AI Video Creation9 min read

Veo 3.1 "Ingredient" Prompts in Flow: A Practical Creator Guide to Adding/Removing Elements Without Breaking the Shot (2026)

Learn Veo 3.1 ingredient prompts: a Lock → Delta → Reconfirm workflow, copy‑paste templates, QA checklist, and a worked example to prevent shot drift.

On this page

TL;DR

“Ingredient prompts” are for surgical edits: add/remove/replace one element while keeping the shot stable. The reliable method is Lock → Delta → Reconfirm:

  • LOCK the shot like a contract (framing/motion, lighting, style, subject, location, action).
  • Make one DELTA (add/remove/replace) with explicit placement + interaction + timing.
  • RECONFIRM everything that must not change (“same camera, same person, no cuts, no new props besides X”).

Flow describes Ingredients to Video as using multiple reference images to control characters, objects, and style—that’s the continuity lever (https://blog.google/innovation-and-ai/products/veo-updates-flow/).

Key takeaways

What “ingredients” are in Flow (and when they beat rewriting the whole prompt)

Google describes Flow as an AI filmmaking tool powered by Veo (https://blog.google/innovation-and-ai/products/veo-updates-flow/). In that same update, Ingredients to Video is described as a way to use multiple reference images to control characters, objects, and style (https://blog.google/innovation-and-ai/products/veo-updates-flow/).

For creators, that implies a practical rule:

  • Rewrite the prompt when you want a new concept or new shot.
  • Use ingredient-style prompting when you want the same shot with one controlled change.

When ingredients are the right tool

Use ingredient edits for:

  • Continuity (same creator identity, same product look, same set).
  • One-variable iteration (swap prop, remove background item, change a shirt color).
  • Compliance cleanup (remove logos/readable text without changing wardrobe or location).

The failure mode that wastes the most time: “one tweak” that re-rolls the whole shot

This is the drift pattern:

“Same video, but make it more premium and add the product in her hand.”

That’s not one change—it’s multiple changes. The Veo prompt guide lists prompt components like shot framing and motion, style, lighting, character descriptions, location, action, and dialogue (https://deepmind.google/models/veo/prompt-guide/). When you don’t lock those components, the model is free to reinterpret them.

Editor rule: If your delta is “add a mug,” do not also ask for “cooler,” “premium,” “cinematic,” or “trendier.” Those are new art direction.

The micro-workflow: Lock → Delta → Reconfirm

This is the workflow to reuse across shots and campaigns.

1) LOCK (write the shot like a contract)

Include only what you need to keep stable, but make it unambiguous:

  • Format: aspect ratio (16:9 or 9:16)
  • Framing + motion: e.g., medium close-up, static camera, no cuts
  • Lighting + style: e.g., soft window light, realistic textures
  • Subject + location + action: who/what, where, what they’re doing

This mirrors the controllable components called out in the Veo prompt guide (https://deepmind.google/models/veo/prompt-guide/).

2) DELTA (one change, bounded)

Pick exactly one:

  • ADD one object
  • REMOVE one object/type of artifact
  • REPLACE A → B

Make it hard for the model to “get creative” by specifying:

  • Placement: right hand, back-left shelf, on the table near the laptop
  • Interaction: held still, tapped, placed down
  • Timing: 00:00–00:02 (or “during the first second”)

3) RECONFIRM (explicitly forbid drift)

End with a clause that says what must remain identical:

  • camera/framing/movement, no cuts
  • same subject identity and wardrobe (unless wardrobe is the delta)
  • same background/set dressing
  • no new objects besides the delta item

If you’re producing lots of variants (prop color swaps, compliance removals, hook variations), Veo3Gen’s developer API can help you standardize this so only the DELTA line changes between runs.

CTA (mid-article): If you want to run this workflow at scale, Veo3Gen provides access to Google’s Veo 3.1 models with text-to-video and image-to-video, including first-and-last-frame control on Veo 3.1, plus native synchronized audio (dialogue, SFX, music) in one pass.

Copy‑paste “ingredient prompt” template (with a worked table)

Use this structure every time.

The universal template

LOCK: [aspect ratio], [framing], [camera motion]. [lighting]. [style].
Subject: [who/what]. Location: [where]. Action: [what happens].

DELTA ([ADD|REMOVE|REPLACE]): [one specific change with placement + interaction + timing].

RECONFIRM: Keep [camera/framing/motion], [subject identity], [wardrobe], [background], [timing] identical. No cuts. No new objects besides [delta].

