Skip to content
TheChatArchiveThe Chat Archive
o3-2025-04-16· OpenAIHugging Face LMArena Human Preference 140K

i have a 6 minute video - want to use ai to turn it into an scetch animation

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with chatgpt-4o-latest-20250326’s answer.
  1. User

    i have a 6 minute video - want to use ai to turn it into an scetch animation

  2. Assistant

    Turning a regular 6-minute clip into a hand-drawn “pencil sketch” animation is a fun project—AI can do most of the heavy lifting for you nowadays. Below is a menu of approaches—pick whichever fits your budget, skill level, and available hardware. Feel free to mix & match!

    ────────────────────────────────────────────

    1. One-Click / No-Code Solutions
      ──────────────────────────────────────────── | Tool | How it Works | Pros | Cons / Cost | |------|--------------|------|-------------| | ⚡ RunwayML “Stylizer” + “Line Drawing” | Upload video → pick “Line Drawing” or “Cartoon” style | Fast, cloud-based, no coding | 1250 credits ≈ $15 for 6 min @24 fps | | 🎬 Adobe Premiere + Sensei (Beta) | Import video → Effects ➜ Artistic ➜ “Sketch” (AI-powered) | Integrated into Adobe workflow | Requires CC subscription; not pure AI | | Clipchamp / Kapwing “Sketch” filters | Online editors with AI filters | Dead simple | Quality varies, usually 720p cap unless paid | | 🔥 VEED.IO “Pencil Sketch” | Select filter, export | Browser-only; zero setup | Paid export for >720p |

    Great if you just want a quick result without touching code.

    ──────────────────────────────────────────── 2. Semi-Pro / “Prosumer” Workflows
    ──────────────────────────────────────────── These give you a bit more control and usually look nicer.

    A. After Effects + “Cartoon Animator” (Red Giant)

    1. Import video, pre-compose.
    2. Apply Red Giant Cartoon plugin or “Find Edges” + Posterize Time.
    3. Tweak stroke width, shading; add a Turbulent Displace for hand-drawn jitter.
    4. Render at 24 fps with a light grain overlay to sell the hand-made vibe.

    B. EbSynth + Keyframe Stylization

    1. Pick 4-8 representative frames → manually convert them to sketches (Photoshop action or Procreate).
    2. Feed originals & stylized keyframes into EbSynth; it propagates the style to the rest of the footage.
    3. 6-minute video = ~8–10 keyframes → render → minor cleanup.
    4. Works well if camera motion is moderate.

    ──────────────────────────────────────────── 3. Fully Automated, Open-Source “AI Pipeline”
    ──────────────────────────────────────────── If you’re comfortable with Python + a GPU (8 GB VRAM or more), you can get high-end results and total creative control.

    Step-by-Step:

    1. Specs check
      • 6 min @ 24 fps ≈ 8,640 frames.
      • 512×512 frames need ~4 GB VRAM; 1080p needs 10-12 GB.

    2. Extract frames

    ffmpeg -i myVideo.mp4 -vf fps=24 frames/frame_%05d.png
    
    1. Convert each frame to “line art” with ControlNet (Stable Diffusion)
    from diffusers import StableDiffusionControlNetPipeline, ControlNetModel
    import torch, glob, cv2, os
    
    controlnet = ControlNetModel.from_pretrained(
         "lllyasviel/sd-controlnet-canny", torch_dtype=torch.float16)
    pipe = StableDiffusionControlNetPipeline.from_pretrained(
         "runwayml/stable-diffusion-v1-5", controlnet=controlnet,
         torch_dtype=torch.float16).to("cuda")
    
    prompt  = "clean black-and-white pencil sketch, thin lines, minimal shading"
    neg     = "colors, realistic, blur, grain"
    
    for f in sorted(glob.glob("frames/*.png")):
        img      = cv2.imread(f)
        canny    = cv2.Canny(img, 100, 200)
        canny    = cv2.cvtColor(canny, cv2.COLOR_GRAY2BGR)
        result   = pipe(prompt, negative_prompt=neg,
                        image=canny, control_image=canny,
                        num_inference_steps=20,
                        guidance_scale=7).images[0]
        result.save(f.replace("frames", "out"))
    

    Tips
    • Keep the same seed for temporal consistency (pipe.seed = X).
    • Lower guidance_scale if flicker appears.
    • For colorized “ink & wash,” swap in a different model or add “light watercolor wash”.

    1. (Optional) Deflicker & smooth
      • Run the DaVinci Resolve “Deflicker” filter OR
      • ffmpeg -i %05d.png -vf deshake,tmix if you’re CLI-savvy.

    2. Re-assemble video

    ffmpeg -r 24 -i out/frame_%05d.png -c:v libx264 -pix_fmt yuv420p sketch_out.mp4
    
    1. Add sound back (if needed)
    ffmpeg -i sketch_out.mp4 -i myVideo.mp4 -map 0:v -map 1:a -c:v copy -c:a copy final.mp4
    

    ──────────────────────────────────────────── 4. Time & Hardware Budget
    ──────────────────────────────────────────── • Consumer RTX 3060 ≈ 0.5 sec / 512×512 frame → 8,640 frames ≈ 72 min.
    • Cloud GPU (Paperspace, Lambda): ~0.60/hr;budget0.60 / hr; budget 1–2.
    • No GPU? Use Google Colab (free tier = slower; Pro tier ≈ $10/mo).

    ──────────────────────────────────────────── 5. Helpful Extras
    ──────────────────────────────────────────── • Consistency: Try ControlNet + TemporalNet (ComfyUI) or invokeAI’s “frame interpolation” node.
    • Speed lines / motion accent: Add an OpenCV motion blur or AE’s Cartoon Motion.
    • Color pop: After line art pass, colorize with another Stable Diffusion run (“flat cel shading”).
    • Legal: Make sure you own or have rights to transform the source footage.

    ──────────────────────────────────────────── Need More?
    ──────────────────────────────────────────── • Want me to write a ready-to-run Colab notebook?
    • Prefer a Toon-style (colored) instead of black-and-white?
    • Need advice on eliminating flicker?

    Tell me and I’ll dive deeper! 🎨✏️

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.