POST /api/v1/inference/submit

This is the shared submit reference for all models. For a focused guide with complete examples, choose MiniMax H3 Turbo or MiniMax H3. H3 uses duration-based output settings; Wan, image and audio parameters below are model-specific.

API version: v1.2.4

Submit a video or image generation task to the queue. The endpoint detects video vs. image mode based on the model field.

Endpoint

POST /api/v1/inference/submit

Authentication

Required. Include your API key in the request header:

X-API-Key: your-api-key-here

Request Body

Video Generation Parameters

Resolution: Use resolution_tier + aspect_ratio (recommended) or explicit width + height.

• MiniMax H3 768p: 1344×768 (16:9), 1024×768 (4:3), 768×768 (1:1), 768×1024 (3:4), 768×1344 (9:16)
• MiniMax H3 480p: 832×480 (16:9), 640×480 (4:3), 480×480 (1:1), 480×640 (3:4), 480×832 (9:16)
• Wan 1080p: 1920×1080 (16:9), 1080×1920 (9:16), 1080×1080 (1:1)
• Wan 720p: 1280×720 (16:9), 720×1280 (9:16), 720×720 (1:1)
• Wan 512p: 960×512 (16:9), 512×960 (9:16), 512×512 (1:1)
• Wan 480p: 854×480 (16:9), 480×854 (9:16), 480×480 (1:1)

Image-to-Video Parameters

Wan Video Models

model defaults to wan/v2.2-a14b/text-to-video when omitted.

Turbo variants ignore num_inference_steps and guidance_scale and use fixed internal defaults. Standard mode caps 1080p at 81 frames; turbo allows 153.

LTX-Video Image-Audio-to-Video

Frame-based, driven by a still image plus an optional audio track.

Allowed sizes: 854×480, 1280×720, 1920×1080, 480×854, 720×1280, 1080×1920, 480×480, 720×720, 1080×1080.

Max frames by tier: 480p 481, 720p 241, 1080p 121. Frame counts are not subject to the Wan 1+4k alignment.

LTX Pricing (per frame): 480p $0.0008, 512p $0.0009, 720p and above $0.0012.

MiniMax H3 FL2VA Parameters

MiniMax H3 is exposed as text-to-video, image-to-video and reference-to-video:

Reference-to-Video input caps — the current Standard and Turbo limits below apply at both tiers. The image cap is higher when sending images only (no reference video or audio). Check the selected variant in GET /api/v1/pricing/config for current limits: minimax_h3_ref2va_limits or minimax_h3_ref2va_turbo_limits.

minimax/h3/reference-to-video and minimax/h3/reference-to-video/turbo

Reference video FPS must be 24 or lower; reference audio 10s or shorter each; reference image aspect ratio 4.0 or lower.

Reference input URLs must be accessible to the service. See upload requirements. Address references in the prompt with one-based tags such as <Picture 1>, <Video 1> and <Audio 1> that correspond to the submitted arrays. Audio references alone do not satisfy the required visual input.

Note this is a combined budget: hitting each per-type cap individually does not mean they can all be hit at once.

MiniMax H3 Pricing (10 credits = $1 USD):

Per output second, regardless of duration:

Reference-to-Video adds the reference inputs on top of the output:

total = duration x base_rate
      + max(0, reference_images - 5) x image_rate  # first 5 images free
      + reference_audios x $0                      # reference audio is free
      + reference_video_seconds x base_rate        # same per-second rate as the output tier

Example — reference-to-video, 5s at 768p with 2 reference images and 3 reference audios: 5 x $0.012 + 0 + 0 = 0.6 credits. 5s at 480p with 9 images only: 5 x $0.0045 + 4 x $0.0075 = 0.525 credits.

Live rates are published at GET /api/v1/pricing/config (reference_image_free_count carries the free allowance). Select minimax_h3_ref2va_pricing for Standard or minimax_h3_ref2va_turbo_pricing for Turbo; their reference-image fees differ. For the full current schedule, see pricing.

Custom LoRA Parameters

Standard (Non-Turbo) Mode Quality Parameters

Apply only to standard models (wan/v2.2-a14b/text-to-video, wan/v2.2-a14b/image-to-video). Turbo models (/turbo suffix) ignore these and use fixed internal defaults.

Safety Checker Parameters

Note: Both checkers are enabled by default. Disabling is intended for enterprise users with their own content moderation pipelines.

