Google’s Gemini Omni Flash 1.2 generates short videos, while Alibaba’s Qwen3.8-Omni-Flash analyses text, images, audio and video but returns text. OrcaRouter argues they are better understood as different tools in a video workflow than as competitors on a single leaderboard.
Watch Desk analysis
What happened
The OrcaRouter comparison says Gemini takes text, up to six reference images and reference video, then generates video with synchronised audio. It supports clips of three to 15 seconds, extendable to 30 seconds. Qwen can take in video, including footage over an hour long in one call, but produces text rather than video.
That makes a direct video-generation scorecard impossible: Qwen does not generate video and is absent from the video leaderboards OrcaRouter cites. Their listed prices also measure different jobs. OrcaRouter gives Gemini a video-output rate of $12 per million tokens and says a 15-second 720p clip costs about $1.05 in output tokens. Qwen’s listed input rate is $0.15 per million tokens; video costs depend on how footage is sampled.
Why it matters
OrcaRouter’s proposed workflow is to use a reader model to describe source footage, then feed that text into a video generator. That could help a team prepare a shot list or check a finished clip for continuity before spending more on generation. It is a suggested workflow, not a vendor-backed integration or a guarantee that the reader will catch every problem.
The comparison also flags a practical restriction: it says Gemini’s editing and video-extension features are unavailable in the UK, EEA and Switzerland. That matters to British video teams considering the service, whatever the headline generation price suggests.
Our read
The useful comparison is not “which model wins?” but “which job needs doing?” Qwen is the reader; Gemini is the renderer. OrcaRouter’s cost estimates and workflow advice are its own analysis, not a neutral benchmark or a promise of production results. Still, the distinction is valuable: buying a video model to analyse footage, or a reader to produce a finished clip, is a category error with an invoice attached.
What to watch
- Whether Google changes the UK availability of Gemini’s editing and extension features.
- How video sampling settings affect Qwen’s cost and usefulness on real footage.
- Whether teams publish results from reader-and-generator workflows, including how often they catch continuity errors.
Discussion spark: For a video workflow, would you pay for a separate AI reader to check footage before generation, or keep the process simpler and accept more trial-and-error?
Sources and evidence
- Gemini Omni Flash 1.2 vs Qwen 3.8 Omni Flash: video vs text – OrcaRouter (11 November 2026, 00:50 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.