Veo 3 Video Generation Compared: How to Choose Across Four Core Scenarios

Introduction: Veo 3.1 and the New Video Generation Stack

Google Veo 3.1 represents an evolved generation stack for AI video: sharper motion coherence, stronger prompt adherence, and a clearer split between “pure creation,” “frame-locked storytelling,” “reference-locked subjects,” and “timeline continuation.” On FuseAI Tools, the same conceptual surface maps to four practical routes—each tuned for a different production moment.

Teams often stall on the same questions: what separates TEXT_2_VIDEO, FIRST_AND_LAST_FRAMES_2_VIDEO, REFERENCE_2_VIDEO, and VEO3_EXTEND, and which one should ship first?

This guide walks through representative parameter shapes, compares constraints side by side, and ends with a decision tree you can reuse in briefings.

Veo hub: /home/veo3

I. Snapshot: Four Scenarios at a Glance

Scenario	TEXT_2_VIDEO	FIRST_AND_LAST_FRAMES_2_VIDEO	REFERENCE_2_VIDEO	VEO3_EXTEND
Core function	Text-only generation	1-2 images + prompt	1-3 refs + prompt	Extend prior clip
Inputs	Prompt	Images + prompt	Refs + prompt	Task ID + prompt
Output profile	Standard duration tiers	Standard duration tiers	Standard duration tiers	Timeline lengthening
Resolution class	720p / 1080p / 4k	720p / 1080p / 4k	720p / 1080p / 4k	fast / quality / lite
Positioning	Zero-to-one ideation	Keyframe control	Look consistency	Continuation

II. Model Stack and Shared Controls

2.1 Core model identifiers

{
  "model": "veo3 / veo3_fast / veo3_lite"
}

Model	Trait	Best for
veo3	Balanced quality	General delivery
veo3_fast	Fast iterations	Speed-sensitive workflows
veo3_lite	Lightweight	Budget or latency caps

2.2 Common generation parameters

Field	Meaning	Typical values
aspectRatio	Frame geometry	16:9 / 9:16 / Auto
resolution	Output sharpness	720p / 1080p / 4k
watermark	Optional overlay text	Custom string
enableTranslation	Prompt translation assist	true / false (often true)

III. Four Scenarios Deep Dive

Scenario 1: TEXT_2_VIDEO

Pure prompt-to-video creation—no reference frames required.

{
  "model": "veo3_fast",
  "generationType": "TEXT_2_VIDEO",
  "prompt": "A dog playing in a park",
  "imageUrls": [],
  "aspectRatio": "16:9",
  "resolution": "720p",
  "enableTranslation": true,
  "watermark": "MyBrand"
}

Notes: keep imageUrls empty; rich prompts beat single-line labels; up to 4k-class output when the route allows.

Strength: maximum creative freedom. Trade-off: no pixel-level anchor for specific on-screen elements.

Route: /home/veo3/text-to-video

Scenario 2: FIRST_AND_LAST_FRAMES_2_VIDEO

Drives motion from one or two still references—either animate a single plate or interpolate between explicit endpoints.

{
  "model": "veo3_fast",
  "generationType": "FIRST_AND_LAST_FRAMES_2_VIDEO",
  "prompt": "The dog continues running through the park",
  "imageUrls": ["http://example.com/first.jpg", "http://example.com/last.jpg"],
  "aspectRatio": "16:9",
  "resolution": "1080p",
  "enableTranslation": true
}

Image rules: one image → bring the plate to life; two images → first frame + last frame with AI-filled middle.

Route: /home/veo3/first-and-last-to-video

Scenario 3: REFERENCE_2_VIDEO

Locks character or object appearance via one to three reference stills while language steers action.

{
  "model": "veo3_fast",
  "generationType": "REFERENCE_2_VIDEO",
  "prompt": "The character is walking through a futuristic city",
  "imageUrls": ["http://example.com/character1.jpg", "http://example.com/character2.jpg"],
  "aspectRatio": "9:16",
  "resolution": "720p"
}

Constraints: commonly limited to veo3_fast; one to three references; contrasts with first/last mode because refs define look, not necessarily exact start/end frames.

Route: /home/veo3/reference-to-video

Scenario 4: VEO3_EXTEND

Continues an existing Veo-generated clip using the originating task metadata plus a continuation prompt.

{
  "task": "Original Veo generation task id",
  "model": "fast",
  "prompt": "The dog continues running, clearing small hurdles and joining other dogs",
  "seeds": 12345,
  "watermark": "MyBrand"
}

Notes: source clip must come from Veo generation; extension tiers map to fast / quality / lite; follow platform rules about resolutions that remain eligible for extend (e.g., restrictions after certain 1080p outputs).

Route: /home/veo3/extend

IV. Comparison Summary

4.1 Capability matrix

Feature	TEXT	FIRST/LAST	REFERENCE	EXTEND
Prompt-only	✅	❌	❌	❌
Still-image inputs	❌	✅ (1-2)	✅ (1-3)	❌
Character / look lock	❌	❌	✅	✅
Spatial keyframe control	❌	✅	❌	❌
Timeline continuation	❌	❌	❌	✅
4k-capable generation	✅	✅	✅	❌
Model flexibility	veo3 / fast / lite	veo3 / fast / lite	Typically fast	fast / quality / lite

4.2 Decision tree

What is your task?
|
|-- No reference assets
|   `-- TEXT_2_VIDEO
|
|-- You have still images
|   |-- Need explicit start/end framing
|   |   `-- FIRST_AND_LAST_FRAMES_2_VIDEO
|   `-- Need subject appearance lock
|       `-- REFERENCE_2_VIDEO (fast-class stack)
|
`-- Already have a Veo-generated clip to lengthen
    `-- VEO3_EXTEND (respect resolution eligibility rules)

V. Why Veo 3.1-Style Workflows Matter

5.1 Four-mode coverage. Ideation, controlled interpolation, identity-consistent heroes, and continuation sit in one policy framework instead of forcing a single generic prompt.

5.2 Resolution headroom. Generation modes often scale through 720p, 1080p, and 4k-class outputs—extend focuses on temporal continuation rather than chasing the same resolution ladder.

5.3 Translation-aware prompting. When enableTranslation stays on, multilingual briefs map more reliably into the model’s preferred instruction language.

5.4 Extend tuning. Fast, quality, and lite continuation modes let producers trade velocity for fidelity when stitching longer narratives.

VI. Final Recommendations

Use case	Mode	Why
Create from scratch	TEXT_2_VIDEO	Pure language freedom
Animate one still	FIRST_AND_LAST (1 image)	Motion without extra anchors
Lock start/end frames	FIRST_AND_LAST (2 images)	Hard endpoints
Keep character look stable	REFERENCE_2_VIDEO	Reference-conditioned identity
Lengthen an approved clip	VEO3_EXTEND	Task-linked continuation

One-line playbook:

No assets → TEXT_2_VIDEO.
Keyframes or still animation → FIRST_AND_LAST_FRAMES_2_VIDEO.
Identity from refs → REFERENCE_2_VIDEO (fast stack).
Continue an existing Veo render → VEO3_EXTEND.

Open every scenario from the Veo overview: /home/veo3.