Video models

Published Aug 24, 2026

Updated Oct 9, 2026Last verified Sep 20, 2026

12 min read

Uncensored AI Image-to-Video Generators: From Still to Spicy Video API

Compare Wan 3, Pro, and the new Prime and Prime Pro tiers by turnaround, resolution, keyframes, reference inputs, and live image-to-video API pricing.

Key takeawayImage-to-video is not merely making a still move; it is preserving identity, composition, and usable motion over time.

Short answer

This page solves one job: turning an approved still into a deliverable clip while identity, composition, and motion stay usable across the whole timeline. The method is to set the starting point with a first frame, add a last frame when the ending must be controlled, constrain style and rhythm with image, video, and audio references, and validate the first pass at five seconds and a lower resolution before raising resolution or duration. Spicy routes require an age-confirmed account, and real-person material requires rights and consent.

Start with the question behind the search

A correct still can drift once it enters a video model: faces and clothing change, backgrounds are redesigned, camera movement overwhelms subject movement, or the opening frame no longer connects to the result. People searching for uncensored image-to-video also need clarity on account eligibility and allowed scope, but production quality still depends on references, motion direction, duration, and retry design.

Choose by task

How the five ModelRush video routes actually differ. Capabilities come from the Registry operation limits; the live pricing table further down this page carries the rates.

Model	Resolution	Duration	Reference inputs	Best stage
Wan 3 Spicy	480P / 720P / 1080P	Fixed duration of 2–30 seconds; omission defaults to 5 seconds, -1 is unavailable; billing uses that duration	First frame, last frame, up to 10 images / 5 videos / 5 audio clips	Low-cost exploration and most delivery
Wan 3 Pro Spicy	1080P / 2K / 4K	The same unified workflow as Wan 3	Same as Wan 3	High-resolution final delivery of approved shots
Wan 3 Prime Spicy	480P / 720P / 1080P	Fixed duration of 2–30 seconds; omission defaults to 5 seconds, -1 is unavailable; billing uses that duration	First frame, last frame, up to 10 images / 5 videos / 5 audio clips	Faster iteration up to 1080P
Wan 3 Prime Pro Spicy	1080P / 2K / 4K	Fixed duration of 2–30 seconds; omission defaults to 5 seconds, -1 is unavailable; billing uses that duration	First frame, last frame, up to 10 images / 5 videos / 5 audio clips	Faster high-resolution delivery up to 4K
Wan 2.7 R2V	720P / 1080P	2–10s	A `media` array of reference material, driven by the reference subject rather than a first frame	Keeping a specific character or object without locking the composition

Start from your image in the Playground

You can try image-to-video without an API key. Open the Wan 3 Playground, then sign in or create an account. Check that Image-to-video and Wan 3 Spicy are selected.

Upload your image as the first frame. Use an image you own or have permission to use. Add a last frame only when you need a specific ending.
Describe one subject action and one camera movement. For example: “The bottle stays centered while the camera slowly moves closer.” This is a suggested prompt, not a tested result.
Start with a fixed 5 seconds at 480P. Check the estimated charge and your balance before you generate. Each generation uses account credit; creating an account does not promise a free generation.
Inspect the start, middle, and end of the clip. If the subject or motion drifts, change one setting and compare the next result. A retry is another generation and can add cost.

Use the current price table below for public rates. Your account quote determines your actual charge.

First frame versus first and last frame

With only a first frame, the model decides where the shot goes. The start is fixed and the ending is inferred probabilistically from the prompt, so repeated runs land differently. That suits exploration, mood shots, and footage that does not need a precise destination.

With both a first and a last frame, both ends are locked and the model only owns the transition. That suits shots that must land on a named composition: a final product position, a character turning to a specific angle, or a cut that has to match the next shot. The cost is that the two keyframes must be coherent with each other. If lighting, wardrobe, or background differ between them, the model forces a bridge and the clip breaks more often, not less.

The practical order is: use a first frame to confirm the subject and motion direction can work at all, then decide whether the ending is worth a second keyframe. Starting with both introduces two variables at once and hides which end caused the failure.

What image, video, and audio references each constrain

The Wan 3 Registry contract allows up to 10 reference images, 5 reference videos, and 5 reference audio clips, and each constrains a different dimension.

Image references constrain appearance: identity, wardrobe, material, palette. Several images from different angles do more than repeating the same person in prose.
Video references constrain motion: rhythm, camera movement, and how large the action is. When a slow push in comes back as a fast dolly, one reference video is faster than ten more adjectives.
Audio references constrain time. Wan 3 supports native audio, so a reference clip affects beat and lip alignment rather than simply attaching a soundtrack.

None of the three is a guarantee. The more references you attach, the more constraints the model has to satisfy at once, and contradictory references — a daylight image reference with a night-time video reference — visibly lower the pass rate.

Duration, resolution, and what a clip actually costs

Video bills per second and resolution sets the per-second rate, so cost is roughly rate x duration x attempts. The third term is what actually decides the budget: a five-second shot that needs four attempts costs more than a ten-second shot that lands first time.

So run short shots at a lower resolution to judge identity, composition, and motion direction, then promote the approved prompt to the target resolution. Wan 3 Spicy and Wan 3 Prime Spicy cover 480P, 720P, and 1080P; Wan 3 Pro Spicy and Wan 3 Prime Pro Spicy cover 1080P, 2K, and 4K. Choose resolution for the delivery requirement, then decide whether faster Prime turnaround is worth a higher per-second rate. Compare costs at the same resolution, duration, and attempt count, and record actual wait times. The live table below carries the rate in effect for each resolution.

