--- title: "How to add a watercolor with Claude (tested prompt and script)" description: "At one second the shot becomes a soft watercolor on grainy cream paper: the holographic backdrop melts into blotchy pale blue and lilac washes, the bottle glows violet with darker pigment along its edges, and a ragged unpainted white border frames the picture. Install one skill and ask Claude Code, or run the Python script we tested yourself." url: https://www.recharm.com/effects/watercolor --- # How to add a watercolor with Claude At one second the shot becomes a soft watercolor on grainy cream paper: the holographic backdrop melts into blotchy pale blue and lilac washes, the bottle glows violet with darker pigment along its edges, and a ragged unpainted white border frames the picture. Made by the script on this page, from our own test clip. ## Get this effect 1. ### Switch to Code in your [Claude Desktop App](https://claude.com/download) 2. ### Install the skill (once) ``` Install the Recharm video effects skill: download https://recharm-growth-staging.s3.us-east-2.amazonaws.com/website/downloads/recharm-video-effects.zip and unzip it into ~/.claude/skills ``` 3. ### Ask for the effect ``` Add a watercolor to input.mp4 Second the picture turns into a painting: 1.0 Paper color, showing through the washes and around the painting: (250, 247, 240) ``` ## Several videos, several effects, one ask Put your clips in the folder, mention them with @, and ask for everything at once. Claude runs one tested script after another on each clip, then joins them. Add a watercolor to **@clip1.mp4**, put a sale badge on **@clip2.mp4**, give **@clip3.mp4** a VHS look, then join all three with crossfades. ## Settings you can change Ask Claude to change any of these, or edit them at the top of the script. | Setting | Default | What it does | | ------------ | ----------------- | --------------------------------------------------------------------------------- | | `START` | `1.0` | Second the picture turns into a painting. | | `PAPER` | `(250, 247, 240)` | Paper color, showing through the washes and around the painting. | | `WASH` | `0.8` | Average strength of the paint. Lower is paler and more see-through. | | `SATURATION` | `2.2` | Color boost so the pale washes still read as color. | | `POOLING` | `0.25` | How much darker pigment pools along the edges of each wash. 0 turns the rims off. | | `MARGIN` | `0.07` | Unpainted paper border, as a share of the frame width. | | `BLANK` | `0.93` | Washes this light (0-1) or lighter are left as bare paper. | | `SEED` | `3` | Random seed for the paper grain, the blotches and the ragged border. | ## Without the skill Don’t want to install the skill? Paste this prompt into Claude Code and Claude writes the script itself. ``` Give input.mp4 (540x960) a watercolor painting look from 1.0 seconds to the end; before 1.0 seconds leave every frame untouched. Use Python with numpy and OpenCV (opencv-python-headless) for the frames and ffmpeg to read and write the video. Make the paper once, with a fixed random seed so it stays put: a grain map of 0.92 + 0.04 x fine noise (random values on a 3 pixel grid, scaled up bicubic) + 0.04 x coarse noise (20 pixel grid), all in 0-1; a wash-strength map of 0.8 x (0.8 + 0.4 x noise on an 80 pixel grid); and a paint mask that is 0 at the frame edges and 1 inside: take each pixel's distance to the nearest side divided by the frame width (540), add (noise on a 60 pixel grid - 0.5) x 0.07 so the border wobbles, then map it with clip((d - 0.042) / 0.056, 0, 1). For each frame from 1.0 s: shrink to 270x480 (area filter), run cv2.stylization with sigma_s 40 and sigma_r 0.35, Gaussian-blur it with sigma 1.2 and scale back to 540x960 (bicubic); multiply its HSV saturation by 2.2 (capped at 255); take the gradient length of its grayscale (Sobel x and y), blur that with sigma 3 and divide by 120 (capped at 1) as a rim map. Paper is (250, 247, 240). Paint = paper - (paper - color) x wash strength, then multiply by 1 - 0.25 x rim so pigment pools darker at wash edges; where the grayscale brightness is 0.93 of full or more, fade to bare paper (blend weight clip((brightness/255 - 0.85) / 0.08, 0, 1)). Blend paper and paint with the paint mask, multiply by the grain map and clip to 0-255. Keep the size, 30fps, the 6 second length and the audio (don't use ffmpeg's drawtext if you add any text), and save it as output.mp4. ``` ## Run it without Claude This is the exact script we tested. Save it as `watercolor.py` and run: ``` python3 watercolor.py input.mp4 output.mp4 ``` ``` #!