What to put in each line (fast reference)

Line Include Avoid
LOCK framing/motion, lighting, style, subject, location, action (and dialogue if needed) (https://deepmind.google/models/veo/prompt-guide/) extra “vibe” adjectives that invite redesign
DELTA one change, with placement + interaction + timing multiple edits bundled together
RECONFIRM “same camera,” “same person,” “no cuts,” “no new objects” “make it better,” “clean it up,” “more cinematic”

Worked example: add the product in the first second (without breaking the shot)

Problem: you already have a UGC-style hook, but you need the creator to hold the product immediately. Social tests suggest clips that work tend to show a recognizable subject within the first second, keep one clear motion/transition, and end before degradation (https://www.vidu.com/blog/social-media-video-ai). So we’ll add a prop without changing anything else.

Before (drift-prone)

A creator talks to camera in her kitchen. Make it cinematic. Add the product in her hand.

Why it fails: “cinematic” changes lighting, lensing, motion, color; “add the product” is underspecified.

After (bounded delta)

LOCK: Vertical 9:16, medium close-up, static camera, no cuts. Soft window light from camera-left. Realistic textures, natural color.
Subject: same creator speaking directly to camera. Location: tidy kitchen. Action: speaking continuously.

DELTA (ADD): Add a small [YOUR PRODUCT] container in the subject’s right hand, held still near chest height from 00:00 to 00:02.

RECONFIRM: Keep the same face, hairstyle, outfit, kitchen background, framing, and lighting unchanged. No new props besides the product container. No cuts.

Why it works:

  • The LOCK includes the prompt components Veo expects (framing/motion, lighting, character, location, action) (https://deepmind.google/models/veo/prompt-guide/).
  • The delta is one object, with placement and time bounds.
  • RECONFIRM removes “helpful” additions (new props, new camera moves).

Ingredient specificity: the three constraints that prevent drift

1) Anchor the subject (“who/what”)

Google’s video prompt guidance defines the subject as the “who” or “what” the action revolves around (https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/video/video-gen-prompt-guide). In ingredient edits, restate the subject plainly (e.g., “same creator,” “same product reviewer”) so identity doesn’t shift.

2) Make placement geometric

For objects, include:

  • position (“right hand,” “back-left shelf”)
  • scale (“palm-sized,” “small box”)
  • interaction (“held still,” “placed down,” “tapped twice”)

3) Don’t mix “edit” language with “new direction”

Replace:

  • “make it more premium”
  • “make it more cinematic”
  • “add cool accessories”

With bounded deltas:

  • “Increase contrast slightly” (expect some drift risk)
  • “Add a plain silver watch on the left wrist; no other changes”
  • “Remove the red toy on the coffee table; everything else unchanged”

Copy‑paste patterns (ADD / REMOVE / REPLACE)

Keep the order: LOCK → DELTA → RECONFIRM.

ADD: handheld prop

LOCK: Vertical 9:16, medium close-up, static camera, no cuts. Soft window light from camera-left. Realistic textures.
Subject: creator speaking to camera. Location: tidy kitchen. Action: speaking continuously.
DELTA (ADD): Add a matte-white ceramic mug in the subject’s right hand, held near chest height.
RECONFIRM: Keep the same person, face, hairstyle, outfit, kitchen background, framing, and lighting unchanged. No new objects besides the mug.

ADD: background product (no readable text)

LOCK: 16:9, locked-off tripod shot, eye-level, minimal motion, no cuts. Bright neutral daylight.
Subject: product reviewer at a desk. Location: home office. Action: talking and gesturing lightly.
DELTA (ADD): Add a small [PRODUCT BOX] on the back-left shelf, label facing camera but with no readable text.
RECONFIRM: Keep camera position, depth of field, desk items, wardrobe, and color identical. Do not add any other new props.

REMOVE: logos/readable text without wardrobe changes

LOCK: Vertical 9:16, medium shot, handheld micro-shake, no cuts. Warm indoor lighting.
Subject: person walking and talking on a sidewalk.
DELTA (REMOVE): Remove any visible logos, brand marks, or readable text from clothing and accessories; replace with plain unbranded surfaces.
RECONFIRM: Keep the same jacket color, fabric type, fit, sidewalk location, camera movement, and pacing.

REPLACE: swap a prop, preserve timing

LOCK: 16:9, close-up of hands on a tabletop, fixed top-down angle, no cuts. Soft diffuse lighting.
Action: hands slide an object into frame, pause, then tap it twice.
DELTA (REPLACE): Replace the object with a [NEW OBJECT] of similar size, keeping the same hand path and timing.
RECONFIRM: Keep camera angle, lighting, tabletop texture, and hand appearance identical.

2-minute QA pass (catch continuity breaks fast)

Run this checklist every iteration:

  1. Framing: did the crop, angle, or camera motion change?
  2. Hands/contact: is the object actually held/placed (not floating/warped)?
  3. Text/logos: did readable text appear unintentionally?
  4. Lighting continuity: did key light direction or color temperature shift?
  5. Motion continuity: any surprise cut, push-in, or pacing change?

