Real outputs from the public examples of this model on Replicate, with the prompt and settings that produced them. They were not generated live on this page.
Aerial drone shot slowly rising above a misty old-growth forest at dawn, revealing a winding river cutting through the valley below, golden sunlight breaking through clouds
Close-up of a woman's face as she opens her eyes and looks into the camera, shallow depth of field, soft golden backlight, hair gently moving in the breeze, cinematic 24fps
Google DeepMind is the merged AI research organisation formed in April 2023 when Google Brain and DeepMind (acquired by Google in 2014) were combined into a single unit led by Demis Hassabis (CEO). DeepMind was founded in 2010 by Hassabis, Shane Legg and Mustafa Suleyman, and is famous for AlphaGo, AlphaFold and the Gemini language-model family. On the video side, DeepMind shipped Imagen Video (2022), Lumiere (2024), Veo (May 2024, Google I/O), Veo 2 (December 2024), Veo 3 with native audio (May 2025) and Veo 3.1 (late 2025). Veo is exposed via Vertex AI, the Gemini API, Google Labs (VideoFX, Whisk) and YouTube's Dream Screen.
Latent video diffusion / Diffusion Transformer hybrid with cascaded super-resolution
Veo 2 is a closed text-to-video and image-to-video model that builds on DeepMind's Imagen and Lumiere research. It uses a spatio-temporal latent diffusion architecture with a transformer-based denoiser conditioned on rich text embeddings from Gemini-family encoders and optional image embeddings. Veo 2 generates clips up to 8 seconds at up to 4K resolution (extended versions reach minutes), with strong understanding of camera language ('dolly', 'aerial', '35mm anamorphic') and real-world physics. DeepMind report substantial gains over Veo 1 on motion realism, prompt adherence and detail. The pipeline combines a base diffusion model with cascaded spatial and temporal super-resolution and an upscaler. Veo 2 was trained on a curated YouTube-and-partner video corpus with synthetic captioning at multiple granularities, and post-trained with reward models for prompt adherence and aesthetic quality.
Rich cinematographic prompt vocabulary (lenses, lighting, camera moves)
Strong real-world physics and object permanence
Available via Vertex AI, Gemini API, VideoFX and YouTube Dream Screen
SynthID watermarking embedded in every frame
Multilingual prompts via Gemini text encoders
Lower hallucination of artefacts than Veo 1
Best for: high-quality marketing, film previs, broadcast inserts, enterprise creative.
Entraînement et licence
Curated mix of licensed video, public web video (including YouTube under Google's terms) and synthetic data, with multi-granularity captioning produced by Gemini vision models. Exact dataset size undisclosed.
Licence: Proprietary commercial licence via Google Cloud / Vertex AI and the Gemini API; commercial usage permitted under Google's generative AI terms; SynthID watermark required on all output.
curl https://railwail.com/api/v1/videos/generations \
-H "Authorization: Bearer $RAILWAIL_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "google-veo-2",
"prompt": "A slow drone shot over a misty pine forest at sunrise"
}'
# Response: {"job_id": "...", "status": "queued", ...}
# Poll until status is completed, failed or cancelled:
curl https://railwail.com/api/v1/jobs/JOB_ID \
-H "Authorization: Bearer $RAILWAIL_API_KEY"
import os
import time
import requests
API = "https://railwail.com/api/v1"
headers = {"Authorization": f"Bearer {os.environ['RAILWAIL_API_KEY']}"}
job = requests.post(
f"{API}/videos/generations",
headers=headers,
json={
"model": "google-veo-2",
"prompt": "A slow drone shot over a misty pine forest at sunrise",
},
).json()
while True:
status = requests.get(f"{API}/jobs/{job['job_id']}", headers=headers).json()
if status["status"] in ("completed", "failed", "cancelled"):
break
time.sleep(5)
print(status["status"], status.get("output_url"))
const API = "https://railwail.com/api/v1";
const headers = {
Authorization: `Bearer ${process.env.RAILWAIL_API_KEY}`,
"Content-Type": "application/json",
};
const job = await fetch(`${API}/videos/generations`, {
method: "POST",
headers,
body: JSON.stringify({
model: "google-veo-2",
prompt: "A slow drone shot over a misty pine forest at sunrise"
}),
}).then((r) => r.json());
let status;
do {
await new Promise((r) => setTimeout(r, 5000));
status = await fetch(`${API}/jobs/${job.job_id}`, { headers }).then((r) => r.json());
} while (!["completed", "failed", "cancelled"].includes(status.status));
console.log(status.status, status.output_url);
Proprietary commercial licence via Google Cloud / Vertex AI and the Gemini API; commercial usage permitted under Google's generative AI terms; SynthID watermark required on all output.
Kuaishou's Kling v2.1, generating 5 and 10 second videos at 720p or 1080p from text or an image. Known for cinematic camera work and realistic physical motion, available on Replicate via the official KwaiVGI account.
