Runway Gen‑4 “Cast / Scout / Block” in Veo3Gen: An 8‑Shot Pre‑Production Workflow for Consistent Ads (as of 2026-02-26)

Why most Veo3Gen ads fall apart: no pre‑production (not “bad prompts”)

If your AI video ad starts strong and then “drifts” by shot 3—different face, wardrobe swapping, props vanishing, lighting changing—it’s usually not because you used the wrong adjective.

It’s because you’re trying to direct and edit without doing the lightweight pre‑production that real commercials rely on:

• a consistent cast (who we’re filming)
• a consistent location (where the world exists)
• a consistent block/shot plan (what happens, in what order, with what camera rules)

Runway’s Gen‑4 research emphasizes consistent characters, locations, and objects across scenes (https://runwayml.com/research/introducing-runway-gen-4). That same idea—consistency as a production system, not a one‑off prompt—can be applied in Veo3Gen even if you’re not using Runway.

This post gives you a beginner, tool‑agnostic workflow you can run inside Veo3Gen: Cast → Scout → Block, followed by an 8‑shot generation and QC loop.

The Gen‑4 framing: Cast → Scout → Block (translated to Veo3Gen)

There’s been ongoing attention around Runway Gen‑4/4.5 and motion quality improvements; for example, Runway’s Gen‑4.5 post describes “state‑of‑the‑art motion quality, prompt adherence and visual fidelity” (https://runwayml.com/research/introducing-runway-gen-4.5), and CNBC reported Runway announced Gen 4.5 and discussed independent benchmark positioning among major video models (https://www.cnbc.com/2025/12/01/runway-gen-4-5-video-model-google-open-ai.html).

But the most portable lesson is not the model. It’s the pre‑production structure.

Cast (in plain language)

Cast means: decide exactly who/what the viewer will recognize across every shot (person, product hero, mascot, key object). In Veo3Gen terms, this becomes:

• a cast card you paste into every prompt
• a small asset kit (your best reference images) you reuse

Scout (in plain language)

Scout means: decide what “world” your ad lives in—one primary location plus optional alternates. In Veo3Gen terms:

• a 5‑image mood/reference kit for the location(s)
• a short location lock paragraph you paste into every prompt

Runway highlights the idea of setting look & feel and preserving style, mood, and cinematographic elements (https://runwayml.com/research/introducing-runway-gen-4). Your goal is to create that as a spec, then reuse it.

Block (in plain language)

Block means: plan the action and camera like a director—who moves where, what the camera sees, and what must stay the same.

In Veo3Gen terms:

• an 8‑shot list with consistent constraints
• one variable change at a time (shot size, angle, action), not everything at once

Step 1 — Cast: build a 6‑line “cast card” you’ll paste into every prompt

You’re not “training” a character. You’re standardizing description so every generation starts from the same identity spec.

Copy‑paste template (fill in brackets):

CAST CARD (paste into every prompt)

1. Identity: [age range], [gender presentation], [ethnicity/skin tone if relevant], [hair color/style], [eye color], [distinct facial feature]
2. Wardrobe: [top], [bottom], [shoes], [one accessory], no wardrobe changes
3. Distinguishing marks: [freckles/scar/tattoo/glasses], [hairstyle locked], [no facial hair change]
4. Tone & performance: [energy level], [facial expression baseline], [speaking style if needed]
5. Camera distance rules: only [close‑up / medium / wide] as specified per shot; keep lens look consistent
6. Continuity constraints: same person, same outfit, same hair, same product model/color; do not alter identity

Tip: If your ad is product‑first (no on‑camera person), your “cast” can be the product hero: model name, colorway, materials, any visible logo placement, and “no design changes.”

Step 2 — Scout: lock your brand location(s) with a 5‑photo mood/reference kit

Pick one primary location for the entire 8‑shot sequence whenever possible. (Consistency beats variety in short ads.)

