Gemini Omni Flash Reference to Video API

google/gemini-omni-flash/reference-to-video

Gemini Omni Flash Reference to Video builds 4–10 second clips from exactly three reference images plus a text prompt, with locked character consistency, native lip-synced audio, and resolution from 720p to 4K. It carries face, wardrobe, and style cues across the new scene while adding directed motion, camera language, and synchronized sound.

Input

Prompt

441/20000

Reference images

Drag and drop files, paste from clipboard, or provide a URL.

3/3 files added

Resolution

Duration

Aspect ratio

Output

Ready

Continue with

Gemini Omni Flash Text to Video Gemini Omni Flash Image to Video Gemini Omni Flash Video Edit

Examples

Use the reference images only for this astronaut's face, short red hair, and white-and-orange flight suit. New scene: she steps into a vast alien crystal cave, giant luminous crystals refracting teal and violet light, she turns her head looking around in wonder. Camera follows behind her in one steady tracking shot. Audio: echoing footsteps and a low crystalline hum. Keep her identity consistent. No logos, no readable text, no watermark.

Use the reference images only for this border collie's markings, blue eye, and identity. New scene: the dog herds a small flock of sheep across a misty hillside, trotting and turning to guide them, fog drifting over green grass. Documentary-style steady wide shot. Audio: soft wind, distant sheep bells. Keep the dog's appearance consistent. No logos, no readable text, no watermark.

Use the reference images only for this white ceramic fox figurine with gold accents, its shapes and colors. New scene: the figurine comes alive in a snowy pine forest, it shakes snow off its back, blinks, and steps carefully between the trees, gentle snowfall. Magical realism, steady shot. Audio: soft snow ambience and tiny footsteps. Keep its design consistent. No logos, no readable text, no watermark.

REST API

Quick Start

Authenticate with the API, submit three references and a prompt, then retrieve the video using the task ID.

Connect to the Vidgo API

Create an API key, keep it only on your server, and send Authorization: Bearer VIDGO_API_KEY.

Endpoint: POSThttps://api.vidgo.ai/api/generate/submit
Authentication: Authorization: Bearer VIDGO_API_KEY

Submit one generation task

Fill in the parameters for this endpoint using the request example, then save the returned task_id to query progress and results.

Submit request

REQUEST_BODY=$(cat <<'JSON'
{
  "model": "google/gemini-omni-flash/reference-to-video",
  "input": {
    "prompt": "Use the reference images only for this astronaut's face, short red hair, and white-and-orange flight suit. New scene: she steps into a vast alien crystal cave, giant luminous crystals refracting teal and violet light, she turns her head looking around in wonder. Camera follows behind her in one steady tracking shot. Audio: echoing footsteps and a low crystalline hum. Keep her identity consistent. No logos, no readable text, no watermark.",
    "duration": 8,
    "resolution": "720p",
    "aspect_ratio": "16:9",
    "image_urls": [
      "https://cdn.vidgo.ai/apis/models/google/gemini-omni-flash/reference-to-video/v1/01/input-01.png",
      "https://cdn.vidgo.ai/apis/models/google/gemini-omni-flash/reference-to-video/v1/01/input-02.png",
      "https://cdn.vidgo.ai/apis/models/google/gemini-omni-flash/reference-to-video/v1/01/input-03.png"
    ]
  }
}
JSON
)

RESPONSE=$(curl --silent --show-error --fail-with-body \
  --request POST \
  --url "https://api.vidgo.ai/api/generate/submit" \
  --header "Authorization: Bearer $VIDGO_API_KEY" \
  --header "Content-Type: application/json" \
  --data "$REQUEST_BODY")

CODE=$(printf '%s' "$RESPONSE" | jq -r '.code // empty')
if [ "$CODE" != "0" ] && [ "$CODE" != "200" ]; then
  printf 'API error: %s
' "$RESPONSE" >&2
  exit 1
fi
printf '%s
' "$RESPONSE"

Wait for the result

Query with task_id, continue for not_started/running, and stop for finished/failed. On success, read data.files[].file_url.

