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 majestic lion drifts slowly upward through an enchanted forest, suspended weightlessly as golden-hour sunlight spills between towering trees, turning the morning mist into glowing ribbons of amber and gold. He looks very tiny compared to the size of the redwoods
PixVerse is the generative-video product of AISphere, a startup founded in 2023 by former senior researchers from ByteDance and Tencent. Headquartered between Singapore and Beijing, AISphere targets international short-form-video creators with a consumer-friendly web and mobile experience. PixVerse v1 shipped in early 2024, followed by v2, v3, v4 and v5 (mid-2025) and v5.6 (late 2025). The product emphasises stylised animation (anime, 3D pixar-style, paper craft, etc.), short-clip generation at 720p-1080p and integrated lip-sync and templates. AISphere is backed by Hillhouse, GGV Capital and IDG Capital, with a reported valuation of around $300M in 2025.
Latent video diffusion (DiT-style) with style adapters
PixVerse v5.6 is a closed latent-video-diffusion model with a transformer denoiser operating on a learned spatio-temporal latent space. The model supports text-to-video, image-to-video, character-reference mode and a library of pre-trained style adapters (anime, 3D animation, paper craft, claymation, comic, neon, etc.) that operate as fine-tuned LoRAs or auxiliary cross-attention heads on top of the base DiT. Conditioning uses a bilingual Chinese/English text encoder and image embeddings for first-frame conditioning. v5.6 generates 5-8 second clips at 720p-1080p / 24 fps with a lip-sync module that aligns mouth motion to a user-provided audio track. PixVerse has not released a formal technical paper; behaviour and feature surface suggest a multi-billion-parameter DiT trained on a curated multilingual video corpus with extensive synthetic stylised data.
curl https://railwail.com/api/v1/videos/generations \
-H "Authorization: Bearer $RAILWAIL_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "pixverse-v5",
"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": "pixverse-v5",
"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: "pixverse-v5",
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);