Content Policy Violation Response (400 Bad Request)

If input safety checker detects a violation, the request is rejected immediately, before any processing starts:

{
  "error": "content_policy_violation",
  "message": "Your prompt contains content that violates our content policy. Please revise your prompt and try again.",
  "reason": "nsfw_prompt"
}

For input image violations:

{
  "error": "content_policy_violation",
  "message": "The input image contains explicit content that violates our content policy. Please use a different image.",
  "reason": "explicit_content_nudenet"
}

Image Generation Parameters

Supported Image Models:

Allowed Image Resolutions: 512×512, 1024×1024, 1024×768, 768×1024, 1280×768, 768×1280

flux2/dev/text-to-image and flux2/dev/image-edit have their own set — 2048×1536, 1536×2048, 2048×1152, 1152×2048, 2048×2048, 1024×1024.

openai/gpt-image-2/text-to-image has its own set — 1024×1024, 1920×1072 (16:9 1080p), 1072×1920 (9:16 1080p), 2560×1440 (16:9 2K), 1440×2560 (9:16 2K). Note 1080p uses 1072, not 1080: width and height must both be multiples of 16 (1080 ÷ 16 = 67.5). It always returns exactly 1 image (num_images is ignored, as with FLUX.2-dev), and also ignores num_inference_steps, guidance_scale and output_format; the only parameters it accepts are prompt, model, width, height, and seed.

Any other size is rejected with 400 Invalid parameters.

Image Pricing (10 credits = $1 USD):

Flat per-image pricing — all resolutions same price. Price multiplied by num_images for supported multi-image models; FLUX.2-dev is fixed at 1 image per request.

Response

Success (200 OK)

{
  "request_id": "550e8400-e29b-41d4-a716-446655440000",
  "status": "queued",
  "message": "Task submitted successfully"
}

Error Responses

Insufficient Credits (402 Payment Required)

{
  "error": "Insufficient credits",
  "detail": "You need $0.05 credits to generate this video"
}

Invalid Request (400 Bad Request)

{
  "error": "Invalid request",
  "detail": "num_frames must be between 1 and 200"
}

Unauthorized (401 Unauthorized)

{
  "error": "API key required",
  "message": "X-API-Key header is required"
}

Examples

MiniMax H3 Text-to-Video

curl -X POST https://phosor.ai/api/v1/inference/submit \
  -H "X-API-Key: your-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "minimax/h3/text-to-video",
    "prompt": "A cinematic wide shot of a fantasy kingdom at sunrise, sweeping camera movement, 4K film look",
    "resolution_tier": "480p",
    "aspect_ratio": "16:9",
    "duration": 5,
    "seed": 42
  }'

MiniMax H3 Image-to-Video

curl -X POST https://phosor.ai/api/v1/inference/submit \
  -H "X-API-Key: your-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "minimax/h3/image-to-video",
    "prompt": "The character turns toward the camera as wind moves through the scene, cinematic lighting",
    "image_url": "images/img-abc123",
    "resolution_tier": "768p",
    "aspect_ratio": "16:9",
    "duration": 5
  }'

Reference-to-Video (MiniMax H3)

curl -X POST https://phosor.ai/api/v1/inference/submit \
  -H "X-API-Key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "minimax/h3/reference-to-video",
    "prompt": "Use <Picture 1> and <Picture 2> as sequential keyframes; slow push-in, cinematic 35mm look.",
    "reference_image_urls": [
      "https://your-storage/keyframe-1.webp",
      "https://your-storage/keyframe-2.webp"
    ],
    "resolution_tier": "768p",
    "aspect_ratio": "16:9",
    "duration": 5
  }'

Reference each input explicitly in the prompt with <Picture n>, <Video n> and <Audio n>. References that are uploaded but never named in the prompt are still billed, and the model treats them only as loose hints. Indexes follow upload order; when use_ref_video_audio is on, each reference video's own soundtrack takes an <Audio n> slot before your uploaded audio.

Basic Text-to-Video (Wan)

curl -X POST https://phosor.ai/api/v1/inference/submit \
  -H "X-API-Key: your-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "A serene mountain landscape at sunrise",
    "negative_prompt": "blurry, low quality",
    "num_frames": 81,
    "frames_per_second": 16
  }'

Image-to-Video Generation

import requests

# First, upload the image
with open("input.jpg", "rb") as f:
    upload_response = requests.post(
        "https://phosor.ai/api/v1/storage/image/upload",
        headers={"X-API-Key": "your-api-key"},
        files={"image": f}
    )

image_url = upload_response.json()["s3_url"]

# Then submit with image_url
response = requests.post(
    "https://phosor.ai/api/v1/inference/submit",
    headers={
        "X-API-Key": "your-api-key",
        "Content-Type": "application/json"
    },
    json={
        "prompt": "The scene comes alive with movement",
        "image_url": image_url,
        "num_frames": 81,
        "frames_per_second": 16
    }
)

print(response.json())

Using LoRA Models

import requests

response = requests.post(
    "https://phosor.ai/api/v1/inference/submit",
    headers={
        "X-API-Key": "your-api-key",
        "Content-Type": "application/json"
    },
    json={
        "prompt": "A character in anime style walking in a park",
        "lora_id": "abc123-def456-ghi789",
        "lora_scale": 0.8,
        "num_frames": 81,
        "frames_per_second": 16,
        "seed": 42
    }
)

print(response.json())

