PhosorVideo Generation Limits
API version: v1.2.4
Choose the limits for the model you send to
POST /api/v1/inference/submit. MiniMax H3 uses
output duration; Wan uses frame counts. Their dimensions and tuning controls
are model-specific.
Resolution & Frame Limits
MiniMax H3
Standard and Turbo support 4–15 second output at fixed 24fps. Use
resolution_tier, aspect_ratio and duration for all three input modes:
text-to-video, image-to-video and reference-to-video. An image-to-video request
also needs image_url; end_image_url is optional. A reference-to-video
request needs at least one reference image or video.
| Resolution Tier | Aspect Ratio | Dimensions | Duration | FPS |
|---|---|---|---|---|
| 480p | 16:9 | 832×480 | 4–15s | 24 fixed |
| 480p | 4:3 | 640×480 | 4–15s | 24 fixed |
| 480p | 1:1 | 480×480 | 4–15s | 24 fixed |
| 480p | 3:4 | 480×640 | 4–15s | 24 fixed |
| 480p | 9:16 | 480×832 | 4–15s | 24 fixed |
| 768p | 16:9 | 1344×768 | 4–15s | 24 fixed |
| 768p | 4:3 | 1024×768 | 4–15s | 24 fixed |
| 768p | 1:1 | 768×768 | 4–15s | 24 fixed |
| 768p | 3:4 | 768×1024 | 4–15s | 24 fixed |
| 768p | 9:16 | 768×1344 | 4–15s | 24 fixed |
MiniMax H3 Reference Inputs
Current Standard and Turbo limits at both 480p and 768p:
| Reference input | Limit |
|---|---|
| Images, when sending images only | 9 |
| Images, when combined with video or audio | 5 |
| Videos | 3 |
| Audio clips | 3 |
| Combined duration of all reference videos | 15.1 seconds |
| Reference video frame rate | At most 24fps |
| Reference image longest edge | At most 2048px |
| Reference image aspect ratio | At most 4:1 in either orientation |
| Duration of each reference audio clip | At most 10 seconds |
These counts are not a promise that every per-type maximum can be combined. Audio alone is not a valid visual reference. Input video dimensions need not match the output dimensions; duration, frame-rate and readable-file requirements still apply. Reference URLs must be accessible to the service. See upload requirements and the reference parameters and prompt tags.
The current limits are available in
GET /api/v1/pricing/config:
minimax_h3_ref2va_limits and minimax_h3_ref2va_turbo_limits.
Wan 2.2 Text-to-Video and Image-to-Video
Wan Standard and Turbo have different maximum frame counts at 1080p:
| Resolution | Standard max frames | Turbo max frames |
|---|---|---|
| 480p | 161 | 161 |
| 512p | 161 | 161 |
| 720p | 161 | 161 |
| 1080p | 81 | 153 |
These are Wan text/image-to-video limits. For LTX, speech-to-video and animation inputs, consult the model-specific sections of the submit reference.
Frame Rate
For Wan, frames_per_second accepts 4–60. Set it explicitly when calculating
duration from num_frames. MiniMax H3 output is fixed at 24fps; use duration
to choose its length.
Frame Alignment
Wan frame counts are aligned down to 1 + 4k, with a minimum of 5 frames.
For example, a request for 80 frames becomes 77. This Wan alignment rule does
not define H3's output duration; use H3's duration-based parameters above.
Parameter Constraints
| Parameter | Applies to | Guidance |
|---|---|---|
prompt | Video generation | Describe the subject, action, setting and camera movement |
negative_prompt | Supported Wan modes | Optional; not supported by MiniMax H3 |
duration | MiniMax H3 | 4–15 seconds; default 5 |
resolution_tier | MiniMax H3 | 480p or 768p |
aspect_ratio | MiniMax H3 | 16:9, 4:3, 1:1, 3:4 or 9:16 |
num_frames | Wan | Respect the selected tier and Standard/Turbo frame cap |
frames_per_second | Wan | 4–60; H3 output remains fixed at 24fps |
width, height | Model-specific | Prefer the supported tier and aspect-ratio combination |
seed | Supported models | Integer; omit for randomized generation |
lora_id, lora_scale | Supported LoRA workflows | Use a ready LoRA with a compatible model |
Supported Resolutions
H3 dimensions are listed above. Common Wan tier/aspect combinations are:
| Wan tier | Landscape 16:9 | Portrait 9:16 | Square 1:1 |
|---|---|---|---|
| 480p | 854×480 | 480×854 | 480×480 |
| 512p | 960×512 | 512×960 | 512×512 |
| 720p | 1280×720 | 720×1280 | 720×720 |
| 1080p | 1920×1080 | 1080×1920 | 1080×1080 |
Standard (Non-Turbo) Mode Quality Parameters
The optional Wan Standard tuning parameters in the submit reference apply to supported Wan models. They are not MiniMax H3 controls. Use the H3 input-mode, duration, resolution and reference parameters for H3 requests.
Pricing
Use current pricing or
GET /api/v1/pricing/config for rates. The billing
units differ:
| Model family | Billing unit |
|---|---|
| MiniMax H3 text/image-to-video, Standard or Turbo | Per output second at the selected tier |
| MiniMax H3 reference-to-video, Standard or Turbo | Output seconds plus billable reference inputs |
| Wan video workflows | Per frame, with model/tier-specific LoRA pricing |
For H3 reference-to-video, the first five reference images are free under the current schedule. Additional images cost $0.0075 each for Standard or $0.0025 each for Turbo. Reference audio is free; reference video is billed per input second at the selected variant and output tier's reference-video rate. See the complete reference fee table.
10 credits = $1 USD. Recorded account billing is the source for actual charges; an example or estimate is not a completed request's bill.
Concurrency Limits
Concurrency depends on your account's current allowance. Do not assume that all accounts can submit the same number of simultaneous jobs. If a request exceeds the allowance, follow the returned error and wait for active requests to finish. See common API errors.
Queue Limits
Queue and processing times depend on the selected model and capacity; this page does not promise a fixed completion time. Poll request status for progress. A client or network timeout does not prove that submission failed or that an accepted job was cancelled. Check request history before submitting again to avoid duplicate paid generations.
Image-to-Video (I2V) Requirements
Provide a usable image_url. You can upload an image
or import an image URL first. See
upload requirements for supported file formats and
size limits. For H3, an optional end_image_url supplies an ending image.
LoRA Model Requirements
Use LoRA only with a model that supports it. The LoRA must be ready and compatible with the selected workflow; see the LoRA guide. LoRA availability and fees must not be inferred from another model family.
Error Examples
Exceeds Frame Limit
Check the selected Wan tier and variant: 1080p Standard accepts at most
81 frames, while Turbo accepts at most 153. Reduce num_frames accordingly.
For H3, select an output duration within 4–15 seconds instead.
FPS Out of Range
For Wan, choose frames_per_second between 4 and 60. For H3 reference inputs,
use reference videos at 24fps or below; H3 output stays fixed at 24fps.
Concurrency Limit
Wait for active requests to finish, then retry according to the returned error. Keep each accepted request ID so it can be checked through status and history without creating another generation.