Can you use a hand-drawn sketch as a reference input for AI video generation?

Yes. A hand-drawn sketch works as a reference input — upload the drawing alongside your text prompt and the model treats it as a visual anchor for composition, character arrangement, or camera framing. Inside the invideo agent, the sketch gets routed into an image model first to build a clean reference frame, then into a video model like Seedance 2.0 or Kling for motion.

Use sketches when text prompting alone won't get the model to visualize a specific physical arrangement — two characters in contact, an unusual prop configuration, a tricky camera angle. The invideo agent is an agentic video tool with all the current image and video models available, so you upload the sketch once and it routes the file to the right model for each stage.

How to feed the sketch in. Upload the drawing directly into the chat with the invideo agent and tell it what the sketch represents ("this is the configuration I want for the two characters") and what to ignore (rough proportions, shading, paper texture). The agent attaches it to an image model — Nano Banana or GPT-Image-2 work well for translating line art into a clean character sheet or reference frame — then uses that generated image as the reference for video generation in Seedance 2.0, Kling, or Veo depending on the shot. In one documented production, the team couldn't get Nano Banana to visualize a complex two-character carry from text alone; a hand-sketch of the configuration uploaded as reference unblocked it on the next pass.

What makes a sketch actually work. Clean, high-contrast lines on a plain background. Simple shapes the model can read as forms. Mark what matters (pose, relative position, prop placement) and leave out what doesn't — the model will invent texture, lighting, and finish regardless. If you have a specific style in mind, pair the sketch with 1–2 style reference images and tell the agent which file controls structure (the sketch) and which controls look (the style refs). As Hridaye, invideo's creative director, put it: "He hand sketched how we want juice box character attached to our vampire character. We took that drawing and we uploaded that to our agent one who then in turn took that and then attached that to Nano Banana and prompted his way to finally get us the perfect character sheet."

Sketch → image → video, not sketch → video direct. Most current video models read motion better from a finished reference frame than from raw line art, so the reliable path is: sketch → generate a locked reference image → use that image (plus a motion prompt describing camera and action) as the input to video generation. Research models like VidSketch and SketchVideo (CVPR) generate video directly from sketches with temporal consistency, and commercial sketch-to-video tools exist, but for production work the two-step path through an image model gives you a cleaner reference to iterate on and re-use across shots.

When to reach for a sketch. Multi-character physical contact shots, unusual prop configurations, specific staging the model keeps misreading, or any moment where you can draw the answer faster than you can describe it. Keep the sketch in the agent's context — once the resulting reference image is locked, the agent can pull it back for related shots so you're not re-uploading the drawing each time.

A few practical limits to know: line fidelity drops as scenes get visually dense, motion control still comes from your prompt (not the sketch), and finished video clips from current models run in short segments — generate the shot, then extend or stitch as needed.

Watch some of these to see what works for you:

He hand sketched how we want juice box character attached to our vampire character. We took that drawing and we uploaded that to our agent one who then in turn took that and then attached that to Nano Banana and prompted his way to finally get us the perfect character sheet.

— Hridaye, invideo's creative director