How to evaluate beyond a polished demo

Test portrait close-ups, full-body action, product rotation, and complex backgrounds separately.
Fix the first pass at five seconds and a lower resolution to judge identity, composition, and motion direction.
Use first and last frames when the endpoint matters instead of describing the final frame only in prose.
Convert price per second into cost per approved shot, including unusable outputs and human review.

ModelRush options for the job

Wan 3 Prime Spicy targets faster iteration at 480P, 720P, or 1080P, with keyframes, multimodal references, native audio, and a specified 2–30 second duration.
Wan 3 Prime Pro Spicy targets faster delivery at 1080P, 2K, or 4K while keeping the same keyframe, reference, audio, and duration controls.
Wan 3 Pro Spicy is the current high-quality image-to-video route, with keyframes, image/video/audio references, native audio, and 1080P, 2K, or 4K output.
Wan 3 Spicy keeps the same unified reference workflow and up-to-30-second duration at 480P, 720P, or 1080P for lower-cost exploration.
Use Wan 3 Spicy for short first-pass tests, then move approved identity, motion, and composition to Wan 3 Pro Spicy for high-resolution delivery.

Minimum working request

This request uses modelrush/wan3-prime-spicy for a five-second 720P keyframe video. For Prime Pro, change model to modelrush/wan3-prime-pro-spicy and set resolution to 1080P, 2K, or 4K; Prime Pro does not accept 720P. The endpoint returns 202 with a generation ID and a poll URL. Save the first response so a network retry does not create a second billable job.

bash
curl https://api.modelrush.ai/v1/videos/generations \
  -H "Authorization: Bearer $MODELRUSH_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "modelrush/wan3-prime-spicy",
    "prompt": "The subject turns toward camera and the light warms, one slow push in",
    "first_frame": "https://example.com/approved-still.png",
    "last_frame": "https://example.com/approved-final-pose.png",
    "duration": 5,
    "resolution": "720P",
    "ratio": "16:9",
    "audio": true
  }'

A reproducible production workflow

Save the source-image hash, rights basis, and subject consent before creating a five-second test. Split the prompt into subject, action, environment, and camera layers, changing one variable at a time. Preserve generation ID, model, resolution, duration, cost, and review decision. Only approved shots move to a longer duration or higher resolution, with status synchronized by webhook or polling. See the complete request contract in the video API documentation.

Frequently asked questions

When should I choose Prime or Prime Pro?

Choose Wan 3 Prime Spicy for faster iteration at 480P, 720P, or 1080P, and Wan 3 Prime Pro Spicy for faster delivery at 1080P, 2K, or 4K. Both support keyframes, multimodal references, native audio, and up to 30 seconds. Prime tiers have higher per-second rates than their corresponding non-Prime tiers; check the live pricing table for current costs.

Can image-to-video preserve a character perfectly?

It cannot be guaranteed. A reference improves retention, while motion intensity, occlusion, camera movement, and duration all increase drift risk.

Should I use Wan 3 or Wan 3 Pro?

Use Wan 3 at 480P, 720P, or 1080P for lower-cost exploration. Choose Wan 3 Pro for final delivery at 1080P, 2K, or 4K. Both support keyframes, multimodal references, and native audio.

Can Spicy routes process any person's photo?

No. Real-person material requires clear rights and consent; minors and non-consensual intimate use are strictly prohibited.

Start with a five-second test, inspect the controlled Wan 3 benchmark, and put your real acceptance rate into the video cost calculator. Then confirm current capabilities and asynchronous states on the Wan 3 Spicy model page and in the image-to-video API documentation.

Current pricing for the models in this article

Rates come from the ModelRush Registry and use the same immutable price version as billing.

Model	Capability	Price variant	Customer price
Wan 2.7 R2V · modelrush/wan2.7-r2v	Video generation	1080P	$0.165 / second
Wan 2.7 R2V · modelrush/wan2.7-r2v	Video generation	720P	$0.110 / second
Wan 3 Prime Pro Spicy · modelrush/wan3-prime-pro-spicy	Video generation	1080P	$0.234 / second
Wan 3 Prime Pro Spicy · modelrush/wan3-prime-pro-spicy	Video generation	2K	$0.252 / second
Wan 3 Prime Pro Spicy · modelrush/wan3-prime-pro-spicy	Video generation	4K	$0.279 / second
Wan 3 Prime Spicy · modelrush/wan3-prime-spicy	Video generation	1080P	$0.252 / second
Wan 3 Prime Spicy · modelrush/wan3-prime-spicy	Video generation	480P	$0.061 / second
Wan 3 Prime Spicy · modelrush/wan3-prime-spicy	Video generation	720P	$0.126 / second
Wan 3 Pro Spicy · modelrush/wan3-pro-spicy	Video generation	1080P	$0.144 / second
Wan 3 Pro Spicy · modelrush/wan3-pro-spicy	Video generation	2K	$0.160 / second
Wan 3 Pro Spicy · modelrush/wan3-pro-spicy	Video generation	4K	$0.184 / second
Wan 3 Spicy · modelrush/wan3-spicy	Video generation	1080P	$0.180 / second
Wan 3 Spicy · modelrush/wan3-spicy	Video generation	480P	$0.045 / second
Wan 3 Spicy · modelrush/wan3-spicy	Video generation	720P	$0.090 / second

Registry revision · 2026-10-14.apipod

Next steps

Move straight from this article to model details, current pricing, API documentation, and the Playground.

Wan 3 Spicy model details Wan 3 Pro Spicy model details Wan 3 Prime Spicy model details Wan 3 Prime Pro Spicy model details Wan 2.7 I2V Spicy model details Wan 3 Spicy API pricing Image-to-video API documentation Open image-to-video in the Playground

Compare callable models

Apply the article's framework to live models by capability, I/O, price, and region.