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.
Prompt
a woman is walking through a busy Tokyo street at night, she is wearing dark sunglasses
Close-up of chocolate being poured over a stack of fresh strawberries in slow motion, the chocolate coating each berry as it cascades down, warm studio lighting
A person walking confidently through a colorful mural-covered alley, camera tracking alongside them, vibrant graffiti art on both walls, casual street fashion, golden hour
MiniMax (Shanghai Xiyu Technology) was founded in December 2021 by Yan Junjie, former SenseTime general manager, alongside several senior researchers from SenseTime and Microsoft Research Asia. It is one of China's 'four AI tigers' along with Zhipu, Moonshot and Baichuan, and has raised over $1B from investors including Tencent, Alibaba, HongShan and Mihoyo, with a $2.5B+ valuation in 2024. MiniMax developed the abab LLM family and the Hailuo AI consumer chat product. The 'Minimax Video' product (a.k.a. Hailuo Video on the consumer surface) generates short cinematic clips and was, at launch in September 2024, one of the first Chinese video AIs widely available to international users. This entry covers the developer-facing MiniMax Video API which exposes the underlying Hailuo Video models to third-party platforms.
Latent video diffusion (Diffusion Transformer) family
MiniMax Video is the API-tier branding for the Hailuo Video model family (currently Video-01, Video-01-Live, Video-01-Subject, Hailuo 02 and Hailuo 2.3). All variants share a closed latent-video-diffusion architecture: a 3D VAE compresses video into a spatio-temporal latent grid, and a transformer-based denoiser is conditioned on bilingual Chinese/English text embeddings (from a MiniMax abab-family LLM encoder) plus optional image embeddings. Different variants emphasise different conditioning modalities: text-only (Video-01), live2D-style anime animation (Video-01-Live) and subject-reference / identity locking (Video-01-Subject). Generation produces 6-10 second clips at 720p-1080p / 24-30 fps. MiniMax publish no model card; behaviour and reported throughput suggest a multi-billion-parameter DiT trained on a curated multilingual video corpus with synthetic dense captions and reward-model post-training.
curl https://railwail.com/api/v1/videos/generations \
-H "Authorization: Bearer $RAILWAIL_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "minimax-video",
"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": "minimax-video",
"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: "minimax-video",
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);
MiniMax Hailuo 02 on Replicate. Text-to-video and image-to-video producing 6s or 10s clips at 768p standard or 1080p pro. Known for accurate real-world physics and stable motion.
