GEMINI OMNI FLASH + EZUGC

Gemini Omni Flash is Google's multimodal video model, available in EzUGC through Runware. It generates video with synchronized audio and accepts text, images, and video as input. Alongside text-to-video and image-to-video, it can edit a clip you already have and continue one of its own earlier generations.

Yes. Audio is produced in the same pass as the picture, so dialogue, effects, and ambience come back attached to the clip. Every demo on this page was generated with the model and kept the audio track it returned. Because the sound is generated rather than added later, it is worth directing: put the spoken line in quotes and say what should be audible underneath it.

You can point a new generation at a previous Omni clip and keep going from it rather than starting fresh. The character, wardrobe, and location carry across, so you can add a second beat to a take that already works. The two multi-turn samples above are exactly that: one generation, then a continuation of it.

It can. Give it a source clip and describe the change, and it rewrites the scene while holding the original camera move and timing. The paired product samples above show the same generated clip before and after being moved from a dark studio backdrop into a sunlit kitchen. Duration and framing are inherited from the source, so an edit returns a clip the same length as the one you sent.

It outputs 720p in landscape (1280x720) or vertical (720x1280), at durations between 3 and 10 seconds. Vertical is native rather than a crop, so hooks for paid social come out the right shape. It also accepts up to seven reference images, which drops to five when you are editing or continuing an existing video.

In EzUGC both run at 720p and both generate their own audio, so the choice is not about resolution. Veo 3.1 gives you dual-frame control, meaning you can pin the last frame as well as the first, and it runs at a fixed 8 seconds. Omni covers 3 to 10 seconds, takes up to seven reference images, and adds two things Veo does not do here: editing a source video and continuing one of its own clips. Reach for Veo 3.1 when you need a shot to land on a specific closing frame, and Omni when you expect to iterate on the take.