IAMCCS Cine-PP AntiAlias 1:1

Upscaled video frames come back with a tell: hard jagged edges on diagonals, shimmer on thin lines, everything looking slightly crunchy. A plain blur smooths the jaggies but also turns the whole image to mush. IAMCCS_CinePPAntiAlias is a post-production anti-aliasing pass that knows where the edges are - it detects them, smooths them, and leaves the flat detail alone. Drop it between your upscaler and your combine step and the final clip stops shimmering.

How it works

It's a 1:1 per-frame post node. The core uses a Sobel edge detector (cached per dtype/device) to find edges, then applies edge-guided blurring with knobs for how much to protect detail and clean up chroma:

• preset - Auto picks Balanced normally, switching to Low VRAM for 4K or 97+ frame batches, and to Strong for small, high-quality batches. The others (Preview, Light, Balanced, Strong, Crisp Lines, Low VRAM) are manual starting points.
• aa_strength (0–2, default 1) - overall strength multiplier on the preset.
• edge_threshold (-1 = auto, else 0–1) - how aggressively a pixel counts as an edge.
• blur_radius (0–4) - manual blur radius override.
• temporal_stability (-1 = auto, else up to 0.75) - pulls neighboring frames into the smoothing so edges don't flicker between frames. This is the one that makes video look steadier instead of just softer.
• detail_protection and chroma_cleanup (-1 = auto, else 0–1) - protect fine texture from being flattened; clean up color fringing on edges.
• hardware_mode (auto/low_vram/balanced/quality/cpu_safe) and max_frames_per_chunk - how many frames process at once. Auto estimates chunk size from your free VRAM; cpu_safe forces a single frame at a time.

Outputs are images and a report string describing the effective preset and chunking it chose.

Where it fits

This is squarely in the Cine-PP post-production bucket of the pack - the "finish the shot" stage, not the generation stage. The workflow shape is: upscale (or generate) your frames, run them through this node, combine to video. The temporal_stability control is the reason you'd use this instead of a per-image sharpen/blur in Photoshop-style nodes: it's video-aware. One honest note: it smooths aliasing but doesn't invent resolution - if your upscaler's output is genuinely soft, run a real detail pass after AA, not instead of it.

Install

Part of IAMCCS-nodes:

• ComfyUI Manager → search "IAMCCS" → install → restart.
• Or:

cd ComfyUI/custom_nodes
git clone https://github.com/IAMCCS/IAMCCS-nodes.git

Restart ComfyUI. Requirements: ComfyUI ≥ 0.3.0, Python ≥ 3.12, Torch ≥ 2.8. Pure torch ops - no models, no downloads.