Depth Slice Mask Video

Depth maps are great for separating things, but they're even better when you let a "depth window" travel through them. Eden_DepthSlice_MaskVideo is the animated cousin of DepthSlicer: instead of cutting the depth map into static layers with k-means clustering, it sweeps a sliding slice from near to far (or far to near) across n_frames, producing a mask video where each frame reveals a different depth band. One depth map in, a whole animation out.

It's the mechanism behind the "focus travels through the scene" look - objects pop into mask existence one depth layer at a time - and it composes beautifully with frame-by-frame inpainting to make content reveal or regenerate in depth order.

How it works

The node takes a depth map (bright = close, as the KB's depth essay frames it) and normalizes it to 0–1, then optionally applies gamma_correction - raising the value brightens mid-tones and shifts where the slice bands land, which is your main way to bias the sweep toward near or far detail. It then marches a window of width slice_width from min_depth to max_depth across the depth range, one step per frame, and each frame becomes a binary mask showing which pixels fall inside the current window.

Two animation controls round it out:

• reverse - flips the direction of the sweep.
• bounce - runs the sweep forward, then back, so the animation loops instead of snapping. The code halves the frame budget for the forward pass and concatenates the reverse, giving you a smooth ping-pong.

Inputs and outputs

• depth_map - IMAGE, the depth map to sweep.
• slice_width (default 0.1) - how wide the depth band is. Narrower = more slices visible across the animation.
• min_depth / max_depth (0 / 1) - the depth range to sweep. Tighten these to restrict the sweep to part of the scene.
• gamma_correction (default 1) - depth curve shaping, as above.
• n_frames (default 100) - length of the mask animation.
• reverse, bounce - direction and looping.

Outputs: depth_slice_masks (the MASK video - one frame per depth step) and gamma_corrected_depth_map (the depth map after gamma, as a MASK), which is handy if you want to see exactly what the gamma did.

How to use it

The pattern is the same "mask → inpaint" recipe as DepthSlicer but animated: convert each mask frame to a latent mask, run a masked inpaint pass per frame, and you get a video where content regenerates in strict depth order - the background changes first, then the mid-ground, then the foreground, like a camera focus pulling through the scene. The KB's inpainting essay backs the approach: masked inpainting guarantees the unmasked region stays pixel-identical, which is exactly what keeps this frame-stable. Feed it a good depth map (MiDaS-class preprocessors) and start with a wide slice_width (0.2–0.3) to avoid a too-thin, noisy reveal.

Installing

One of 70+ nodes in edenartlab/eden_comfy_pipelines (Eden.art nodesuite):

cd ComfyUI/custom_nodes
git clone https://github.com/edenartlab/eden_comfy_pipelines.git
cd eden_comfy_pipelines
pip install -r requirements.txt

Or ComfyUI Manager → search "Eden". Pure torch math - no extra model downloads beyond your depth map source.

Gotchas

The output is MASK, not IMAGE - so unlike DepthSlicer you're already in mask territory, which is convenient but means you can't just preview it as an image without conversion. Watch n_frames × resolution: a 100-frame mask batch at high res eats VRAM like any stacked animation. And remember the depth map gets normalized internally, so extremely flat or extremely contrasty maps will behave - but you'll get better sweeps from a clean, well-lit depth source. If the sweep looks wrong, gamma_correction is your first lever, not slice_width.