Realtime Video · fast generation
AI video in seconds, not minutes
Realtime AI video is a generation loop measured in seconds: type a prompt, watch the clip, change a word, go again. On Percify realtime clips run 5, 10 or 15 seconds at 480P or 768P, and a verified 15 second run returned 15.1 seconds of video for 60 credits at 480P. The prompt stays after you generate, every take is kept in a strip you can scrub back through, and one generation can hold several shots, hard cuts included.
- 5, 10, 15 s
- clip lengths
- 480P, 768P
- resolutions
- 60
- credits, verified 15 s run at 480P
- 45 to 90 s
- the wait on other surfaces
Speed changes the tool, not just the wait
Every other generation surface is built around a 45 to 90 second wait: you fill in a form, commit, leave and come back. At five seconds that shape is wrong. The loop becomes type, watch, retype, and a form that clears itself after each run throws away the prompt exactly when you want to change one word of it.
So this screen behaves differently. The prompt stays after you generate, every result is kept in a strip you can scrub back through, and clicking an old take reloads the prompt that made it. The unit of work is the iteration, not the submission.
What you can set
Two dials, and both ranges come from the model provider rather than from us:
| Setting | Options |
|---|---|
| Duration | 5, 10 or 15 seconds |
| Resolution | 480P or 768P |
Several shots in one generation
The model can compose more than one shot inside a single generation, hard cuts included. It is invisible unless someone tells you, and it is useful: describe the shots in order and one generation can carry a small sequence. If a cut appears that you did not want, describe one continuous shot instead.
What it is for
Realtime video is the fast half of a project, not a replacement for the slow half:
- Cutaways and product shots around a talking head, instead of stock footage.
- Testing a visual idea in seconds before spending minutes on a full render.
- Short sequences for an ad, generated a few seconds at a time.
What a run costs
The verified benchmark is 15.1 seconds of video for 60 credits at 480P. A talking head render averages 69 credits at 480p for a median 19 second clip, so the two sit in the same range, but one comes back in seconds and the other in minutes. What you are buying is the ability to try. If you need a person speaking in sync, that is the talking head generator.
Work the loop in three steps
- 1
Type the shot
Describe one shot, or a few in order, and pick a length.
- 2
Watch and change one thing
The prompt is still there. Change a word and go again.
- 3
Keep the take you want
Scrub back through the strip; clicking a take reloads its prompt.
Short clips, fast
Real Percify video output, the kind of short clips people use as cutaways.
ASMR style
a close product style shot
Dance
movement in a short clip
Music video
a cinematic sequence
Frequently asked questions
How fast is realtime AI video?
Generation takes a handful of seconds, so the loop is type, watch, retype, instead of the 45 to 90 second waits of other surfaces.
How long can a realtime clip be?
5, 10 or 15 seconds, at 480P or 768P. Both ranges come from the model provider.
What does realtime AI video cost?
A verified 15 second run at 480P returned 15.1 seconds of video for 60 credits.
Can realtime video make a talking avatar?
No. For a person speaking with lips in sync, use the talking head generator, which drives the mouth from the audio.
Why are there cuts in my realtime clip?
The model can compose several shots inside one generation. Describe a single continuous shot if you want none.
Keep reading
Try an idea in seconds
Type it, watch it, change one word, keep the take you like.