This ComfyUI workflow turns two still images into a short video using Google’s Gemini Omni Flash via the GeminiVideoOmni node. You provide a start image and an end image (two LoadImage nodes), then guide the motion, duration, and framing with a natural-language prompt. A built-in Prompt Constructor subgraph (custom node: b83478ba-1f3e-4e62-9869-c8a0c8e9a5f5) helps you explicitly include the required duration and aspect ratio in the final prompt string—both are mandatory for Gemini Omni Video. The GeminiVideoOmni node then generates a single video clip that you can preview with PreviewAny and export with SaveVideo.

Technically, the workflow wires your two image tensors and the constructed text prompt into GeminiVideoOmni, which requests a video from the Gemini Omni Flash model. Duration and aspect ratio are parsed from the prompt (for example: "duration: 4s, aspect_ratio: 16:9"), while the rest of your text describes the motion (e.g., zooms, pans, morphs, reveals). The MarkdownNote nodes embedded in the graph provide quick reminders and tips as you work. The result is a fast, conversational way to iterate on short brand clips, vertical social posts, and before/after reveals—without manual keyframing.