Custom Video Length

import requests

# Generate a 10-second video (160 frames at 16 fps)
response = requests.post(
    "https://phosor.ai/api/v1/inference/submit",
    headers={
        "X-API-Key": "your-api-key",
        "Content-Type": "application/json"
    },
    json={
        "prompt": "A time-lapse of clouds moving across the sky",
        "num_frames": 160,
        "frames_per_second": 16
    }
)

print(response.json())

Text-to-Image (Z-Image Turbo)

curl -X POST https://phosor.ai/api/v1/inference/submit \
  -H "X-API-Key: your-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "A cat sitting in a sunlit garden",
    "model": "z-image/turbo/text-to-image",
    "width": 1024,
    "height": 1024
  }'

Text-to-Image (Qwen Image 2512)

curl -X POST https://phosor.ai/api/v1/inference/submit \
  -H "X-API-Key: your-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Professional portrait, studio lighting, shallow depth of field",
    "model": "qwen-image/v2512/text-to-image",
    "width": 1024,
    "height": 768,
    "num_inference_steps": 28
  }'

Image-to-Image

import requests

# Step 1: Upload input image
with open("input.jpg", "rb") as f:
    upload_response = requests.post(
        "https://phosor.ai/api/v1/storage/image/upload",
        headers={"X-API-Key": "your-api-key"},
        files={"image": f}
    )

image_url = upload_response.json()["s3_url"]

# Step 2: Submit image-to-image
response = requests.post(
    "https://phosor.ai/api/v1/inference/submit",
    headers={
        "X-API-Key": "your-api-key",
        "Content-Type": "application/json"
    },
    json={
        "prompt": "Oil painting style",
        "model": "z-image/turbo/image-to-image",
        "image_url": image_url,
        "strength": 0.7
    }
)

print(response.json())

Speech-to-Video (S2V)

curl -X POST https://phosor.ai/api/v1/inference/submit \
  -H "X-API-Key: your-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "wan/v2.2-a14b/speech-to-video",
    "audio_url": "audio/aud-abc123",
    "image_url": "images/img-def456",
    "resolution_tier": "480p",
    "num_frames": 77
  }'

Animate

curl -X POST https://phosor.ai/api/v1/inference/submit \
  -H "X-API-Key: your-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "wan/v2.2-a14b/animate",
    "video_url": "videos/vid-abc123",
    "image_url": "images/img-def456",
    "resolution_tier": "480p",
    "num_frames": 17
  }'

Speech-to-Video (S2V) Parameters

S2V Pricing (10 credits = $1 USD, LoRA adds 1.2×):

Supported resolutions: 480p, 512p, 720p only (no 1080p). Max 77 frames.

Animate Parameters

Animate Pricing (10 credits = $1 USD, LoRA adds 1.2×):

Supported resolutions: 480p, 512p, 720p only (no 1080p). Max 17 frames.

Audio (TTS) Generation

Generate speech audio from text using Qwen3-TTS CustomVoice.

Example

import requests

response = requests.post(
    "https://phosor.ai/api/v1/inference/submit",
    headers={"Authorization": "Bearer YOUR_API_KEY"},
    json={
        "model": "qwen3-tts/text-to-speech/1.7b",
        "text": "Hello, welcome to Phosor AI.",
        "speaker": "Sohee",
        "language": "English"
    }
)
job = response.json()
request_id = job["request_id"]

Audio Parameters

Available speakers (9 voices): Vivian, Serena, Uncle_Fu, Dylan, Eric, Ryan, Aiden, Ono_Anna, Sohee

Available languages (11): Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish, Italian, Auto

Language shortcodes also accepted: zh, en, ja, ko, de, fr, ru, pt, es, it

Audio Pricing (10 credits = $1 USD)

Example: 200 chars → max($0.003, 200 × $0.00003) = $0.006 (0.06 credits)

The result URL in /api/v1/inference/result/{request_id} will be an .mp3 audio file.

Video Resolution and Frame Limits

MiniMax H3:

• duration: 4–15 seconds
• frames_per_second: fixed at 24fps
• resolution_tier: 480p or 768p
• aspect_ratio: 16:9, 4:3, 1:1, 3:4, 9:16

Wan text-to-video and image-to-video use num_frames instead of H3's duration-based controls. The maximum is 161 frames at 480p, 512p and 720p; at 1080p it is 81 frames for Standard or 153 for Turbo. Wan frame rate: 4–60 fps (default: 16).

For the complete H3 dimension table, Wan dimensions and model-specific input limits, see Video Generation Limits.

Image Resolution Limits

All resolutions are supported at the same price:

• 512×512, 1024×1024, 1024×768, 768×1024, 1280×768, 768×1280

Related Endpoints

• GET /api/v1/inference/status - Check task status
• GET /api/v1/inference/result/{request_id} - Get completed video or image
• POST /api/v1/storage/image/upload - Upload image for I2V or I2I