/usr/bin/env python3 """Watercolor: at one second the shot is repainted as soft, see-through color washes on textured paper with a white margin. Usage: python3 watercolor.py input.mp4 output.mp4 Needs ffmpeg, numpy and OpenCV (pip install numpy opencv-python-headless). """ import json import subprocess import sys import cv2 import numpy as np START = 1.0 # second the picture turns into a painting PAPER = (250, 247, 240) # warm white paper showing through the washes and around the painting WASH = 0.8 # average strength of the paint; lower is paler and more see-through SATURATION = 2.2 # color boost so the pale washes still read as color POOLING = 0.25 # how much darker pigment pools along the edges of each wash MARGIN = 0.07 # unpainted paper border, as a share of the frame width BLANK = 0.93 # washes this light (0-1) or lighter are left as bare paper SEED = 3 # same seed, same paper grain, blotches and ragged border on every frame def noise(shape, cell, rng): """Smooth random field in 0-1: random values on a coarse grid, scaled up.""" small = rng.random((shape[0] // cell + 2, shape[1] // cell + 2)).astype(np.float32) return cv2.resize(small, (shape[1], shape[0]), interpolation=cv2.INTER_CUBIC).clip(0, 1) def make_paper(h, w): """Paper grain to multiply the paint by, uneven wash strength, and the ragged mask of where paint reaches.""" rng = np.random.default_rng(SEED) grain = 0.92 + 0.04 * noise((h, w), 3, rng) + 0.04 * noise((h, w), 20, rng) y, x = np.mgrid[0:h, 0:w].astype(np.float32) edge = np.minimum.reduce([x, w - 1 - x, y, h - 1 - y]) / w # distance to the nearest side edge += (noise((h, w), 60, rng) - 0.5) * MARGIN # make the border wobble reach = np.clip((edge - MARGIN * 0.6) / (MARGIN * 0.8), 0, 1) blotch = WASH * (0.8 + 0.4 * noise((h, w), 80, rng)) # some patches soak up more paint return grain[..., None], blotch[..., None], reach[..., None] def watercolor(frame): """Repaint one frame: flatten it into washes, lighten them toward paper, pool pigment at their edges.""" h, w = frame.shape[:2] half = cv2.resize(frame, (w // 2, h // 2), interpolation=cv2.INTER_AREA) flat = cv2.stylization(half, sigma_s=40, sigma_r=0.35) # smooth areas, keep their edges flat = cv2.resize(cv2.GaussianBlur(flat, (0, 0), 1.2), (w, h), interpolation=cv2.INTER_CUBIC) hsv = cv2.cvtColor(flat, cv2.COLOR_RGB2HSV).astype(np.float32) hsv[..., 1] = np.clip(hsv[..., 1] * SATURATION, 0, 255) flat = cv2.cvtColor(hsv.astype(np.uint8), cv2.COLOR_HSV2RGB).astype(np.float32) gray = cv2.cvtColor(flat.astype(np.uint8), cv2.COLOR_RGB2GRAY).astype(np.float32) rims = cv2.GaussianBlur(np.hypot(cv2.Sobel(gray, cv2.CV_32F, 1, 0), cv2.Sobel(gray, cv2.CV_32F, 0, 1)), (0, 0), 3) rims = np.clip(rims / 120, 0, 1)[..., None] # 1 along a wash's edge paper = np.array(PAPER, np.float32) paint = paper - (paper - flat) * blotch # thin, see-through wash paint *= 1 - POOLING * rims # darker rim where the pigment dried light = (gray / 255)[..., None] bare = np.clip((light - BLANK + 0.08) / 0.08, 0, 1) # highlights stay unpainted paint = paint * (1 - bare) + paper * bare out = paper * (1 - reach) + paint * reach return np.clip(out * grain, 0, 255).astype(np.uint8) def probe(path): info = json.loads(subprocess.check_output(["ffprobe", "-v", "error", "-select_streams", "v:0", "-show_entries", "stream=width,height,r_frame_rate", "-of", "json", path])) s = info["streams"][0] num, den = s["r_frame_rate"].split("/") return s["width"], s["height"], s["r_frame_rate"], float(num) / float(den) src, out = sys.argv[1:3] w, h, rate, fps = probe(src) grain, blotch, reach = make_paper(h, w) reader = subprocess.Popen(["ffmpeg", "-v", "error", "-i", src, "-f", "rawvideo", "-pix_fmt", "rgb24", "-"], stdout=subprocess.PIPE) writer = subprocess.Popen(["ffmpeg", "-y", "-v", "error", "-f", "rawvideo", "-pix_fmt", "rgb24", "-s", f"{w}x{h}", "-r", rate, "-i", "-", "-i", src, "-map", "0:v", "-map", "1:a?", "-c:v", "libx264", "-crf", "18", "-pix_fmt", "yuv420p", "-c:a", "copy", "-shortest", out], stdin=subprocess.PIPE) i = 0 while raw := reader.stdout.read(w * h * 3): if i / fps >= START: raw = watercolor(np.frombuffer(raw, np.uint8).reshape(h, w, 3)).tobytes() writer.stdin.write(raw) i += 1 writer.stdin.close() if reader.wait() or writer.wait(): sys.exit("ffmpeg failed") ``` ## More stylize and art - [Anime style **Anime style**](https://www.recharm.com/effects/anime-style) - [ASCII art **ASCII art**](https://www.recharm.com/effects/ascii-art) - [Blueprint **Blueprint**](https://www.recharm.com/effects/blueprint) - [Charcoal drawing **Charcoal drawing**](https://www.recharm.com/effects/charcoal-drawing) - [Claymation style **Claymation style**](https://www.recharm.com/effects/claymation-style) ## Find 3x _winning ads_ Recharm cuts your ad footage into tagged, searchable clips: hooks, testimonials, product shots. Find the shots that win, add the effect with Claude, and ship 3x more ad edits. [Start 14-day free trial](https://www.recharm.com/get-started)