Track status

GET https://api.vidgo.ai/api/generate/status/{task_id}

Poll status with a 2-second base interval, and increase the interval for longer tasks. Continue only while status is not_started or running, and stop once finished or failed. You can also specify callback_url in the request payload to receive webhook notifications.

not_startedrunningfinishedfailed

Submit response

{
  "code": 200,
  "data": {
    "task_id": "task-unified-...",
    "status": "running",
    "created_time": "2026-09-15T10:00:00Z"
  }
}

Finished response

{
  "code": 200,
  "data": {
    "task_id": "task-unified-...",
    "status": "finished",
    "files": [
      {
        "file_url": "https://storage.vidgo.ai/generated/video.mp4",
        "file_type": "video"
      }
    ],
    "created_time": "2026-08-22T10:00:00Z",
    "progress": 100,
    "error_message": null
  }
}

Complete runnable example

Expand for a complete script with HTTP and business-code checks, task_id validation, polling, terminal-state handling, and a timeout boundary.

Complete runnable example

set -euo pipefail
: "${VIDGO_API_KEY:?Set VIDGO_API_KEY in your environment}"

REQUEST_BODY=$(cat <<'JSON'
{
  "model": "google/gemini-omni-flash/reference-to-video",
  "input": {
    "prompt": "Use the reference images only for this astronaut's face, short red hair, and white-and-orange flight suit. New scene: she steps into a vast alien crystal cave, giant luminous crystals refracting teal and violet light, she turns her head looking around in wonder. Camera follows behind her in one steady tracking shot. Audio: echoing footsteps and a low crystalline hum. Keep her identity consistent. No logos, no readable text, no watermark.",
    "duration": 8,
    "resolution": "720p",
    "aspect_ratio": "16:9",
    "image_urls": [
      "https://cdn.vidgo.ai/apis/models/google/gemini-omni-flash/reference-to-video/v1/01/input-01.png",
      "https://cdn.vidgo.ai/apis/models/google/gemini-omni-flash/reference-to-video/v1/01/input-02.png",
      "https://cdn.vidgo.ai/apis/models/google/gemini-omni-flash/reference-to-video/v1/01/input-03.png"
    ]
  }
}
JSON
)

SUBMIT_RESPONSE=$(curl --silent --show-error --fail-with-body \
  --request POST \
  --url "https://api.vidgo.ai/api/generate/submit" \
  --header "Authorization: Bearer $VIDGO_API_KEY" \
  --header "Content-Type: application/json" \
  --data "$REQUEST_BODY")

TASK_ID=$(printf '%s' "$SUBMIT_RESPONSE" | jq -r '.data.task_id // .task_id // empty')
BUSINESS_CODE=$(printf '%s' "$SUBMIT_RESPONSE" | jq -r '.code // empty')
if [ "$BUSINESS_CODE" != "0" ] && [ "$BUSINESS_CODE" != "200" ]; then
  printf 'Submit failed:
%s
' "$SUBMIT_RESPONSE" >&2
  exit 1
fi
if [ -z "$TASK_ID" ]; then
  printf 'Submit response did not include task_id:
%s
' "$SUBMIT_RESPONSE" >&2
  exit 1
fi

START_TIME=$(date +%s)
POLL_DELAY=2
while true; do
  if [ $(( $(date +%s) - START_TIME )) -ge 600 ]; then
    printf 'Timed out after 600 seconds
' >&2
    exit 1
  fi
  STATUS_RESPONSE=$(curl --silent --show-error --fail-with-body \
    --url "https://api.vidgo.ai/api/generate/status/$TASK_ID" \
    --header "Authorization: Bearer $VIDGO_API_KEY")
  STATUS=$(printf '%s' "$STATUS_RESPONSE" | jq -r '.data.status // .status // empty')
  BUSINESS_CODE=$(printf '%s' "$STATUS_RESPONSE" | jq -r '.code // empty')
  if [ "$BUSINESS_CODE" != "0" ] && [ "$BUSINESS_CODE" != "200" ]; then
    printf 'Status request failed:
%s
' "$STATUS_RESPONSE" >&2
    exit 1
  fi

  case "$STATUS" in
    finished)
      printf '%s' "$STATUS_RESPONSE" | jq -r '(.data.files // .files // [])[]?.file_url'
      break
      ;;
    failed)
      printf '%s' "$STATUS_RESPONSE" | jq -r '.data.error_message // .error_message // "Generation failed"' >&2
      exit 1
      ;;
    not_started|running)
      sleep "$POLL_DELAY"
      if [ "$POLL_DELAY" -lt 10 ]; then POLL_DELAY=$((POLL_DELAY + 1)); fi
      ;;
    *)
      printf 'Unexpected task status: %s
' "$STATUS" >&2
      exit 1
      ;;
  esac
done

Input parameters

The table lists available input parameters, types, and defaults. Request examples also include the required top-level model field.

