Veo reference-to-video vs I2V

Pick Veo reference-to-video when the stills are the brief: this person, this bottle, new scene. Pick image-to-video when the plate is frame one: this kitchen, this light, now roll. The stack is not a first frame. VEO 3.1 Reference to Video is Google’s published R2V row — 8 seconds, up to three reference images, native audio, 64 credits. Versely does not publish a single-still veo-3-1-image-to-video slug. Google’s I2V-adjacent Veo card is VEO First Last Frame (two stills as endpoints, 4 / 6 / 8s, 4K, 40 credits, native audio). Everyone else’s I2V lives under image-to-video. This page is the pick. The meter post is a different URL.

Do not recast a locked plate on R2V. Do not restage a cast on I2V. That mix is the whole waste.

R2V: the stills are the cast

Reference-to-video extracts identity and puts it in a scene the prompt describes. Frame one is not your upload. The kitchen in the photo does not have to survive. The face and the SKU do. Film language: you handed the model a headshot and a pack shot, not a storyboard frame.

Veo’s published R2V contract:

• Inputs: up to three stills. Front of the spokesperson, three-quarter of the SKU, one clean plate of the room — then stop. A fourth near-duplicate averages two lighting setups into a third face.
• Duration: 8s only. No 4s draft. No 15s hold. Write an eight-second shot or change families.
• Audio: on. Write the room. Dialogue, or none; footsteps, or none; music, or none. A silent prompt is not a mute plate.
• Credits: 64. That is the catalog figure. Spend cheap stills first. Re-rolling 64 credits because the label is the wrong color is how a reference model becomes a sketchbook.
• Requires an image. Text-only will not run.

The prompt then names action and camera, not hair color. The stills already answered who. Re-describing the subject in text is how the stack and the sentence fight.

Hygiene: same crop family, same light, no mixing a studio packshot with a phone grab. Mixed refs are how you get a person who is in none of the photos. Register the set once so clip nine pulls the same keys as clip one.

When R2V is the pick:

• The product must appear in a place you did not photograph (beach, night kitchen, trade-show floor).
• The spokesperson must walk a new blocking without a new shoot.
• You have three approved stills and an eight-second cinematic with Veo audio.

When R2V is the wrong pick:

• The approved hero is the first frame and the background must stay. That is I2V.
• You have two endpoints of a move (closed pack / open pack). That is first-last, not a three-still recast.
• You need 20–30 seconds of one hold. That is Seedance 2.5, not 8s of Veo.

The category definition lives at reference-to-video. The Veo card is the Google implementation at 64 credits.

I2V: the still is frame one

Image-to-video anchors generation to the picture you uploaded. Composition, background, and lighting are the opening frame. The model invents motion forward. It does not owe you a new location. Ask I2V for “this bottle on a beach” and you will watch your studio sweep try to become a beach and fail. The image-to-video glossary is the one-sentence version: your image is the first frame.

On this Google roster, the published I2V-shaped Veo row is first-last: two stills, interpolation between them, 4 / 6 / 8 seconds, 4K, 40 credits, native audio. One still is not “close enough” for that card. One still is a different job, and faking the second in prose wastes interpolation on a destination that never existed. If you only have one Veo-looking plate and you needed it to move, you are on a non-Veo I2V row or on Gemini Omni Video, which lists image-to-video among its categories at 4–10s, 9 credits. Do not invent a Veo I2V slug to close the gap.

When I2V is the pick:

• The photograph is the ad. You already fought lighting and styling. You need steam, a slow push, a hand entering — not a new room.
• The client approved a pixel-accurate pack shot. Recasting it is a brand risk.
• You are interpolating two approved frames (Veo first-last).

When I2V is the wrong pick:

• “Same person, different world.” R2V.
• “Same dance, new mascot.” Kling motion control.
• “Same prompt, no still.” Text-to-video: Veo 3.1 is 4 / 6 / 8s, 4K, 40 credits, native audio, no image required.

Cheap I2V exists when Veo’s 40 or 64 credits are the wrong sketchbook: Hailuo 2.3 Fast is 6s / 10s, 28 credits, image required. Use it to see whether the plate even wants motion. Then spend Veo.

The pick, in one pass

Ask one question: must frame one be this file?

Yes → I2V (or Veo first-last if you have two endpoints). The plate is the brief.

No, but these faces and labels must survive a new blocking → Veo R2V. The stack is the brief.

No, and I do not even have a still → Veo 3.1 text-to-video.

No, and the still is only a style hint among several files → Seedance 2.0 reference or Omni Video’s quota, depending on family. Not Veo R2V’s three-still cinematic.

Secondary checks, only after that question:

• Length. Veo R2V is 8s. Veo first-last is 4 / 6 / 8s. Veo T2V is 4 / 6 / 8s. Longer holds leave the Veo length menu.
• Audio. All three Veo cards above have native audio on. Write the track. I2V on Hailuo 2.3 Standard/Pro has audio off — a different finishing plan.
• Money. 64 credits is not a reason to max a stack you have not locked. Generate stills on text to image until you would print them. Then R2V once.
• Omni 1.1 first-last is Google’s 27 August interpolation on gemini-omni-1.1-flash, not a Veo card. Do not mix that control into this pick.

Finish either file like a file: add captions, crop, disclose if Article 50 applies (disclosure template). R2V will not typeset. I2V will not typeset.

The AI video generator is the launcher. Open the R2V card when the stack is the brief. Open I2V when the plate is frame one. Do not open R2V because three upload slots looked like a good place to dump hero JPEGs you already want as frame one.

This pick is not the meter page

There is already a Veo R2V page that is the SKU: eight seconds, 64 credits, three stills, native audio. That page exists so you do not max a slider you have not locked. This page exists so you do not put the hero plate in the recast slot.

If you arrived here from a cost question, go read the meter. If you arrived here from a failed generate that “ignored the background,” you used R2V on an I2V brief. If you arrived here from a failed generate that “changed the kitchen I approved,” same error. If you arrived here from a failed generate that “would not leave the studio sweep,” you used I2V on an R2V brief.

Omni 1.1 first-last (Google, 27 August 2026) is interpolation between two stills on gemini-omni-1.1-flash. It is not Veo R2V. It is not Veo I2V. Do not add it to this pick because both say Google.

Seedance 2.0 reference is a twelve-file-class brief on a different family. Veo R2V is three stills and eight seconds. If your stack is twelve files, you are not in this pick.

If you are routing Veo this week, sign up and confirm R2V only after the three stills are the cast — not the set.

FAQ

Is Veo R2V just I2V with extra pictures?

No. I2V treats your file as frame one. R2V extracts identity and restages. VEO 3.1 Reference to Video is the latter: up to three stills, 8s, native audio, 64 credits. Uploading a hero plate there will recast it.

Where is Veo image-to-video on Versely?

There is no published veo-3-1-image-to-video slug. The Veo I2V-adjacent card is VEO First Last Frame (two stills). Single-still motion is other I2V rows or Omni Video’s image-to-video category.

Why not always R2V “to be safe”?

Because 64 credits and an 8s recast will ignore a plate you already approved. If frame one must be that file, I2V is cheaper in mistakes even when the sticker is lower. Safety is matching the contract, not uploading more photos.