Build a simple “Scout Kit”:

• Photo 1: wide establishing look of the space (layout)
• Photo 2: lighting reference (window direction / practicals)
• Photo 3: surface/texture reference (countertops, flooring)
• Photo 4: color palette reference (brand tones)
• Photo 5: hero angle reference (where the product will be shown)

Then write a Location Lock paragraph:

LOCATION LOCK (paste into every prompt)

• Environment: [e.g., modern sunlit kitchen, matte white cabinets, light oak floor]
• Time/lighting: [e.g., morning, soft natural window light from camera left]
• Palette: [e.g., whites + warm wood + one brand accent color]
• Props: [only these props], no extra clutter

Runway notes that visual references can be combined with instructions to create consistent styles, subjects, and locations (https://runwayml.com/research/introducing-runway-gen-4). You’re recreating that same discipline: references + repeatable instructions.

Step 3 — Block: convert your idea into an 8‑shot list (0–12s) that stays consistent

An 8‑shot structure is a sweet spot for short ads: it’s enough coverage to feel edited, but small enough to keep continuity.

Your 8‑shot list template (copy/paste)

Use this per shot:

• Shot #:
• Subject (who/what is on screen):
• Action (one clear verb):
• Environment (use Location Lock):
• Camera (angle + distance + movement):
• Lighting (keep consistent):
• Constraints (must NOT change):

A practical 0–12s pacing map

• Shot 1 (0–1s): establish location + product present
• Shot 2 (1–2.5s): hands/interaction begins
• Shot 3 (2.5–4s): problem moment (before)
• Shot 4 (4–5.5s): product reveal / key feature
• Shot 5 (5.5–7s): benefit shown (after)
• Shot 6 (7–8.5s): proof/texture close‑up
• Shot 7 (8.5–10s): creator reaction / social proof beat
• Shot 8 (10–12s): CTA frame (clean, legible composition)

Keep the “world” stable. Vary only what an editor would vary: shot size, angle, and action.

Step 4 — Generate: run the shots in Veo3Gen with one variable at a time

Generation rule: carry over the same Cast Card + Location Lock + global style lines into every shot.

Then, per shot, change only:

1. the action, and
2. the camera instruction.

This mirrors the idea that you can regenerate elements from multiple perspectives within the same scene concept (https://runwayml.com/research/introducing-runway-gen-4)—but you’re doing it through structured prompting rather than relying on improvisation.

A simple prompt skeleton:

• CAST CARD
• LOCATION LOCK
• Global style: [UGC creator‑style / premium studio / etc.], [color grade], [realistic]
• Shot‑specific block: Subject + Action + Camera + Lighting + Constraints

Step 5 — QC: a fast consistency checklist (face, wardrobe, props, lighting, camera)

Run this after each shot batch.

Consistency QC checklist (short)

• Face/identity: same person; no sudden age/feature shifts
• Wardrobe/hair: unchanged; accessories consistent
• Product/props: same model, color, logo placement; no extra items
• Location: same layout cues; palette matches Scout Kit
• Lighting/camera: direction and “look” consistent with your plan

If one box fails, don’t redo everything—patch that one dimension.

Step 6 — Patch fixes: 3 re‑prompt patterns to correct drift without restarting

Here are three re‑prompt patterns that tend to be more surgical than “try again.”

Pattern A: “Re‑assert + prohibit”

Add two lines to the end of the prompt:

• Re‑assert: “Same [character/product] as previous shots; match identity and outfit exactly.”
• Prohibit: “Do not change hair, outfit, face shape, product color, or logo placement.”

Pattern B: “Continuity anchor”

Pick 2–3 anchors you never change:

• “Keep the gold hoop earrings, navy hoodie, and white mug in frame.”

Anchors reduce wandering because they give the model more fixed commitments.

Pattern C: “Lock the world, change the camera”

When the location drifts, explicitly freeze it:

• “Same kitchen layout: matte white cabinets, light oak floor, window camera-left. Only change is camera moves from medium to close-up.”

