Microscope-style science clip with Gemini Omni Flash: prompt and cost

To make a microscope-style science clip with Gemini Omni Flash 1.1, write the prompt like a lab caption: the subject, the colors, the lighting, the imaging style, and one sentence telling the model to keep the lens effect for the whole clip. On Sume, send it as plain text-to-video to gemini-omni-flash-1.1 for 3 to 10 seconds. An 8-second draft at 360p costs $0.30, and the same clip at 720p costs $1.00.

Google's launch post uses exactly this genre as its 360p example, and the prompt is worth copying because of how it is built.

The prompt structure in Google's example

The launch post shows a prompt about iridescent marine diatoms with glass-like silica shells, a palette from volcanic amber and warm copper to turquoise and violet, and tiny structures glowing against a clean dark field. It ends with style terms ("high-fidelity scientific imaging, sharp details, organic textures, micro-photography") and then the line "Maintain the microscope lens effect throughout the entire video."

• Subject first, with one physical detail (shell structure, symmetry).
• Palette as a range, not a list of ten colors.
• Lighting and background in one clause (glow against a dark field).
• Style vocabulary last.
• A persistence sentence, so the lens look does not drift into a normal camera look halfway through.

Draft at 360p, finish higher

Google says 360p previews generate up to 60 percent faster and at a third of the cost of 720p. Sume's rates keep that ratio: $0.0375 per second at 360p against $0.125 at 720p. Run three or four prompt variants at 360p, pick one, and only then pay for 720p or 1080p.

Cost of a three-draft, one-final pass

For an 8-second clip, three 360p drafts are 3 x $0.30 = $0.90, and one 1080p final is $1.50, so the pass costs $2.40. If you finish at 720p instead, the total is $1.90. Native audio is always on for this model, and Sume rejects generate_audio: false, so a silent microscope clip needs the audio dropped afterwards or a narration bed placed over it in a timeline.

What to be careful about

A prompt like this produces a plausible image of a microscopic scene, not a measurement. If the clip goes into teaching material, label it as an illustration and do not caption it with species names or scale bars the model invented. Describe what the viewer should see, and check each finished clip against the species or structure you meant.

Keep the prompt well under the 20,000-character limit; the Google example is a few sentences, and short, concrete prompts are easier to adjust one phrase at a time between drafts. Change a single phrase per draft, such as the palette or the lighting, so you can tell which edit caused which difference.

Camera behavior is the other lever. A slow push-in or a drifting focus pull fits the genre; for ideas on wording, see camera prompt words for Omni. For the draft pass in detail, see the 360p draft cost on Sume.