Paste a script into a chat with Supercomputer and ask for a whiteboard explainer. It plans the scenes, picks image, video and voice models from the catalog, renders each step and shows you the result, and you steer it in the same conversation. The hand-drawn look is an approximation, so every fact on screen stays yours to check.
A whiteboard explainer works because a hand seems to draw each idea while a voice explains it, and your eye follows along. Making one used to mean storyboarding, illustrating, recording and syncing by hand. With Supercomputer, the agent in the chat, you paste a script, say what you want, and it plans the scenes, picks models from the catalog, renders each step and shows you the result. The look is an approximation, and every fact on screen is yours to check.
Most explainers start as text: a product description, a lesson outline, a process your team keeps repeating in meetings. Text is accurate but slow to absorb. A whiteboard video changes the pace. One idea appears at a time, the picture is deliberately simple, and the narration carries the logic from one scene to the next.
The expensive part was never the idea. It was the production chain around it: sketching every scene, finding or drawing consistent icons, recording a voice, lining everything up. A catalog of models can do each of those jobs in a rough first pass. What was missing was someone to coordinate them, and that is the job of an agent. You describe the finished video once, in the words you would use with a colleague, and the agent breaks it into scenes, assigns a model to each step and keeps the thread between them.
Supercomputer is a chat in the web app. It is a separate product from the toolkit, which is the manual workspace where you choose the model yourself. In the chat you do not choose anything up front: you describe the result you want in plain words, and the agent plans the job, picks models from the catalog of 502 models, renders the steps and shows you the result. Then you steer it in the same conversation.
For a whiteboard explainer the plan usually has four parts. The script is split into scenes. Each scene gets a simple drawn style picture from a text-to-image model. Each picture is animated into a short clip by a video model. A text-to-speech model reads the narration. Which models do the work is the choice of the agent, and it tells you what it picked, so you can ask for a different one when a result misses.
Supercomputer uses the same models and the same credit system as the toolkit. A generation costs credits, the amount depends on the model and the settings, and the pricing page is where plans live. It needs no key, because it is a chat in the web app. New accounts get free credits, which is enough to test a short script before you commit to a long one.
Honest framing: the drawn look is imitated, not recorded. Video models generate frames that resemble a marker sketch appearing on a board. They do not trace a real pen stroke by stroke in time with your narration, so treat the result as a hand-drawn style, never as a live drawing.
Whiteboard video is a family of looks, not one. Naming the look in your first message saves the most revisions, because the models respond to concrete visual words far better than to the word whiteboard alone. These are the approaches that hold up well, and what to watch in each.
| Approach | What the picture looks like | Best for | Watch out for |
|---|---|---|---|
| Marker sketch on white board | Black and one accent color marker lines, simple stick figures, plain white background | Process explanations, sales pitches | Faces and hands can drift between scenes |
| Black ink line art on paper | Fine pen lines, a little shading, off-white paper texture | Lessons, history, storytelling | Detail looks muddy once the clip is small |
| Doodle icons with spot color | Flat hand-drawn icons, one or two bright fills | Onboarding, feature walkthroughs | Icons repeat unless you fix a style description |
| Diagram and labeled flow | Boxes, arrows and short labels on a clean board | Systems, workflows, comparisons | Generated lettering is unreliable, so keep labels short |
| Cut-out reveal | Finished picture revealed by a wiping hand or marker | Short social explainers | The hand itself is the part models get wrong most often |
Two habits help whichever look you pick. First, ask for a plain background in every scene, so the pictures read as drawings and not as scenes. Second, keep one style sentence and reuse it word for word in each scene prompt, or ask the agent to do that for you. The effects page holds one-click preset looks, a quick way to find words for a style you cannot name.
This is the journey from a finished script to a cut you can review. It assumes you have the script and nothing else.
On Picasso IA the work spreads across catalog models you will recognize from the toolkit: image models such as nano-banana, seedream-4.5 and ideogram-v3-quality for the drawn scenes, video models such as veo-3.1, kling-v3-video and seedance-2.0 for the motion, and a voice model such as speech-2.8-hd for the narration. In the chat you can ask for one by name, or leave the choice to the agent.
