Kling 4 vs Veo 3.1 for consistent characters and products

Status and access

Veo 3.1 is Google's video model with a reference mode Google calls Ingredients to Video. It is available through Google's APIs and through providers such as fal.ai. One change to watch: in the Gemini API, the Veo 3.1 preview model IDs are scheduled to shut down on October 22, 2026. Google points developers to Gemini Omni Flash (gemini-omni-1.1-flash) or to the generally available Veo models on Google Cloud. If you call the preview IDs directly, check your integration now.

Kling 4.0 launched in Kling's app on October 15, 2026, after an early-access phase that began in late September. Its developer API is marked as coming soon. fal.ai lists the model with prices, but API access there is also marked as coming soon.

References: three images versus fifteen inputs

Veo 3.1 accepts up to three reference images of a single person, character or product, per Google's documentation. References work on Veo 3.1 and Veo 3.1 Fast, not on Veo 3.1 Lite. Three images means choosing carefully: typically a front view, a three-quarter view and one detail or profile.

Kling 4.0's Omni Reference takes up to 15 inputs in total: up to 10 images, up to 5 videos with 30 seconds combined, up to 7 elements and up to 3 voice profiles, per Kling's guide. Elements let you define several subjects at once, for example a presenter and the product they hold, each with its own images.

For a single face, three good images can be enough. For two characters in one shot, or a product whose front, back and label all matter, the larger budget is a real advantage, if Kling 4.0 uses the extra references well. We have not tested that yet.

Clip length, resolution and audio

Veo 3.1 generates 4, 6 or 8 seconds, but generations with reference images must be 8 seconds. Output is 720p by default, with 1080p and 4K at 8 seconds only, in 16:9 or 9:16. Audio is always generated. A clip can be extended in 7-second steps up to 148 seconds in total, but extension runs at 720p.

Kling 4.0 generates 3 to 30 seconds in one pass, at 720p, 1080p or 4K, in 21:9, 16:9, 1:1 or 9:16, with stereo audio. A separate keyframe mode takes up to 10 keyframes.

Price

On fal.ai, Veo 3.1 reference-to-video costs $0.20 per second without audio or $0.40 with audio at 720p or 1080p, and $0.40 or $0.60 at 4K. An 8-second 1080p clip with audio is $3.20.

fal.ai lists Kling 4.0 at $0.157 per second at 720p and $0.235 at 1080p, but since the API is not open, nobody can run a production cost comparison yet. For reference, Kling 3.0 Omni on fal.ai costs $0.084 to $0.14 per second, depending on tier and audio.

Where each is stronger

Veo 3.1 suits short, polished shots where one subject must look right: an 8-second product hero shot, a presenter's opening line, a single character beat. It is available, documented and priced, which matters if you deliver to clients.

Kling 4.0 is the better fit on paper for several subjects or long continuous action: a mascot with a product, a two-character dialogue, a 20- to 30-second ad as one take. The catch is access. Until its API opens, you can only use it in Kling's own app and tools.

We have not run the same references through both models, so we make no claim about which holds a face or a logo more faithfully.

Which to pick for consistent characters and products

Pick Veo 3.1 for one subject per shot and short clips you need to produce through an API today. Spend your three slots on a front view, a three-quarter view and the most identifying detail.

Wait for Kling 4.0, or use Kling 3.0 Omni in the meantime, when shots need more than three angles, more than one subject, or longer takes.

RefVid takes the second route. It runs on Kling 3.0 Omni today, with up to five reference images per character or product, and will switch to Kling 4.0 when its API opens. It does not offer Veo 3.1.