Field	Type	Required	Default	Description
prompt	string	Yes	—	1–20,000 characters after trimming.
image_urls	string[]	Yes	—	Exactly 3 public HTTP(S) image URLs.
duration	integer	No	6	4, 6, 8, or 10 seconds.
resolution	string	No	720p	720p, 1080p, or 4k.
aspect_ratio	string	No	16:9	16:9 or 9:16.

Response Fields

A successful submission returns a task ID. Status queries provide progress, output files, and error details when a task fails.

Field	Type	Description
code	integer	Application result code; successful responses use 0 or 200.
message	string	Human-readable message or error detail when present.
data.task_id	string	Task ID used in the status endpoint path.
data.status	string	not_started, running, finished, or failed.
data.created_time	string	Task creation time in date-time format.
data.progress	integer	Task progress from 0 to 100, when included in the response.
data.files[]	array	All output files from a successful task, in response order.
data.files[].file_url	string	Public URL for a generated video.
data.files[].file_type	string	File type, such as video.
data.error_message	string \| null	Failure detail when status is failed.

Task Lifecycle

Continue querying while the status is not_started or running. End polling at finished or failed, then process the output files or error details respectively.

not_started
The task was accepted and is waiting to begin.
running
Generation is in progress. Continue polling the same task_id.
finished
Generation succeeded. Read every video URL from data.files[].file_url.
failed
Generation stopped with an error. Read data.error_message and stop polling.

Polling and Errors

AuthenticationFor a 401 response, check the Bearer API key in Authorization, update the credentials, and retry.
ValidationFor a 400 response, use the response details to check required fields, parameter ranges, and available credits, then adjust and resubmit.
Network and timeoutIf a status query encounters a network error or timeout, retain the original task_id and retry the query, then handle the result according to the returned task status.
Polling intervalPoll status with a 2-second base interval, and gradually increase the interval for longer tasks.
Terminal statesContinue only for not_started or running. Stop immediately on finished or failed.
Callback optionProvide callback_url at the request top level to receive the final flat task object; polling remains available if delivery fails.

Endpoint limits

Specification	Value	Details
Input mode	Prompt + 3 images	Exactly three public reference stills plus a scene and audio prompt.
Output	Video	Returns an asynchronous task ID; finished tasks include a video file with native audio.
Resolution	720p / 1080p / 4k	Default is 720p.
Duration	4 / 6 / 8 / 10 seconds	Default is 6 seconds.
Aspect ratio	16:9 / 9:16	Default is 16:9.
Billing basis	Per generation	720p/1080p: 4s=120, 6s=150, 8s=200, 10s=220 credits. 4k: 4s=250, 6s=300, 8s=350, 10s=450 credits.

View the Gemini Omni Flash Reference to Video API docs

Gemini Omni Flash Reference to Video

Gemini Omni Flash Reference to Video is Google DeepMind’s multimodal model for multi-image consistent video generation. Provide exactly three public image URLs—for example face, wardrobe, and product or style—and a cinematic prompt to output 4–10 second clips with lip-synced audio. Choose 720p, 1080p, or 4K with 16:9 or 9:16 framing—ideal for recurring characters, campaign talent continuity, product line consistency, and branded virtual-host series.

Why Choose This?

Three-image character lockUse exactly three references to anchor face, wardrobe, and style so recurring talent stays recognizable across shots.
Strong multi-subject consistencyName people and products in the prompt while the references supply visual ground truth for appearance continuity.
Native audio with lip syncGenerate dialogue and ambience with the picture for talking characters that match the locked look.
Cinematic staging on locked looksApply push-in, dolly, tracking, and lighting cues without rebuilding identity from text alone.
720p to 4K delivery tiersIterate at 720p or 1080p, then step to 4K when campaign finals need finer texture and lighting detail.
Flexible duration and framingPick 4–10 seconds and 16:9 or 9:16 for hooks, mid-length demos, and vertical social series.

