MiniMax H3 Singularity Reference-to-Video LoRA
minimax-h3-singularity/reference-to-video-lora
Generate videos with native stereo audio, 3–15 second durations and four resolution tiers. Supports custom LoRAs.
Examples
Parameters
| Name | Type | Default | Constraints | Description |
|---|---|---|---|---|
| prompt *Prompt | textarea | — | — | Text description of the desired video. Refer to reference inputs as <Picture 1>..<Picture 9>, <Video 1>..<Video 3>, and <Audio 1>..<Audio 3>. Audio is generated natively together with the video. |
| reference_imagesReference Images | image_upload | — | image/* · 0–9 items | Reference image URLs (up to 9). At least one reference input is required. |
| reference_audiosReference Audios | audio_upload_group | — | audio/* · 0–3 items | Standalone reference audio URLs (up to 3, trimmed to 15 seconds each). |
| aspect_ratioAspect Ratio | select | 16:9 | 16:9 | 9:16 | 1:1 | 4:3 | 3:4 | 21:9 | 9:21 | Output aspect ratio. |
| resolutionResolution | select | 480p | 480p | 540p | 768p | 1080p | Output video resolution. 768p is the model's native canvas; 480p is a faster, lower-cost tier; 540p is a mid tier at 1.5x the 480p price. 1080p is the highest-quality full-HD tier at 2x the 768p price (generation takes longer). |
| durationDuration | select | 5 | 3 ~ 15 · step 1 · 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | … | Output video duration in seconds. |
| lorasLoras | array<object> | — | 0–3 items | List of LoRAs to apply (max 3). |
| ↳path *Path | text | — | — | URL or the path to the LoRA weights. |
| ↳scaleScale | slider | 1 | 0 ~ 4 · step 0.1 | The scale of the LoRA weights. This is used to scale the LoRA weight before merging it with the base model. |
| seedSeed | number | — | step 1 | The random seed to use for the generation. -1 means a random seed will be used. |
Output fields
| Field | Type | Description |
|---|---|---|
| videos | array<string> | Generated video URLs with native audio. |
API
Call this model through one unified REST API. Get a key on the API Keys page.
cURL
# 1) Submit — returns { "task_uuid": "..." }
curl -X POST "https://www.namifusion.com/api/v1/marketplace/run/minimax-h3-singularity/reference-to-video-lora" \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"input": {
"prompt": "A cinematic portrait, dramatic rim light, shallow depth of field",
"aspect_ratio": "16:9",
"resolution": "480p",
"duration": 5
}
}'
# 2) Poll until status is "completed", then read the output URLs
curl "https://www.namifusion.com/api/v1/marketplace/run/tasks/TASK_UUID" \
-H "Authorization: Bearer YOUR_API_KEY"Python
import time, requests
API_KEY = "YOUR_API_KEY"
HEADERS = {"Authorization": f"Bearer {API_KEY}"}
# 1) Submit
resp = requests.post(
"https://www.namifusion.com/api/v1/marketplace/run/minimax-h3-singularity/reference-to-video-lora",
headers=HEADERS,
json={
"input": {
"prompt": "A cinematic portrait, dramatic rim light, shallow depth of field",
"aspect_ratio": "16:9",
"resolution": "480p",
"duration": 5
}
},
)
resp.raise_for_status() # 401/402/429/5xx stop here instead of polling a bad task
task = resp.json()
# 2) Poll until a terminal state (completed / failed / cancelled).
# This model is allowed up to 1800s server-side.
deadline = time.time() + 1860
while task.get("status") not in ("completed", "failed", "cancelled"):
if time.time() > deadline:
raise TimeoutError(f"still {task.get('status')} — keep the task_uuid and poll later")
time.sleep(3)
poll = requests.get(f"https://www.namifusion.com/api/v1/marketplace/run/tasks/{task['task_uuid']}", headers=HEADERS)
poll.raise_for_status()
task = poll.json()
print(task["status"], task.get("output"))JavaScript
const API_KEY = "YOUR_API_KEY";
const HEADERS = { Authorization: `Bearer ${API_KEY}` };
// 1) Submit
const resp = await fetch("https://www.namifusion.com/api/v1/marketplace/run/minimax-h3-singularity/reference-to-video-lora", {
method: "POST",
headers: { ...HEADERS, "Content-Type": "application/json" },
body: JSON.stringify({
"input": {
"prompt": "A cinematic portrait, dramatic rim light, shallow depth of field",
"aspect_ratio": "16:9",
"resolution": "480p",
"duration": 5
}
}),
});
if (!resp.ok) throw new Error(`submit failed: ${resp.status} ${await resp.text()}`);
let task = await resp.json();
// 2) Poll until a terminal state (completed / failed / cancelled).
// This model is allowed up to 1800s server-side.
const deadline = Date.now() + 1860 * 1000;
while (!["completed", "failed", "cancelled"].includes(task.status)) {
if (Date.now() > deadline) throw new Error(`still ${task.status} — keep the task_uuid and poll later`);
await new Promise((r) => setTimeout(r, 3000));
const poll = await fetch(`https://www.namifusion.com/api/v1/marketplace/run/tasks/${task.task_uuid}`, { headers: HEADERS });
if (!poll.ok) throw new Error(`poll failed: ${poll.status}`);
task = await poll.json();
}
console.log(task.status, task.output);Documentation
MiniMax H3 Singularity Reference-to-Video LoRA
Create a video with native stereo audio from reference media.