Most weak whiteboard videos fail in the briefing, not in the models. A script written to be read is not the same as a script written to be drawn. Each sentence needs something concrete to put on the board, and the sentences that name nothing visible become filler scenes. Before you paste, mark every line that has no picture and rewrite it around an object.
These habits do most of the work:
The same discipline applies to the voice. A flat read makes a good drawing feel slow. If the narration is the weak point, the page on AI voiceover for videos goes through voice choices in more detail, and automatic video subtitles covers captions, which many viewers will rely on because they watch with the sound off.
A page that sells a format should also say where it stops. For this one there are specific limits, and a few of them decide whether the format fits your project at all.
The motion is generated, not drawn. A real whiteboard video is a recording of a hand drawing, so the line grows exactly where the narrator points. A model produces frames that suggest a sketch appearing, and the timing between the drawing and your sentence is approximate. If a scene depends on a line appearing at one precise word, expect to trim or re-time it in an editor.
Factual content needs a human. The agent can lay out your script as pictures, but a drawn figure can show a wrong number or an object that does not exist, and any word drawn on the board may be misspelled, with the same confidence as a right one. For anything about health, money, law or safety, treat the video as a draft that a qualified person reviews before it is shared.
Generation also takes time and costs credits. A long explainer is many images, many clips and a voice track, so the cost grows with every scene and every redo, and the models differ in clip length, quality and cost. Check the model pages before planning a long piece. Finally, people: use only your own likeness, your own voice or fictional characters. Do not draw or voice a real, identifiable person without their permission.
It can make a video in a whiteboard style, which is a close cousin of the real thing. Supercomputer splits your script into scenes, generates drawn style pictures, animates them into short clips and adds a narration. What it does not do is film a real hand drawing in sync with each word. For most training, pitch and lesson uses that is enough once you have checked it.
It is an imitation. Image models draw pictures that resemble marker or ink sketches, and video models animate them so the lines seem to appear. Nothing is tracing a pen on a board in real time. If viewers only need the clarity of a sketched explanation, the look does the job.
No. Supercomputer is a chat in the web app, so there is nothing to set up and no key to create. You describe the video you want in the conversation and the agent does the planning. A key from your account matters only for the API or the MCP server, and a key should be treated like a password.
The agent picks models from the catalog for each step: a text-to-image model for the drawn scenes, a video model for the motion and a text-to-speech model for the narration. Candidates include nano-banana, seedream-4.5, veo-3.1, kling-v3-video, seedance-2.0 and speech-2.8-hd. You can ask for a different one if a result misses.
It costs credits, and the amount depends on the model and the settings used for every image, clip and voice track, so a longer script with more scenes costs more. New accounts get free credits to test a short script, and the pricing page is the place to trust for current figures.
Yes. The toolkit runs the same catalog today, so you can try a scene prompt on an image model, an animation prompt on a video model and a voice on a text-to-speech model, and see what each produces. It is a good way to learn which style words work, and what you find can go into the chat as a reference.
Mostly, if you help it. Models do not remember a style on their own, so the reliable method is to fix one style sentence covering line weight, colors and background and reuse it in every scene, or ask the agent to carry it. Faces and hands can drift, so keep figures simple and expect to regenerate the odd scene.
Each clip is as long as the video model allows, and the models differ, so a longer explainer is a chain of short clips joined across scenes. Plan chapters and approve them one at a time in the same chat, and check the model pages for limits.
Only your own face and voice, or fictional characters. Drawing or voicing a real, identifiable person without their permission is not something this page recommends, and it can cause real harm. Simple stick figures carry the story without any likeness at all.
You can usually share what you make, but each platform has its own rules about generated media, disclosure and monetization, and they change, so check the current requirements of the platform you use before publishing. For a course or a company training, review the content first.
Ready to hear your script as a drawn explainer? Open the toolkit to test the scenes and voices on the same catalog today, or ask a question through support.
Choose your language