Parameters

Parameter	Requirement	Description
prompt	Required	String. Scene, action, camera, lighting, and audio cues; 1–20,000 characters after trimming.
image_urls	Required	Array with exactly 3 public HTTP(S) image URLs for character, wardrobe, or style reference.
duration	Optional	Integer. Output length in seconds; the Playground preselects 6. Default`64810`
resolution	Optional	String. Output clarity tier; the Playground preselects 720p. Default`720p1080p4k`
aspect_ratio	Optional	String. Output framing; the Playground preselects 16:9. Default`16:99:16`

How to Use

Prepare three reference imagesGather clear public URLs for face, wardrobe, and product or style so identity cues cover appearance and branding.
Write the scene promptDescribe action, camera, lighting, and audio, and name which reference drives face, outfit, or product look.
Set output durationChoose 4, 6, 8, or 10 seconds (default 6) to match the narrative beat.
Select resolutionPick 720p for iteration, 1080p for clearer delivery, or 4K for high-detail finals.
Choose aspect ratioSelect 16:9 or 9:16 to match landscape storytelling or vertical social series.
Review the cost and runCheck the cost shown on the Run button, finish references and prompt, then click Run.
Preview and download the videoWhen the task finishes, preview picture and synced audio, then select Download video to save the result.

Pricing

Billed per generation by duration and resolution tier, with native audio included. 1 credit = $0.005.

Usage	Rate	Details
720p / 1080p	4s=120, 6s=150, 8s=200, 10s=220 credits	Default 720p / 6s costs 150 credits ($0.75).
4k	4s=250, 6s=300, 8s=350, 10s=450 credits	4k / 6s costs 300 credits ($1.50).

Best Use Cases

Recurring character seriesKeep the same talent look across episodic social shorts and campaign chapters.
Product line consistencyLock packaging and hero product appearance while staging new environments and camera moves.
Brand virtual-host contentCombine face and wardrobe references with dialogue cues for ongoing host performances.
Multi-shot campaign storyboardsPrototype story beats that must share one visual identity before full production.

Pro Tips

Assign roles to the three images in the prompt—face reference, wardrobe reference, product or style reference.
Keep reference lighting consistent and faces unobstructed so identity signals stay strong.
Reuse the same appearance wording across iterations when building a series of related clips.
Specify push-in, dolly, or tracking moves separately from character action to control staging.
Add Audio lines for dialogue language and ambience so lip sync matches the locked performer.

Usage notes

Gemini Omni Flash Reference to Video requires prompt plus image_urls with exactly three public image URLs.
Describe speech or ambience in the prompt; native audio with lip sync is included in the result.
Duration, resolution, and aspect_ratio configure output length, clarity, and framing.
After an API submission, save the returned task_id to query progress and retrieve the final media URL.

Related Models

PixVerse V6 Max Text to VideoPixVerse V6 Max Text to Video turns a prompt into 1–15 second videos at 360p to 1080p, with eight aspect ratios and an optional native audio track for dialogue, effects and music. Describe subjects, actions and sound in one prompt, then pick the length and format for each placement.

PixVerse V6 Max Image to VideoPixVerse V6 Max Image to Video animates one JPEG, PNG or WebP image into a 1–15 second video at 360p to 1080p, with an optional native audio track. The image sets the subject and framing, and a prompt of up to 2,048 characters directs motion, camera and sound.

OmniHuman 2.0 Audio to VideoOmniHuman 2.0 Audio to Video turns one character image and 4–300 seconds of speech or music into a 720p video, with lips, expression and gestures following the audio. The image keeps the character's look and framing, while an optional prompt adds camera moves and actions.

FLUX 3 Max Text to VideoFLUX 3 Max Text to Video turns a written shot into a 5–20 second 720p or 1080p video with native audio and eight aspect ratios. It follows the camera moves, lighting and sound cues you describe, so one prompt returns picture and audio together.

FLUX 3 Max Image to VideoFLUX 3 Max Image to Video animates a start image into a 5–20 second 720p or 1080p clip with native audio, optionally landing on an end frame you supply. It keeps the composition, subject and palette of your images while adding the motion and sound in the prompt.

FLUX 3 Max Keyframes to VideoFLUX 3 Max Keyframes to Video turns 1–10 images pinned to exact frames into a single 5–20 second 720p or 1080p clip with native audio. Each image appears at the frame_index you set on a 24 fps timeline, so the story beats land where you planned them.

