Video
The same composer as Image, tuned for clips — with longer waits and much bigger numbers next to the send button.
Video works exactly like Image — one composer, one feed, controls built from the model you picked — so this page only covers what's different. The page describes itself as "Type a scene. Get a 5-10s clip in seconds." and its prompt field suggests the level of detail that suits it: "A drone shot over a misty pine forest at golden hour…"
Read Image first if you haven't. Everything there about references, advanced parameters, plan-locked models and models that vanish applies here unchanged.
What's different
Many video models start from a picture
Animating a still is a whole category, so an empty required slot is common here. The button asks for the exact thing the model wants — "Add the video this model needs", or the image equivalent — and send stays held until it has it.
Length is a setting, not a sentence
Where a model exposes duration, it's a control above the composer rather than something you ask for in the prompt. The starting ideas on the page carry their length with them, so you can see what a five-second clip and a ten-second one cost before writing anything.
It takes longer
An image comes back in seconds; a clip can take a minute or more. The placeholder tile in the feed is the run — it becomes the video when the video exists.
Cost is the thing to watch here
Video is the expensive end of the catalog by a wide margin. A single clip from a top-end model can cost more than a whole week of credits on the free allowance, while a cheap image is a credit or two. The estimate next to the send button is there for exactly this — read it before your first run with any model you haven't used.
The estimate prices the outputs you asked for; the charge prices the ones that come back, and the run is charged when it completes rather than when you press send. A run that fails isn't charged. If the workspace or your own limit can't cover it, Endless refuses before starting.
Credits and spending
Where your balance lives, and the three levels that decide what you can spend.
Which models you can reach
Plan locks, workspace restrictions, and why a model you used last week might be gone.
Waiting for a clip
Once a run starts it belongs to the workspace, not to your browser tab. Endless records it before the work begins, so a run that's still going is still there after a reload — and runs someone starts elsewhere show up in the feed the same way.
A clip you start on a canvas is handed off entirely: the board gets a placeholder square immediately, and the finished video is dropped into that same square whenever it arrives — for everyone looking at the canvas, whether or not the person who started it is still there. See Canvas.
A failed run says "Couldn't generate. Try again." If video generation has been switched off for your workspace the refusal is explicit instead: "Video generation isn't enabled for this workspace" — and the tool won't be in the sidebar at all.