How to Make an AI Video, Step by Step
Making an AI video is not one button. It is a sequence of decisions, each with a cost, and the difference between a cheap project and an expensive one is the order you make those decisions. This is the workflow that keeps the cost per kept clip in the cents.
Step 1 — Write the prompt
The prompt is the cheapest part of the whole pipeline, and the one people under-invest in. A good video prompt states the subject, the action, the setting, the camera move, and the mood in a single pass. Vague prompts produce generic motion that you will re-render five times. Specific prompts produce a keeper on the first or second attempt.
Write the prompt so that a single sentence could be a storyboard beat: "a white-robed swordsman walks up a bamboo ridge in swirling mist, slow dolly-in, backlight." That is a clip, not a hope.
Step 2 — Generate a free key image
Before you spend video credits, generate a still image from the same prompt. Standard images are free — 10 a day for everyone, unlimited during membership on Pro and Studio. The key image does three jobs:
- It tells you whether the composition and lighting work before you commit to motion.
- It becomes a reference image for the video, so the clip stays faithful to the still.
- It is a deliverable on its own, so even if the video never works, you have a usable image.
This step costs nothing. Use it.
Step 3 — Extend the image into video
Now run the reference pipeline: feed the key image (and up to 8 more references if the subject needs several angles) and generate the clip. Pick the quality for the test:
- 480P, 5 seconds = 5 credits ($0.064) — the cheapest way to check the motion.
- 720P, 15 seconds = 28 credits ($0.36) — the final-quality pass.
Generate the short, low-quality version first. If the motion is wrong, you have spent $0.064, not $0.36.
Step 4 — Iterate with references
The first render is rarely the keeper. Re-render with adjusted references and prompt tweaks. Because the reference images lock the subject's identity, your re-renders stay consistent — you are fixing the motion, not re-rolling a face that changes every take. That is what keeps the cost per kept clip bounded.
Step 5 — Add a soundtrack (optional)
A short with a matching 90-second track is a finished piece, not a clip. Music is generated from a prompt — mood, genre, length — and a 90-second track costs 10 credits ($0.13), a 180-second track 20 credits ($0.26). Because images, music, and video share one account, the soundtrack costs credits from the same wallet as the video. No second tool, no second balance.
The cost of a realistic session
A realistic short-form session is: one free key image, two or three 480P 5-second tests (6–9 credits), one 720P 15-second final (14 credits), and maybe a 90-second track (5 credits). That is roughly 25–28 credits, or $1.25–$1.40, for a finished 15-second piece with a soundtrack. On a $10 top-up pack (200 credits, never expires), that covers several finished pieces.
The habits that cut the bill
- Always generate the free key image first. It is free and it catches composition errors before they cost video credits.
- Test at 480P / 5s, finalize at 720P / 15s. Never burn the expensive pass on an untested idea.
- Use references for the final pass. They lock identity and kill the re-render waste.
- Top up in small packs. Credits never expire, so a $10 pack is enough for a month of short-form work without a commitment.
See how we price AI video generation for the exact cost model behind every number here, and AI image generation credits explained for how the free image quota works.
Want to try it? Create a free account