Blackmagic doesn't call Fusion's Depth Map an AI tool

The shot

Blackmagic Design's product page for Fusion, as retrieved on 16 September 2026, describes its tracking and warping tools in traditional terms: 'Fusion uses advanced optical flow algorithms to generate motion vectors so you can fix shots, retime, steady and morph images.' The company's own DaVinci Resolve 19.1 Reference Manual, dated February 2025, documents a Depth Map node inside the 'Fusion Page Effects' section's Matte Nodes chapter that 'creates an Alpha channel based on the perceived distance of objects in your clip,' warning it 'is a very computationally intensive effect.' The same manual places Magic Mask, the tool Blackmagic elsewhere brands with its 'DaVinci Neural Engine' language, in a separate chapter under the Color page, not Fusion.

What the documents show

Read together, Blackmagic's own materials draw a line the company does not narrate directly but its documentation structure makes visible. Tools it explicitly attributes to the 'DaVinci Neural Engine' and describes as using 'advanced machine learning algorithms,' such as Magic Mask, the Dialogue Separator, and the Music Remixer, sit under the Color, Edit, and Fairlight pages. Fusion's own nodes, including Depth Map, a single-image distance-estimation tool another vendor might label AI-based, are described using process language, such as 'computationally intensive' and named quality modes, rather than machine-learning language. Blackmagic's current what's-new page confirms this split, attributing AI capabilities to the Photo, Color, and general pages without naming Fusion. This is a documentation choice by one vendor, not independent evidence about what any node computes internally.

The workflow

A compositor working in Fusion should not assume a Fusion-page node carries the same neural-network branding, licensing, or explainability caveats Blackmagic attaches to its Color-page tools, and vice versa. Depth Map still requires a supervisor's judgment about the Mode setting, the Near and Far limits, and whether the resulting matte needs the Map Finesse post-processing controls before it isolates a grade or effect region. Which node handles which task in a pipeline is a decision the artist makes shot by shot; the manual documents the controls, not the decision itself.

What the tool does not change

Depth Map produces a distance estimate from a single 2D image, not a measured depth channel from set data; the manual's own preview and adjustment controls exist because the estimate needs manual correction. Whether or not a tool is marketed as AI, a Fusion artist still checks a generated matte against the plate before it goes downstream. This is editorial: a feature's marketing label is separate from what a compositor still verifies by eye.

• Does a specific Fusion node's documentation use machine-learning language, or only process and performance language?
• Which DaVinci Resolve page, Color, Edit, or Fusion, does a given AI-branded feature actually live on?
• Has a Depth Map or similar estimate been checked against the plate before it drives a grade or effect?

Blackmagic's own manual is more precise than its marketing shorthand suggests: it reserves 'Neural Engine' language for specific Color, Edit, and Fairlight tools, and describes Fusion's comparable capabilities in the older vocabulary of algorithms and processing modes.