ms-0.47 GBKeeps up with video112 MB

* Measured on an AMD Radeon RX 7900 XTX. Speed is the fastest of 40 to 100 timed runs per setting, so a best case rather than a typical one; memory is a middle reading rather than a peak. Your hardware will shift the absolute numbers - what carries over is how the settings compare to each other. Video marks the settings that estimate depth within one 24 fps frame here; slower ones still play, they just refresh depth less often than the picture.