Come scrivere prompt per Gemini Omni: guida pratica con esempi

I migliori prompt per Gemini Omni sembrano un brief di produzione compatto: obiettivo, riferimenti, soggetto, movimento, camera e vincoli.

Per imparare a promptare Gemini Omni, la prima cosa da abbandonare è l'idea di una frase magica.

Gemini Omni è utile perché nasce per la creazione multimodale video-first. Il prompt può definire lo scopo del clip, spiegare cosa preservare da immagini o video di riferimento, controllare il movimento camera, impostare il mood e guidare revisioni successive in linguaggio naturale.

Prompt debole:

Make a cinematic video of my product.

Prompt più forte:

Create a 6-second horizontal product reveal video for the uploaded image.
Use the product shape, color, and front-facing angle from the reference.
Start with the product in soft shadow, then use a slow push-in as warm studio light reveals the surface texture.
Style: premium editorial product film, clean background, realistic reflections.
Keep the logo position unchanged. Avoid warped text, extra labels, sci-fi effects, and camera shake.

Il secondo prompt dà a Gemini Omni le sei informazioni principali: risultato, guida sugli input, soggetto, movimento, camera e vincoli.

Questa guida mostra come scrivere prompt specifici ma non gonfi. Include formule, template riutilizzabili, prompt di editing, consigli sui riferimenti ed esempi per video prodotto, social clip, storyboard, demo UI e modifiche video.

Risposta rapida

Scrivi il prompt come un brief video:

1. Dichiara il risultato e lo scopo del clip.
2. Spiega come usare ogni immagine, video, audio o testo.
3. Blocca il soggetto e ciò che deve restare coerente.
4. Descrivi scena, luce, mood e stile.
5. Dirigi il movimento dal primo all'ultimo frame.
6. Dirigi la camera: angolo, lente, movimento, tipo di inquadratura e ritmo.
7. Aggiungi durata, formato, dettagli da preservare e problemi da evitare.
8. Itera con prompt successivi stretti invece di riscrivere tutto.

Un buon prompt è chiaro su ciò che deve accadere sullo schermo.

Perché il prompting di Gemini Omni è diverso

L'output non è testo, ma media temporale. Devi descrivere primo frame, cambiamento durante il clip, comportamento della camera e coerenza.

Google posiziona Gemini Omni per creare e modificare da input diversi, partendo dal video. Tre abitudini contano:

• Dare a ogni riferimento un compito.
• Descrivere movimento, non solo aspetto.
• Revisionare con modifiche mirate.

Non serve solo un prompt più lungo: serve pensare come un regista.

Formula prompt di Gemini Omni

Un prompt forte divide il brief in parti controllabili.

Formula:

Create a [duration] [aspect ratio] video for [purpose].
Use [reference input] to preserve [specific details].
The subject is [who or what].
The scene is [setting, lighting, mood, style].
During the clip, [action or transformation].
Camera: [shot type, angle, movement, pacing].
Keep [details that must remain consistent].
Avoid [specific failure modes].

Esempio completo:

Create an 8-second vertical video for a TikTok product teaser.
Use the uploaded product image to preserve the bottle shape, label placement, cap color, and matte texture.
The subject is a minimalist skincare bottle on a wet stone surface.
The scene is a clean bathroom counter with soft morning light and subtle steam in the background.
During the clip, water droplets form on the bottle while the camera slowly pushes in from a medium shot to a close-up.
Camera: stable macro-product style, shallow depth of field, no handheld shake.
Keep the label layout unchanged.
Avoid misspelled text, extra logos, distorted packaging, floating objects, and neon sci-fi effects.

Funziona perché ogni frase ha una funzione di produzione.

Parti dal lavoro, non dallo stile

Molti iniziano con parole di stile:

Cinematic, high quality, ultra realistic, viral, beautiful...

Non sono inutili, ma non sono un brief. Parti dall'azione:

Create a 6-second video ad showing how a messy desk becomes a calm productivity setup.

