Generate shots that keep the same character and location, within a credit budget.
Covers: How to generate consistent shots: the short answer · From shot list to keyframes · Animate with motion-only prompts
Turn a shot list into consistent AI shots: a still per shot from locked references, motion-only prompts, takes and selects, and a credit budget for a one-minute film.
Method. Image-to-video guidance from Google Cloud and Runway, and Google Flow clip lengths and credit prices, checked November 18, 2026; the credit table is our arithmetic and the prompts are our illustration. Vinc makes an AI film studio and describes its own workflow here.
How to generate consistent shots: the short answer
Generate every shot as a still first, from the same character and location references, and fix the framing and the face on the still. Then animate each approved still with a prompt that describes only the motion. Make two or three takes per shot at low resolution, pick the best one (your select), and render only the selects at final quality. Budget your credits before you start: a one-minute film needs several minutes of generated video.
This is step 8 of 10 in our "Start here" path. It follows making a shot list.
From shot list to keyframes
Keyframe
In AI video, a keyframe is a still image that fixes what a shot looks like before it's animated: its framing, character, set and light. The video model then starts from it. Google's best-practice guide for image-to-video says "Think of your source image as the first frame of your film: the stronger the start, the better the finish" (
Paste the fixed descriptions word for word. Don't reword them shot to shot.
Generate two or three stills, pick one, and check it against your storyboard: size, screen direction, props, wardrobe details.
Only then animate.
Animate with motion-only prompts
Once the still is right, the video prompt should describe what moves, not what's there. Google's guide says "Focus your prompt on the motion you want to see" and warns that "Redundant prompts confuse the model and lead to poor results" (Google Cloud). Runway's image-to-video guide says the same, quoted in our prompting guide: prompts "focus almost exclusively on motion".
A motion prompt answers three questions:
Question
Example
What does the subject do?
She turns the envelope over
What does the camera do?
The camera stays still
What else moves?
Steam drifts from the kettle
Keep each clip within your tool's length limit. On Google Flow's free tier, Omni Flash 1.2 makes "3 to 10s" clips; Veo 3.1 Lite and Fast make 4, 6 or 8 second clips, and Flow retires Veo 3.1 Fast on 5 December 2026 (Google Flow, checked 19 Nov 2026). Generate a second or two longer than the shot list says, so the editor has room to trim.
Keep characters and locations consistent
Problem
Fix
The face changes between shots
Same reference images every time; check the face on the still before animating
Wardrobe drifts (the earring moves, the colour shifts)
Put the details in the fixed description; reject stills that miss them
The room changes (window moves, furniture swaps)
Use a location reference for every shot; keep the screen-direction note in the prompt ("window on the right")
The light changes
Name the light the same way every time ("cold blue window light from the right, one warm lamp")
Written words come out garbled
Generate the shot without legible text and add the words in the edit, or photograph real handwriting
Takes and selects
Take
A take is one attempt at a shot. On a set, each take is one run of the camera; in AI work, each generation is a take.
Selects
Selects are the best takes, chosen for the edit. Mark them as you go ("1D take 2, select") so you don't regenerate shots you already have.
A simple routine per shot:
Make 2–3 video takes at draft resolution.
Watch each one twice: once for performance, once for errors (hands, faces, objects that melt).
Pick a select, or note what to change and make one more.
Move on. Perfect isn't the goal; a shot that holds for its planned seconds is.
Shooting ratio
The shooting ratio is how much footage you make compared with how much ends up in the film. If you generate five minutes to use one, the ratio is 5:1. Tight shot lists and approved stills bring it down.
Budget your credits
Here's our arithmetic for Google Flow's free tier, which gives "50 Google Flow credits per day" that don't roll over, with Omni Flash 1.2 priced at 1 credit per second at 360p and 3 credits per second at 720p (Google Flow credits). We assume 12 shots, each generated 1 second longer than planned (72 seconds per pass), three draft takes at 360p and two final takes at 720p. Image credits for the stills aren't counted.
Pass
Seconds generated
Credits per second
Credits
Free days at 50 a day
Drafts, 3 takes at 360p
216
1
216
5
Finals, 2 takes at 720p
144
3
432
9
Total
360
648
13
That's a shooting ratio of 6:1 against the 60-second film. The take counts are assumptions; one creator in our what an AI shot really costs piece reported running sequences 3 to 7 times. If your stills are right first, you'll need fewer takes.
Example prompts for three shots from the shot list. These are our illustration, written to the Google and Runway guidance above: a still prompt with the fixed descriptions, then a motion-only prompt. Attach Maya's look sheet and the kitchen sheet to every still.
Shot 1A: wide, static, 6 s.
STILL: Wide shot, eye level, wide lens. A small, tidy rented kitchen at dawn,
pale grey walls, cold blue window light from the right, one warm lamp over
the cooker. Door on the left, window on the right, small table in the middle,
a mustard raincoat on the chair nearest the door. Maya, a woman in her
mid-thirties, dark hair in a low knot, oatmeal cardigan over a faded blue
T-shirt, small silver hoop in her left ear, standing in the doorway on the left.
MOTION: She walks from the doorway across the room to the counter. The camera
stays completely still.
Shot 1E: insert, top-down, 4 s.
STILL: Top-down insert of a wooden kitchen table in cold blue dawn light. A
woman's left hand holds a cream envelope, front side up, a stamp in the
corner. Oatmeal cardigan sleeve at the edge of frame.
MOTION: The hand turns the envelope over to show the back. The camera stays
still.
The back of the envelope reads "From Maya, age 16". Generated handwriting can come out garbled, so we'd generate the back blank and add the words in the edit, or film a real envelope on a phone.
Shot 1K: medium, static, 5 s.
STILL: Medium shot, eye level, normal lens. The same kitchen at dawn, a little
brighter, kettle steaming on the counter. Maya, a woman in her mid-thirties,
dark hair in a low knot, oatmeal cardigan over a faded blue T-shirt, small
silver hoop in her left ear, reaching for the kettle switch. The mustard
raincoat on the chair beside her, on the left.
MOTION: She switches off the kettle, lifts the raincoat from the chair and
walks out of frame to the left. Steam keeps rising. The camera stays still.
Note what repeats: Maya's description is identical in 1A and 1K, the room's light is named the same way, and "on the left / on the right" holds the screen direction from the blocking plan.
Do it in Vinc
On Frames, then video, approve each keyframe, then animate it and press Generate video. Your locked cast and world sheets keep Maya and the kitchen the same across all twelve shots; Vinc says "faces, wardrobe, set and grade hold from shot to shot". Still look at every keyframe before you animate it.
How we compiled this
Guidance, clip lengths and credit prices come from Google Cloud, Google Flow and Runway help pages, read on 18 November 2026. The credit table, take counts and prompts are our own arithmetic, assumptions and illustration. Vinc makes an AI film studio.
Aspect Ratios in Film Explained: 1.33 to 2.39 and 9:16, With the History and Delivery Specs
Aspect ratios in film are width divided by height. See where 1.375, 1.85, 2.39, 16:9, IMAX and 9:16 came from, and the containers cinemas and Netflix expect.