Inputs
| Parameter | Required | Type / limits / default |
|---|---|---|
prompt | Yes | Motion, scene and sound instructions. |
reference_images | No | array; at most 9 image URLs. |
reference_videos | No | array; at most 3 video URLs; combined reference duration capped at 15 seconds; soundtracks are used. |
reference_audios | No | array; at most 3 audio URLs; each audio is trimmed to 15 seconds. |
aspect_ratio | No | string; 16:9 / 9:16 / 1:1 / 4:3 / 3:4 / 21:9 / 9:21; default: 16:9 |
resolution | No | string; 480p / 540p / 768p / 1080p; default: 480p |
duration | No | integer; 3–15; default: 5 |
loras | No | array; at most 3; path is required, scale is optional (0–4, default 1). LoRA loading adds no separate fee. |
seed | No | integer; -1 = random |
Provide at least one visual reference (image or video); audio is additional guidance. Use <Picture 1>, <Video 1> and <Audio 1> tags in the prompt; numbering follows each array.
Listing price
| Resolution | USD / second | Credits / second |
|---|---|---|
| 480p | $0.06 | 6 |
| 540p | $0.09 | 9 |
| 768p | $0.12 | 12 |
| 1080p | $0.24 | 24 |
USD list prices before account discounts; 1 USD = 100 credits. Reference videos use the same per-second rate as output video. Add $0.02 (2 credits) per reference image and per standalone audio. Sum all reference-video durations, cap at 15 seconds, then round up to whole seconds.
Charge = rate × (output seconds + ceil(min(total reference-video seconds, 15))) + $0.02 × (image count + standalone audio count).
Nami rounds the final credit amount upward to 0.01 credit. Native generated audio has no separate charge.
API example
Replace the example media URL with your uploaded asset. Submit once, then poll the returned task_uuid; do not automatically retry submission.
POST https://www.namifusion.com/api/v1/marketplace/run/minimax-h3-singularity/reference-to-video-lora
Authorization: Bearer YOUR_NAMIFUSION_API_KEY
Content-Type: application/json
{
"input": {
"prompt": "Animate the subject from <Picture 1> with a slow camera move and ambient sound.",
"duration": 5,
"resolution": "480p",
"reference_images": [
"https://example.com/subject.png"
]
}
}
GET https://www.namifusion.com/api/v1/marketplace/run/tasks/{task_uuid}
Authorization: Bearer YOUR_NAMIFUSION_API_KEY
Successful submission returns task_uuid and status. Query the task until completed or failed; successful output.videos contains the generated video URLs.
Related models
xAI Grok Imagine Video v1.5 Reference to Video
Generate a 1-15 second video from a prompt and 1-7 reference images at 480p or 720p.
Vidu Q4 Preview Reference to Video
Generate a 3–16 second video from a prompt with optional references: up to 15 images and 3 audio clips. Supports aspect ratio selection, audio generation and 540p–4k resolution.
MiniMax H3 Reference to Video
MiniMax H3 Reference to Video generates coherent 2K videos from natural-language prompts and multimodal references, including images, videos, and audio, guiding subject consistency, motion, timing, visual style, and scene continuity. Ready-to-use REST inference API, best performance, no coldstarts, affordable pricing.
MiniMax H3 Singularity Reference-to-Video
Generate videos with native stereo audio, 3–15 second durations and four resolution tiers.
Gemini Omni Flash Reference to Video API
Gemini Omni Flash Reference to Video creates short AI videos with synchronized audio from one or more reference images and a text prompt, preserving visual identity and following the provided references for guided multimodal video generation. Ready-to-use REST inference API, best performance, no coldstarts, affordable pricing.
Seedance 2.0 Mini Reference to Video
Seedance 2.0 Mini Reference-to-Video is a high-speed, cost-effective video generation model developed by ByteDance, engineered specifically for creating multi-shot videos with a cinematic feel. By seamlessly combining reference images with text prompts, it generates highly narrative video segments. The model features advanced AI camera control and ensures robust character consistency across different scenes. It supports multiple output resolutions from 480P up to 4K, durations ranging from 4 to 15 seconds, and flexible aspect ratios. Complete with an out-of-the-box REST inference API, it delivers exceptional performance with zero cold starts, offering an affordable yet premium solution for professional video creation.
Seedance 2.5 Reference to Video
The "Reference-to-Video" feature of Seedance 2.5 is the ultimate solution for visual stylistic unity. It precisely extracts artistic styles, lighting tones, or compositional intents from reference materials and seamlessly integrates them into newly generated videos, ensuring a highly consistent visual language for your creative series.
Seedance 2.0 Fast reference-to-video
Seedance 2.0 Fast (Video-Edit) edits an input video from a natural-language prompt at a faster, cheaper tier. Built on ByteDance Seed's unified multimodal architecture, it preserves subject identity, composition, and motion while rewriting lighting, style, weather, environment, or specific elements as instructed. Ready-to-use REST API, best performance, no coldstarts, affordable pricing.