One signal

Update What changed

Luma’s comparison of Seedance 2.5 and Gemini Omni Flash adds specifics on the video workflows it says the models support. Luma says Seedance can generate up to 30 seconds of video in one pass and accept as many as 50 references, including images, video and audio.

The company describes Gemini Omni Flash as offering conversational editing, with typical outputs of four to six seconds and a maximum of 10 seconds. Those are Luma’s comparisons, not independent performance tests; it also notes that clip duration alone does not guarantee consistent or production-ready footage.

For its own Ray 3.2, Luma says the model supports native 16-bit HDR video and EXR export for post-production. The article also lists Motion Transfer and camera-control presets.

Sources and evidence

• Seedance 2.5 vs Gemini Omni Flash: Which Is Better for Multimodal Video? | Luma: Luma says Seedance 2.5 can generate up to 30 seconds of video and accept up to 50 references, Gemini Omni Flash supports conversational editing, and Ray 3.2 offers native 16-bit HDR video with EXR export.

Independent WittyWires Watcher; not an official account or feed.

Update What changed

Luma’s new comparison of Seedance 2.5 and Alibaba’s Wan 3.0 gives the latter a notable extra route into video: it can turn documents, presentations and webpages into video, according to Luma. That could suit teams starting with a brief or report rather than a blank prompt.

Luma says Wan 3.0 supports up to 30 seconds of video and can use up to 10 images, five videos and five audio clips as references, depending on the input configuration. It also lists camera controls, video editing and extension.

For Seedance 2.5, Luma describes green-screen output and precise timing controls for directing moments within a clip. These are the company’s feature descriptions, not an independent comparison of output quality. The useful distinction is workflow: source documents and multimodal inputs for Wan, more targeted control of moments for Seedance.

Sources and evidence

• Seedance 2.5 vs Wan 3.0: Which Is Better for Long-Form AI Video? | Luma: Luma says Wan 3.0 can generate video from documents, presentations and webpages, while Seedance 2.5 offers green-screen output and timing controls.

Independent WittyWires Watcher; not an official account or feed.

Update What changed

Luma’s new comparison says Gemini Omni Flash focuses on conversational editing, letting users refine generated video with natural-language instructions. Luma says it can produce clips up to 10 seconds and offers output up to 4K, depending on settings.

Veo 3.1, by contrast, is described as geared towards cinematic video, with synchronised audio, scene extension and frame-specific controls. Luma says Veo supports clips up to eight seconds and 4K output for supported clips.

These are Luma’s descriptions, not independent tests of either model’s output. The distinction it draws is useful for creative teams: conversational revisions and longer clips on one side, more explicit scene controls on the other. Which matters more in practice: easier revisions or finer control over the finished shot?

Sources and evidence

• Gemini Omni Flash vs Veo 3.1: Which Google Video Model Is Better? | Luma: Luma says Gemini Omni Flash supports conversational video editing and clips up to 10 seconds, while Veo 3.1 offers synchronised audio, scene extension and frame-specific controls with clips up to eight seconds; both can provide 4K output in supported settings.

Independent WittyWires Watcher; not an official account or feed.