Veo first-last is not Veo R2V

VEO First Last Frame is 64 credits, image-to-video, native audio true, 4K, 4 / 6 / 8 seconds, image required. VEO 3.1 Reference to Video is 128 credits, reference-to-video, native audio true, 8 seconds only, image required, resolution unset. First-last interpolates between two frames. R2V extracts identity from a stack and puts it in a scene the prompt describes. Sixty-four is the morph at its 4s floor. One hundred twenty-eight is the cast, which only comes in eight seconds.

This is not Veo 3.1 text-to-video. That card is also 64 credits, native audio, 4K, 4 / 6 / 8 seconds, no image required. Same sticker as first-last, different input. This page is first-last versus R2V.

Both rows speak. audio is true on both. Write whether anyone speaks. SKU interpolates usually want Foley and no voice. Say “no speech.” A surprise monologue is still a wasted morph, 128 credits at 4s with sound. On R2V it is a more expensive wasted cast: 256 for the full eight seconds with sound.

First-last is two frames and a path

Bring a start still and an end still you would print. The model fills the time between them. Frame one is your upload. Frame last is your upload. The kitchen in the first photo should survive if the last photo is the same kitchen. That is interpolation, not a new location.

Duration: 4s, 6s, 8s. There is no 10s, no 15s, no 3s test. A 4s closed-to-open is a snap. An 8s closed-to-open is the long morph. Cutting an 8s take down is fine. Paying 8s because you wanted 10s and Veo felt adjacent to Omni is not. Gemini Omni Video is 29 credits, 4 / 6 / 8 / 10 seconds, 4K, audio null, and it can take references. Different flag, different sticker, different job.

Resolution: 4K on first-last. That is the ceiling listed. Do not treat 4K as optional personality. If the delivery is 1080p, you can still run 4K and downscale, or you can leave this family. The card you picked is the 4K I2V morph.

Sixty-four credits is the catalog sticker for the 4s silent floor. Same as Veo T2V. You are not paying extra for two frames versus a sentence. You are paying the Veo 3.1-class rate, 16 credits a second silent and 32 with sound, for a different gate. R2V bills the same rate; its 128 is the eight-second-only length, not a sharpness tax.

When first-last is the pick:

You have the before and the after as stills.
The motion in the middle should be a path, not a new performance in a new room.
4, 6, or 8 seconds at 4K with native audio is the file.
Identity is already in those two frames. You do not need a third still of the same bottle.

When first-last is the wrong pick:

The product must appear in a place you did not photograph. That is R2V.
You have a stack of pack angles and a talent still, and the scene is a sentence. That is R2V.
You have only a sentence. That is Veo T2V at 64.
You needed 8 seconds of a new scene with three references, not a morph. That is R2V at 128.

R2V is a stack, not frame one

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 stack, not a storyboard frame.

Duration: 8s only. There is no 4s draft on this slug. There is no 6s. You buy the 8-second cinematic clip whose subjects match up to three reference images, with native audio — that is the published description. A 4s look-dev does not exist here. If you are not sure the stack works, you will still spend 128 on an 8s take. That is why the stack has to be ready.

Resolution is unset on R2V. Do not write 4K into the brief as if the field were filled. First-last lists 4K. R2V does not. Do not treat a missing token as 720p either. Crop the stills to the delivery ratio before you upload so the reasoning pass is not also a reframing pass.

One hundred twenty-eight credits is the catalog sticker. Do not turn it into a new per-second Veo rate: the rate is the same 16 silent and 32 with sound as the other Veo rows. 8 seconds at 128 is not “more Veo.” It is the reference door.

When R2V is the pick:

The product must appear in a place you did not photograph, and you have pack stills.
The spokesperson must walk a new blocking without a new shoot, and you have a talent still.
You are building a series and the same bottle has to survive clip nine.
You already paid for photography. The sentence is the cheap part. The stack is the brief.

When R2V is the wrong pick:

You have two frames that already are the shot. First-last at 64 will interpolate them.
You have no stills. R2V requires an image. Veo T2V does not.
You needed 4s or 6s. This slug only sells 8s.

64 versus 128 is the category and length gap

Do not run R2V to “see if the morph is better.” R2V will not morph your two frames as a path. It will recast them. Do not run first-last to “lock identity in a new kitchen.” The kitchen in frame one will fight the kitchen in frame last. If you need a new kitchen, bring kitchen as a reference on R2V, or generate the kitchen on T2V and lose the pack.

Flux 3 First Last Frame to Video is 34 credits, native audio true, 5–20 seconds, if 64 is too much Veo for a morph and 4K is not the brief. Seedance 2.0 is 44 credits to start, 1080p, 4–15 seconds, text-to-video and reference-to-video, native audio, if you need a longer ref door than Veo’s 8s. Those are other families. This page stays on Veo’s two doors.

Google Flow will run Veo. Flow is a surface. Keep Veo. Do not live in Flow because both of these models are Google. The catalog shows which job. First-last versus R2V is the job.

The AI video generator is the launcher. Captions still belong on a talking 4K morph; mute autoplay hides native audio. Captions for silent autoplay is the typesetting. The AI caption generator burns the line.

Article 50 still applies to photoreal people in the stills. Sixty-four credits does not exempt a generated face. Label at first exposure for EU audiences since 2 August 2026. The Article 50 disclosure template is the copy. YouTube’s 27 May 2026 labelling pass overlays photoreal AI on Shorts. Labels do not turn an interpolate into a reference stack.

Sora 2 Pro storyboard was I2V, native audio, 10 / 15 / 25 seconds, 240 credits. The Sora API shuts 24 September 2026 (OpenAI Help Centre, September 2026). A leftover Sora two-frame brief is closer to first-last at 64 if 4K and 8s are enough. It is not closer to R2V unless you have a stack and want a new scene.

After the morph or the cast

Lock picture, then caption. If you still need more pixels on a 720p cousin from another family, that is an upscale row, not a reason to rerun R2V. If the morph smeared the label, fix the stills. Rerunning R2V will not un-smear an interpolate you never ran.

Sixty-four credits is the path. One hundred twenty-eight credits is the stack. Same native-audio flag, same per-second rate. Different input math. Pick the math.

Run the Veo door you mean in Versely.

FAQ

Is Veo first-last the cheap R2V?

No. First-last is I2V interpolation at 64 credits, 4K, 4 / 6 / 8 seconds. R2V is a 128-credit 8s reference stack. Different category.

Does R2V output 4K like first-last?

First-last lists 4K. R2V does not list a max resolution on the published card. Do not invent 4K for R2V.

Can I draft on first-last and promote to R2V?

Only if the job changed from a two-frame path to a new scene with a stack. Promoting a morph to R2V is a new brief, not a quality step.