If something breaks: tighten the RECONFIRM clause and reduce the delta (don’t rewrite the entire prompt).

Common issues and fixes

Object warps in the hand

Constrain interaction: “held still,” “object remains rigid,” “no finger deformation.” If the hand already holds something, consider REPLACE instead of ADD.

Style shifts (suddenly more animated/filmic)

Remove new-direction adjectives and restate the LOCK style (“realistic textures,” “natural color”) (https://deepmind.google/models/veo/prompt-guide/).

Camera changes

Add: “same framing, same camera height/angle, same movement (or lack of movement), no cuts.”

New props appear

End with: “No new objects added besides [X].”

Prompt gets blocked or sanitized

Google’s platform applies safety filters, and prompts that violate responsible AI guidelines can be blocked (https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/video/video-gen-prompt-guide). Rewrite with neutral physical description; remove sensitive context.

Mini playbook: 5 creator use-cases you can ship this week

  1. UGC ads: keep the same hook and gestures; REPLACE “blue bottle” → “green bottle.”
  2. Product shots: REMOVE background logos/text so your product stays the hero.
  3. Real estate reels: REMOVE personal items while reconfirming room layout and lighting.
  4. Shorts hooks: ADD one first-second visual anchor (recognizable subject + one clear motion) (https://www.vidu.com/blog/social-media-video-ai).
  5. Variant testing: keep LOCK identical; rotate only the DELTA line (prop, wardrobe, background hero item).

Checklist

  • Write a LOCK line: aspect ratio, framing/motion, lighting, style, subject, location, action.
  • Choose one DELTA: add or remove or replace.
  • Include placement + interaction + timing in the DELTA.
  • Add a RECONFIRM clause: same camera, same subject identity, same background, no cuts, no new objects besides X.
  • Run the 2-minute QA: framing, hands/contact, text/logos, lighting, motion.
  • If it drifts: delete vague “vibe” adjectives; tighten RECONFIRM; shrink the delta.

FAQ

### What are “ingredient prompts” in Flow—are they just normal prompts?

Flow describes Ingredients to Video as using multiple reference images to control characters, objects, and style (https://blog.google/innovation-and-ai/products/veo-updates-flow/). Ingredient prompting (as a creator practice) means treating your request as a bounded edit anchored by what must remain consistent.

### How do I add a prop without changing the character or room?

Use LOCK → ADD → RECONFIRM. Put placement + timing in the ADD line and end with “No new objects besides the prop; keep the same person/outfit/background/framing/lighting; no cuts.”

### How do I remove logos or readable text without the model changing wardrobe?

Make the DELTA explicitly about logos/text, then RECONFIRM the wardrobe: “keep the same jacket color, fabric type, and fit.” Otherwise the model may “solve” removal by swapping clothing.

### How do I keep the camera angle identical across iterations?

Lock framing/motion in LOCK (“static camera, no cuts”), then restate it in RECONFIRM (“same camera height/angle, same movement”). Avoid adding new style direction that implies new lensing.

### Why does replacing a shirt sometimes change the person’s face?

Because “wardrobe change” can be interpreted as a new character design unless you restate identity. Put “same person, same face, same hair” in LOCK and RECONFIRM.

### What if my prompt is blocked or the output is sanitized?

Safety filters can block prompts that violate responsible AI guidelines (https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/video/video-gen-prompt-guide). Rewrite the request using neutral physical details and remove sensitive context.

Create stable ingredient variants with native audio (without enterprise pricing)

If you’re doing repeated ingredient edits (prop swaps, compliance removals, hook variants), Veo3Gen is an affordable way to access Google’s Veo 3.1 video models without Google’s enterprise pricing. It supports text-to-video and image-to-video, offers Veo 3.1 Fast / Quality / Lite modes, and generates native synchronized audio (dialogue, SFX, music) in a single pass. Supported resolutions include 720p, 1080p, and 4K (4K on Fast/Quality) with 16:9 and 9:16 aspect ratios.

Closing CTA: Start with Veo3Gen’s free credits, then scale with pay-as-you-go credits (purchased credits don’t expire) or optional monthly plans—and if you want to automate batches, use the developer API to run Lock → Delta → Reconfirm variants programmatically.

Start creating with Veo3Gen

Veo3Gen gives you affordable Veo 3.1 video generation with native audio, up to 4K, and credits that never expire — with free credits to start.

Limited Time Offer

Try Veo 3 & Veo 3 API for Free

Experience cinematic AI video generation at the industry's lowest price point. No credit card required to start.

Get Started FreeView Cheapest Pricing
← Back to all posts