Poi aggiungi stile:

Style: bright editorial lifestyle video, realistic lighting, clean modern workspace, calm morning mood.

“Cinematic” è mood; il desk che cambia è un evento. Gemini Omni ha bisogno prima dell'evento.

Spiega l'uso di ogni riferimento

Se carichi prodotto, logo, storyboard e moodboard, assegna priorità:

Use Image 1 for product identity: preserve the shape, color, logo position, and front-facing angle.
Use Image 2 for visual style only: match the lighting and background mood, but do not copy the objects.
Use Video 1 for motion reference: copy the slow camera push-in and pacing, not the subject.
Use the text brief for claims and audience context, but do not render the text inside the video.

Ogni riferimento deve rispondere: cosa deve preservare Gemini Omni?

Buoni target: identità, forma prodotto, posizione logo, palette, movimento camera, luce, layout, timing, mood, sequenza storyboard.

Vago:

Use these images as inspiration.

Meglio:

Use Image 1 for the exact product silhouette and label position. Use Image 2 only for the warm studio lighting and beige background.

Descrivi il movimento come timeline

Un prompt per immagine descrive composizione; un prompt Gemini Omni descrive tempo:

First frame: the product is centered on a clean desk beside a closed laptop.
Middle: the laptop opens and the screen softly lights the product edge.
Final frame: the camera lands on a close-up of the product with the desk blurred in the background.

Più chiaro di:

Show a cool product video with a laptop and nice lighting.

Per clip semplici:

Beat 1: establish the subject.
Beat 2: show the change.
Beat 3: hold the final result.

Per scene complesse:

Shot 1: wide establishing shot of the studio desk.
Shot 2: medium shot as the product rotates slightly.
Shot 3: close-up of the texture and label.
Shot 4: final hero frame with clean negative space on the right.

Più chiaro è l'ordine temporale, meno il modello deve inventare.

Istruzioni camera significative

Usa indicazioni che controllano l'attenzione: slow push-in, locked tripod shot, top-down flat lay, macro close-up, low-angle reveal, smooth tracking shot, over-the-shoulder view, wide establishing shot, shallow depth of field, handheld documentary feel.

Deboli: make it cinematic, great camera, professional angle, viral movement.

Camera: start with a locked medium shot, then use a slow push-in over 5 seconds until the product fills the center third of the frame. Cinematic but restrained, no fast cuts.

Così dai ritmo, framing e vincolo.

Vincoli negativi mirati

I prompt negativi devono essere specifici:

Avoid misspelled text, changing logo placement, extra fingers, warped product geometry, flickering labels, fast camera shake, and random sci-fi effects.

Prodotto:

Keep product shape, color, label layout, and logo position consistent. Avoid extra labels, fake claims, distorted packaging, and unreadable text.

Personaggio:

Preserve the person's face, hairstyle, clothing color, and body proportions. Avoid identity drift, extra limbs, plastic skin, and sudden outfit changes.

UI:

Keep the interface layout clean and readable. Avoid fake tiny text, broken icons, warped browser chrome, and random extra buttons.

Cinema:

Avoid overdone lens flares, shaky camera movement, extreme slow motion, neon cyberpunk lighting, and unrealistic physics.

Non elencare ogni errore possibile: blocca quelli che rovinerebbero questo clip.

Iterare senza ricominciare

Il buon prompting è iterativo: genera, controlla, modifica un punto, ricontrolla.

Il primo output è una direzione, non l'asset finale. Conserva ciò che funziona e cambia una cosa alla volta.

Keep the same subject and background. Make the camera movement 50% slower and remove the handheld shake.

Preserve the product shape and label position. Change only the lighting from dramatic side light to soft morning light.

Keep the first three seconds. Replace the final shot with a cleaner close-up where the object is centered and fully in frame.

Use the same scene and action, but make the motion more physically realistic. Reduce floating objects and keep the product touching the table.

Preserve the composition. Remove the fake text from the background and keep the interface panels visually clean.