Runway frames “set your look and feel” as a way to preserve mood and cinematographic elements (https://runwayml.com/research/introducing-runway-gen-4). This is your prompt-level equivalent: lock look & feel, then vary shot grammar.

Example: a complete 8‑shot template (product + creator‑style UGC)

Below is a fillable template you can adapt.

Global (paste into every shot):

• CAST CARD: [your filled 6 lines]
• LOCATION LOCK: [your filled paragraph]
• Style: creator-style UGC, natural skin texture, handheld feel but stable, realistic color

Shot 1:

• Subject: creator holding product near counter
• Action: quick smile, places product down
• Camera: wide, slight handheld, eye level
• Lighting: soft window light camera-left
• Constraints: same outfit/hair; product label facing camera

Shot 2:

• Subject: hands + product
• Action: opens/activates product
• Camera: close-up top-down 30° angle
• Lighting: same
• Constraints: same countertop texture; no new props

Shot 3:

• Subject: creator + problem moment
• Action: reacts to [pain point] (confused/annoyed)
• Camera: medium, eye level
• Lighting: same
• Constraints: keep background layout identical

Shot 4:

• Subject: product hero
• Action: feature demonstration (single clear motion)
• Camera: close-up, slow push-in
• Lighting: same, highlight product surface
• Constraints: product color/material unchanged

Shot 5:

• Subject: creator using product successfully
• Action: satisfied nod
• Camera: medium, slight angle from right
• Lighting: same
• Constraints: keep brand accent color visible in background

Shot 6:

• Subject: detail/texture shot
• Action: slow rotation in hand
• Camera: macro-ish close-up
• Lighting: same, soft specular highlights
• Constraints: logo crisp, not distorted

Shot 7:

• Subject: creator testimonial beat
• Action: points to product, enthusiastic but not exaggerated
• Camera: medium close-up
• Lighting: same
• Constraints: no wardrobe change; earrings/glasses remain

Shot 8:

• Subject: product + clean negative space
• Action: product centered, subtle motion only
• Camera: wide/medium wide, static
• Lighting: same
• Constraints: leave space for on-screen text; no extra objects

FAQ

Does this require training or fine‑tuning?

No. The workflow is about repeatable specs (cast card, location lock, shot list). Runway’s Gen‑4 page explicitly notes consistency without fine‑tuning or additional training (https://runwayml.com/research/introducing-runway-gen-4).

Do I need Runway to use Cast/Scout/Block?

No. It’s a pre‑production mindset inspired by how Gen‑4 talks about consistent characters/locations and look & feel (https://runwayml.com/research/introducing-runway-gen-4). You can apply it inside Veo3Gen with prompts and reference assets.

What if my character changes slightly from shot to shot?

Treat it like continuity: re‑assert identity, prohibit changes, and add 2–3 continuity anchors (earrings, jacket, signature prop). Then regenerate only the drifting shot.

How many locations should an 8‑shot ad use?

Begin with one. Multiple locations increase drift risk. Add a second location only after you can keep one location stable.

Minimum viable pre‑pro (15 minutes for solo creators)

If you’re on a deadline, do just this:

1. Write the 6‑line Cast Card (3 minutes)
2. Pick one location and collect 5 reference images (7 minutes)
3. Draft an 8‑shot list using the template, one action per shot (5 minutes)

That’s enough structure to stop 80% of “shot 3 drift.”

Related reading

• Veo 3 API pricing comparison
• Veo 3 API features guide
• Veo 3.1 vs Sora 2 comparison

CTA: turn this workflow into a repeatable pipeline

If you want to automate multi‑shot generation, store your cast/location kits, and programmatically iterate on shot lists, see the Veo3Gen API docs: /api.

When you’re ready to scale beyond one‑offs (more variants, more campaigns), review plan options here: /pricing.

Try Veo3Gen (Affordable Veo 3.1 Access)

If you want to turn these tips into real clips today, try Veo3Gen:

• Start generating via the API: /api
• See plans and pricing: /pricing