Kling 4.0 Text to VideoKling 4.0 Text to Video turns a prompt or multi-shot script into one 3–30 second 720p or 1080p video with stereo audio and five aspect ratios. It keeps scenes coherent across long takes and shot changes while following your camera, dialogue and sound directions.

Kling 4.0 Image to VideoKling 4.0 Image to Video animates a start image into a 3–30 second 720p or 1080p video with stereo audio, optionally finishing on an end frame you choose. It keeps the subject, framing and look of your image while adding the motion, camera work and sound in your prompt.

Kling 4.0 Reference to VideoKling 4.0 Reference to Video turns reference images, videos and reusable elements into a 3–30 second 720p or 1080p video with stereo audio, cited as @Image1, @Video1 or @Element1. It keeps characters, products and settings consistent with your references inside the new scene you describe.

Kling 4.0 Video EditKling 4.0 Video Edit transforms an existing 3–30 second video from a written instruction, restyling scenes, swapping objects or adding elements at 720p or 1080p. It keeps the original timing, people and camera path that you ask to preserve while applying the change across every frame.

Vidu Q4 Image to VideoVidu Q4 Image to Video turns one start image and an optional motion prompt into a 3–16 second clip at 540p up to 4K. Your image sets the opening frame, so subject, setting and aspect ratio carry through while the prompt directs movement and camera.

Vidu Q4 Reference to VideoVidu Q4 Reference to Video combines a prompt with up to 12 reference images and 3 voice clips into 3–16 second videos at up to 4K. Tag each reference in the prompt to keep characters, products and places recognisable, with optional generated audio at no extra cost.

Gemini Omni 1.1 Flash Text-to-Video preview

Gemini Omni 1.1 Flash Text-to-VideoGemini Omni 1.1 Flash Text-to-Video. Create 4–10 second videos with native audio from text, directing camera, lighting, and sound.

Gemini Omni 1.1 Flash Image-to-Video preview

Gemini Omni 1.1 Flash Image-to-VideoGemini Omni 1.1 Flash Image-to-Video. Animate a start image with native audio and add an optional end frame to guide the closing shot.

Gemini Omni 1.1 Flash Reference-to-Video preview

Gemini Omni 1.1 Flash Reference-to-VideoGemini Omni 1.1 Flash Reference-to-Video. Use 1–7 images to guide characters, products, and style in new video scenes with sound.

Gemini Omni 1.1 Flash Video Edit preview

Gemini Omni 1.1 Flash Video EditGemini Omni 1.1 Flash Video Edit. Edit 3–10 second videos through natural language with up to 5 reference images for visual guidance.

Gemini Omni Flash Reference to Video API frequently asked questions

What is the Gemini Omni Flash Reference to Video API?

Gemini Omni Flash Reference to Video is a Google DeepMind multimodal model for generating video from multiple reference images. It creates 4–10 second clips from exactly three reference stills plus a text prompt, with locked character consistency, native lip-synced audio, and 720p to 4K output. Built on Gemini’s unified multimodal architecture, it carries face, wardrobe, and style cues into a new scene while adding directed motion and synchronized sound. You can call it programmatically or try it from the playground above.

How many reference images does Gemini Omni Flash Reference to Video require?

Submit exactly three public image URLs in image_urls. Typical setups pair face, wardrobe, and product or style references; naming each role in the prompt strengthens consistency.

How does Gemini Omni Flash Reference to Video keep character consistency?

The three stills supply visual ground truth for appearance, while the prompt restates face, outfit, and branding details. Reuse the same reference set and appearance wording when generating related clips in a series.

Does Gemini Omni Flash Reference to Video support native lip sync?

Yes. Dialogue and ambience are generated with the picture and embedded in the MP4. Add language and tone cues so the locked performer speaks in sync with mouth motion.

When should I choose Gemini Omni Flash Reference to Video over Image to Video?

Use Reference to Video when one still is not enough and you need multi-image locks for face, wardrobe, and style. Use Image to Video when a single key visual already carries the full look you want to animate.

Can Gemini Omni Flash Reference to Video output 4K?

Yes. Choose 4K for campaign finals that need sharper texture and lighting. Draft at 720p or 1080p first to validate consistency, then promote selected takes.

How are Gemini Omni Flash Reference to Video credits calculated?

Credits are charged per generation by duration and resolution. At 720p/1080p, 6 seconds costs 150 credits ($0.75); at 4K the same length costs 300 credits ($1.50). See the Pricing section for full rates.