color-correction

Measure a frame with get_image action analyze_color to confirm the washed-out signature: lifted blackPoint, whitePoint below 235, and contrast under 45. Then add an AdjustContrast node after decode with factor around 1.4 and re-measure until whitePoint reaches roughly 248-255 with clipped highlights under 2%.

Call the get_image MCP tool with action analyze_color on a filename, path, or asset_id. It returns black and white points, contrast, saturation, RGB channel means, clipping percentages, heuristic flags, and an optional RGB/luma histogram overlay.

A contrast or levels stretch fixes compressed tonal range, which is the usual cause of washout. Reference color-match nodes like ColorMatchV2 only transfer the reference's tone, so they cannot add contrast a flat reference lacks; use them for shot-matching to an already well-graded frame.

No, the tool is image-only with no ffmpeg dependency. Extract a frame first using the ComfyUI venv's cv2 VideoCapture, typically the middle frame, then run analyze_color on the extracted PNG.

Long WAN-Animate clips render in temporal windows that can drift in saturation or brightness. Measure a frame from the first and a late window; if they differ, use the between-window colormatch in the embeds rather than a final global stretch.

Skip it when blackPoint and whitePoint already sit near 0 and 255 with healthy saturation, since the color is fine. Also avoid over-stretching: watch clippedHighPct, because blown highlights above roughly 2% are unrecoverable.