Un buon prompt di revisione dice cosa mantenere, cosa cambiare e cosa evitare.

Template riutilizzabili

Product reveal

Create a [duration] [aspect ratio] product reveal video for [platform or use case].
Use the uploaded product image to preserve [shape, color, label position, material, angle].
First frame: [initial setup].
Middle: [reveal action or camera movement].
Final frame: [hero shot].
Style: [premium studio, editorial lifestyle, clean SaaS, documentary, etc.].
Camera: [locked shot, slow push-in, macro close-up, tracking shot].
Keep [brand/product details] consistent.
Avoid [distorted packaging, misspelled text, extra labels, unwanted style].

Esempio:

Create a 7-second horizontal product reveal video for a landing page hero.
Use the uploaded device render to preserve the exact screen shape, silver frame, and keyboard layout.
First frame: the laptop is closed on a clean white desk.
Middle: the laptop opens and the screen glows softly.
Final frame: a close-up of the screen with clean negative space around it.
Style: premium SaaS editorial, bright neutral lighting, realistic reflections.
Camera: stable slow push-in, no handheld shake.
Keep the device proportions and keyboard layout consistent.
Avoid fake readable UI text, extra logos, warped edges, and sci-fi glow.

Social video

Create a [duration] vertical video for [TikTok, Reels, Shorts, X, LinkedIn].
The hook is: [visual hook in the first second].
The subject is [person, product, object, scene].
Show [specific action or transformation].
Style: [creator-style, documentary, studio, educational, editorial].
Camera: [phone-style handheld, locked tripod, overhead, close-up].
Pacing: [fast, calm, single take, three quick cuts].
Avoid [failure modes].

Esempio:

Create an 8-second vertical video for a short-form productivity post.
The hook is a chaotic desk instantly becoming organized in the first second.
The subject is a notebook, laptop, coffee cup, and task cards on a real desk.
Show the clutter moving into a clean weekly planning layout.
Style: bright creator-style desk video, realistic morning light, satisfying but not magical.
Camera: top-down locked shot with subtle natural movement.
Pacing: three clean beats, no fast glitch transitions.
Avoid floating objects, warped text, fake brand logos, and unrealistic hands.

Storyboard

Create a [duration] video concept based on this shot list:
Shot 1: [wide scene].
Shot 2: [subject action].
Shot 3: [detail or reaction].
Shot 4: [final hero frame].
Style: [visual direction].
Camera: [movement and shot types].
Keep [continuity details].
Avoid [continuity failures].

Esempio:

Create a 12-second cinematic concept video based on this shot list:
Shot 1: wide shot of a small design studio at night, rain on the window.
Shot 2: over-the-shoulder shot of a designer dropping a rough sketch into an AI image tool.
Shot 3: close-up as the rough sketch becomes a polished campaign image.
Shot 4: final hero frame of the designer reviewing three clean variations.
Style: grounded editorial tech, warm desk light, realistic workspace, no sci-fi holograms.
Camera: slow controlled movement, shallow depth of field, no fast cuts.
Keep the same desk, laptop, and designer across all shots.
Avoid fake readable UI text, extra screens, random neon effects, and identity drift.

Editing video esistente

Edit the uploaded video.
Preserve [subject, timing, composition, identity, product details].
Change only [background, lighting, camera pacing, style, object, mood].
Keep the original [motion, face, product shape, scene layout] consistent.
Avoid [specific artifacts].

Esempio:

Edit the uploaded 6-second product clip.
Preserve the bottle shape, label placement, rotation timing, and camera angle.
Change only the background from a dark studio to a bright marble bathroom counter with soft morning light.
Keep the bottle physically grounded on the surface.
Avoid changing the logo, adding extra text, warping the cap, or introducing floating reflections.

Demo UI

Create a [duration] video demo for [app or feature].
Use the uploaded screenshot as the interface reference.
Show [specific interaction or transformation].
Camera: [screen recording style, slight push-in, over-the-shoulder, product mockup].
Keep the layout simple and readable.
Avoid fake detailed text, broken UI, extra buttons, distorted browser elements, and unreadable labels.

