How to make an AI short film: node canvas × one-line-to-film
Making short films with AI from scratch: wiring the node canvas, turning an image into video, continuing from an existing clip, laying a whole script out as shots, plus prompt writing and common questions.
What Mussel does
In one line: chain your ideas into a film. Write a single sentence and get a short clip, paste a whole multi-episode script and let it split into shots, or wire it yourself on an infinite canvas — an image feeds a video, and that video continues into the next shot.
No editing software, no timeline, nothing to install — it all runs in the browser. Finished films can be subtitled, reframed to portrait or landscape, and have a cover pulled from them; all of that is done locally and costs nothing extra.
Node canvas: chain your results
Most AI tools give you a single box: type, wait, download, start over — the last result has nothing to do with the next one. The canvas fixes exactly that: every result is material for the next step。
Add an image node and describe the shot
Generate the character or the location first. This image is the basis for every later shot, so it's worth a few tries.
Drag a line from it to a video node
Once connected, that image becomes the first frame of the shot, and the video starts moving from it.
Several images are handled in two steps
A video model only takes one image. When several are connected, the system first composes them into a single first frame with an image model, then animates that — it happens automatically; you only see “Step 1: composing the first frame” and “Step 2: bringing it to life”.
Stitch the shots into one film
“Stitch film” in the bottom toolbar joins every finished clip on the canvas, left to right.
Continue from an existing clip
This is where most people get stuck first: you cannot wire one video node straight into another. A video model takes a single image as its first frame — feeding it a clip does nothing.
The right move is to pull a frame out of the clip and use that as the new first frame. Select the video node and use “Continue” in its toolbar — three ways to grab a frame:
- First — continue from the start of the shot
- Current — scrub to the second you want, then click; that exact frame is taken. The best-looking frame is usually neither the first nor the last
- Last — continue from the end; the most common choice
One click does three things: grabs the frame, drops a node holding it, and wires up an empty video node after it. All you write is what happens next. Setting, lighting and framing carry over, because it is literally the same frame. Grabbing frames is done locally and is free.
Paste a script, get a storyboard
If you already have a script, you don't need to type shot by shot. Use “Script” in the canvas toolbar, or “Import script” on the series list, and paste it in — it splits into shots automatically.
Multiple episodes work too: put headings like “Episode 1” and “Episode 2” in the script, or separate episodes with blank lines, and each becomes its own film.
The difference is what happens after the split — importing into the canvas lays out a row of image → video nodes you can edit and generate one by one; importing into the series list creates one storyboarded draft per episode for the workspace.Both only split — nothing is generated and nothing is charged, so you can read and edit before deciding to shoot.
Writing prompts
There are no magic words. A formula that's good enough: who + where + doing what (one main action)。
Better
Xiaoyu stands outside a late-night convenience store, looking down at her phone, neon light on her face
One person, one place, one action — enough to build a frame.
Likely to go wrong
Xiaoyu walks into the store, buys something, argues with the owner and runs out
Four actions in one shot; the model can only play one — split it into four shots.
A few more that actually help:
- Write character names directly — naming them is how the system knows who is in the shot, and it brings their character sheet along.
- Put dialogue in quotes, e.g. Xiaoyu says: “Why are you here?”. The system uses it to estimate shot length and to build subtitles.
- Camera language works — close-up, medium shot, high angle, handheld shake: write it and you get it.
- Don't compute the duration yourself — the system sets it from the dialogue length and warns you if the line would get cut off.
Characters keep changing? Read this
This is the most common complaint about AI video: the person changes in the second shot. Both the cause and the fix are concrete.
Only one reference face can be locked at a time.When several images are connected, only one of them is actually sent to the model as reference; the other characters are redrawn from the text — which is why the second person is often “not the same person”. On the canvas, the face that gets locked is the one from the topmost node, so drag whoever matters to the top.
To keep two characters consistent, the reliable approach is to shoot them separately: his close-up as its own shot (him on top), her close-up as its own shot (her on top), and the two-shot wide where faces don't carry the frame anyway. One locked face per shot, and both hold up once cut together.
Setting continuity comes from “Continue” above — carry on from the last frame and the lighting and set won't jump.
FAQ
Do I need editing skills?
No. There is no timeline and nothing to install. You describe the shot; framing, camera work and rendering are handled for you, and the clips are stitched into one film at the end.
How long does a film take?
A shot usually takes a minute or two. Rendering runs in the background — close the tab and the progress is still there when you come back.
How is it billed?
Credits, no subscription. Video by the second, images per piece, text by token, charged only when something is actually generated. Splitting scripts, burning subtitles, reframing, pulling covers and grabbing frames all run locally and are free.
Can I post the film directly?
Yes. From the film window you can burn subtitles, convert to 9:16 / 16:9 / 1:1 and pull a cover in one click — no need to reshoot for a different platform.