Esempio:

Create a 6-second horizontal video demo for an AI image editor.
Use the uploaded screenshot as the interface layout reference.
Show a rough product photo turning into three clean campaign variations inside the workspace.
Camera: screen-recording style with a very slight push-in, stable and readable.
Keep the panels, toolbar, and preview area simple.
Avoid fake tiny text, random extra buttons, distorted browser chrome, and sci-fi effects.

Esempi prima e dopo

Esempio 1: viaggio

Debole:

Make a beautiful travel video of the coast.

Meglio:

Create an 8-second horizontal travel video of a quiet coastal road at sunrise.
First frame: wide aerial view of the road along the cliffs.
Middle: the camera glides forward as a single car follows the curve of the road.
Final frame: the coastline opens into a warm sunrise over the ocean.
Style: natural documentary travel film, warm light, realistic waves, no exaggerated saturation.
Camera: smooth drone-style movement, slow and stable.
Avoid crowded traffic, fake landmarks, shaky motion, and overdone lens flare.

Perché funziona: soggetto, luogo, timeline, camera, mood e lista errori sono chiari.

Esempio 2: personaggio fashion

Debole:

Make this person walk in a cool city.

Meglio:

Create a 6-second vertical fashion video using the uploaded portrait as the identity reference.
Preserve the person's face shape, hairstyle, jacket color, and overall proportions.
Scene: quiet city street after rain, soft reflections on the pavement, early evening light.
Action: the person walks slowly toward camera for three steps, then turns slightly to the side.
Camera: chest-height tracking shot, stable, shallow depth of field.
Style: editorial streetwear film, realistic skin texture, natural motion.
Avoid identity drift, extra fingers, changing clothing color, plastic skin, and distorted background signs.

Perché funziona: separa identità preservata e nuova scena.

Esempio 3: lancio app

Debole:

Make a video of my app looking futuristic.

Meglio:

Create a 7-second horizontal launch video for an AI research app.
Use the uploaded screenshot for layout only: preserve the sidebar, search panel, and result cards.
Show a user query turning into three organized research summaries.
Style: clean professional SaaS demo, bright neutral background, restrained blue and green accents.
Camera: screen-recording style with a subtle push-in, no dramatic 3D rotation.
Keep the interface readable as a layout, but do not generate detailed fake text.
Avoid cyberpunk colors, holograms, extra buttons, warped browser chrome, and unreadable tiny labels.

Perché funziona: evita il caos da UI futuristica e imposta limiti utili.

Errori comuni

Errore 1: chiedere una vibe invece di un evento

“Make it cinematic” non basta. Di cosa succede.

Leave clean negative space for a headline. Do not render the headline inside the video.

Errore 2: riferimenti senza istruzioni

Un riferimento può significare identità, stile, luce, composizione, movimento o mood. Specifica il ruolo.

Errore 3: troppi lavori in un prompt

Non mescolare annuncio, personaggio, UI, logo e storia in un clip breve. Scegli il lavoro principale.

Errore 4: fidarsi del testo generato

Testo, logo, claim e label UI precisi vanno aggiunti dopo con strumenti deterministici.

Leave clean negative space for a headline. Do not render the headline inside the video.

Errore 5: cambiare tutto nel follow-up

Se un'inquadratura funziona, preservala:

Keep the composition and product position. Change only the camera speed and reduce the background clutter.

Errore 6: saltare la revisione finale

Controlla identity drift, mani deformate, prodotto incoerente, logo tremolante, UI rotta, fisica impossibile, testo indesiderato, marchi accidentali e movimento artificiale.

Workflow pratico

1. Scrivi un obiettivo creativo in una frase.
2. Raccogli riferimenti e segna cosa preservare.
3. Usa la formula.
4. Genera una versione.
5. Guarda due volte: feeling e dettagli.
6. Scrivi una revisione mirata.
7. Ripeti finché il problema principale è risolto.
8. Esporta dopo controlli su artefatti, claim e brand.

Se l'ultima risposta è no, riscrivi il brief.

Consigli avanzati

Usa “preserve” e “change only”:

Preserve the subject, composition, and lighting. Change only the background from a kitchen to a clean studio desk.

Chiedi meno tagli:

Use one continuous shot. No fast cuts. The camera should move slowly enough that the product remains readable.

Usa spazio negativo:

Leave clean negative space on the right side for a headline. Do not render text inside the video.

Separa stile e contenuto:

Use Image 2 for lighting and color mood only. Do not copy its objects, layout, or people.

Controlla il realismo:

Keep the product resting on the table throughout the clip. Reflections should follow the surface naturally. No floating, bending, or melting.

Usa la durata come vincolo: 5-8 secondi per un movimento, 10-15 per pochi shot, oltre serve storyboard.

Cosa non affidare solo a Gemini Omni

Attenzione a testi legali esatti, claim medici/finanziari/sicurezza, label prodotto pixel-perfect, logo perfetti, testo UI preciso, advertising regolato, tipografia finale e compliance di brand.

Usa Gemini Omni per esplorazione visiva, poi finalizza in editing, design o compliance.

Prompt finale copiabile

Create a [duration] [aspect ratio] video for [specific use case].
Use the uploaded references as follows:
- Image 1: preserve [identity/product/layout].
- Image 2: use only for [style/lighting/mood].
- Video 1: use only for [motion/pacing/camera reference].

The subject is [main subject].
The scene is [setting, lighting, mood].
First frame: [what we see at the start].
Middle: [what changes].
Final frame: [where the clip should land].

Camera: [shot type, movement, angle, pacing].
Style: [visual direction].
Keep [must-preserve details] consistent.
Avoid [specific artifacts, unwanted styles, text problems, and brand issues].

Per revisioni:

Keep [what worked] exactly the same.
Change only [one specific issue].
Make the result more [desired direction].
Avoid [the failure seen in the last version].

Verdetto finale

Il modo migliore per promptare Gemini Omni è pensare da regista: definisci lavoro, riferimenti, primo frame, movimento e finale; assegna un ruolo alla camera; aggiungi vincoli reali; itera con modifiche focalizzate.

FAQ

Struttura migliore?

Risultato, istruzioni sui riferimenti, soggetto, scena, movimento, camera, stile, regole di preservazione e lista da evitare.

Lunghezza?

Quanto basta per definire il lavoro. Di solito 120-250 parole per clip brevi; storyboard più lunghi.

Prompt negativi?

Sì, specifici: label che cambiano, identity drift, UI deformata, camera instabile o effetti sci-fi indesiderati.

Può usare immagini o video?

Google lo posiziona come modello multimodale per creazione e editing. Verifica Gemini, Flow o AI Studio prima della produzione.

Come preservare un personaggio?

Carica la migliore reference identitaria e specifica viso, capelli, vestiti, proporzioni ed espressione.

Per video prodotto?

Preserva forma, materiale, label, colore e angolo. Usa movimento semplice ed evita loghi extra, packaging distorto, claim falsi e testo illeggibile.

Testo o loghi esatti?

Con cautela. Meglio aggiungerli dopo con editor deterministici.

Errore principale?

Chiedere uno stile invece di un evento sullo schermo.

Fonti controllate

Articolo preparato con informazioni pubbliche controllate il 31 agosto 2026:

• Google DeepMind Gemini Omni model page: https://deepmind.google/models/gemini-omni/
• Google Gemini API Omni documentation: https://ai.google.dev/gemini-api/docs/omni
• Google AI for Developers changelog: https://ai.google.dev/gemini-api/docs/changelog
• Google Flow: https://labs.google/fx/tools/flow
• Google AI Studio: https://aistudio.google.com/
• Google AI for Developers pricing: https://ai.google.dev/gemini-api/docs/pricing