The volume engine: uncensored 5-8s clips with first+last frame control.
Spicy 1.0 Lite I2V (Wan 2.2 LoRA Spicy) is the mass-production tier of the Spicy lineup: fixed 5s/8s silent clips at 480p/720p, first and last frame control for precise start-to-end transitions, and a price point that makes thousands of variants affordable. Silent output pairs freely with your own TTS, BGM or platform audio trends.
Model Type:
No output yet. Run the model to see results.
Click 'Run Model' to generate output
Exact unit prices and billing rules for this model. All prices in USD.
| Resolution | No reference video |
|---|---|
| 480p | $0.04 |
| 720p×2 | $0.08 |
480p · 8s output → 8 × $0.04/s
≈ $0.320
Endpoint
https://api.spicyrouter.ai/v1/videosAll requests require a Bearer API key.
Submit Task
curl -X POST \
"https://api.spicyrouter.ai/v1/videos" \
-H "Content-Type: application/json" \
-H "Authorization: Bearer YOUR_API_KEY" \
-d '{
"model": "wan2.2-i2v-spicy",
"duration": 5,
"image_urls": "",
"prompt": "your prompt here",
"prompt_extend": true,
"resolution": "480p"
}'Poll Status
# Replace TASK_ID with the data.task_id from the submit response
curl -X GET "https://api.spicyrouter.ai/v1/status/TASK_ID" \
-H "Authorization: Bearer YOUR_API_KEY"Request Parameters
| Field | Required | Type | Values | Description |
|---|---|---|---|---|
duration | No | enum | "5", "8" Default: | Duration |
image_urls | Yes | array | Default: | Image Urls |
prompt | Yes | string | 1 ~ 4000 chars Default: | Prompt |
prompt_extend | No | boolean | Default: | Prompt Extend |
resolution | No | enum | "480p", "720p" Default: | Resolution |
Battle-tested prompts to get production-grade output in one shot.
A 5-second video. The scene opens with a medium shot of a woman fully clothed in casual wear, standing in a softly lit bedroom, smiling at the camera. She rips her top off, then shoves her jeans down and kicks them away. Now completely naked, she keeps her eyes locked on the camera. Static shot. Soft bedroom lighting, detailed skin texture, 4K, photorealistic.
Why Lite
The rule of thumb in the Spicy lineup: quality jobs go to Pro, volume jobs go to Lite. A 5-second 480p clip lands in about 50 seconds and costs pocket change — which means A/B tests, content libraries and whole feeds of motion are finally affordable.
Per-second billing this low means whole batches of five-second clips for the price of a coffee. Generate every variant, keep the winners, throw the rest away without regret.
Pass the first frame via image and an optional end state via last_image — the model choreographs a stable transition between them.
The v2.4 sampling overhaul cut runtime in half: a 480p 5s video renders in roughly 50 seconds. Queue hundreds and let the pipeline run.
No baked-in audio — pair every clip with your own TTS, trending BGM or platform sound library. One video, infinite soundtracks.
Use Cases
Breathing, blinking, subtle motion for character dialogue scenes — micro-animations that make static sprites feel alive.
Living avatars in your chat interface — pre-generate an emotion-state video library, match it to conversation tags, respond in milliseconds.
Fast, eye-catching clips built for TikTok / Reels / Shorts — silent by design, so every trending sound is yours to ride.
Replace static thumbnails with 5-second motion covers across your content platform — auto-generated on every upload.
Prompt Craft
The model will not reliably generate nudity from "already naked" — show her removing her clothes first. A medium shot in casual wear works best; avoid fixating on specific garments.
Never describe her turning around — write A jumpcut instantly places her... instead. Mandatory for rear-facing positions and scene changes; skip it when she faces forward throughout.
Specify gaze in every scene: eyes locked on the camera the entire time. Facing away? looking back over her shoulder at the camera. Eye contact is what sells intimacy.
Vague body language produces vague bodies. Name it: vulva, nipples erect, trimmed pubic hair — the last one alone kills the plastic look.
Unspecified lighting defaults to flat and unflattering. Close every prompt with Static shot. Soft lighting, detailed skin texture, 4K, photorealistic. — or a variant like candlelit / golden hour.
5–8 seconds fits exactly one continuous action: strip (0:00–0:01), optional jumpcut, then a single act to the end. Cramming more in causes tearing. And never write audio cues — Lite is silent.
Master Template
Fill the brackets, keep the skeleton. Built-in prompt expansion is on by default, but a well-structured prompt still outperforms a lucky one — batch your variants from this base.
A 5-second video. The scene opens with a medium shot of a woman fully clothed in casual wear, standing in [LOCATION], [EXPRESSION] at the camera. She [STRIP ACTION], then [STRIP ACTION]. Now completely naked, her bare breasts, erect nipples, and vulva with trimmed pubic hair are fully exposed. [JUMPCUT IF NEEDED]. [SEX ACT — one continuous action]. She [EYE CONTACT: eyes locked on the camera / looking back over her shoulder]. [LIGHTING: soft bedroom / candlelit / golden hour], detailed skin texture, 4K, photorealistic. Static shot.
Master Template
Position sub-templates
Don’t
FAQ
That's a feature, not a bug. Silent video decouples motion from audio: pair clips with your own TTS engine for companions, ride trending sounds on social platforms, or license BGM once and reuse it across your whole library. If you need native audio generation, use Wan 2.7 I2V Spicy instead.
The base image model rarely produces reliable nudity from a static "already naked" description — the undressing motion is what unlocks anatomically correct exposure. Alternatively, start from an already-nude first frame via image and skip the opener entirely; the image defines the starting state.
Pass the starting image via image (required) and the desired end state via last_image (optional, URL or Base64). The model generates a stable transition between the two — ideal for controlled reveals, pose-to-pose motion, and seamless loops where the last frame matches the first.
A 480p 5s clip normally renders in ~50 seconds; 720p and 8s take longer. If a task stalls: check that both images are under 20MB with sides between 240–8000px, simplify an overloaded prompt (one act only), and retry — transient queue congestion clears on resubmission. Set client timeouts to at least 3–5 minutes for 720p/8s jobs.
Simple rule: volume goes to Lite, quality goes to Pro. Lite gives you 5/8s silent clips for batch production and traffic content. Pro gives you 2–15s cinema-grade 1080p with native audio for premium, monetizable content. Most platforms run both.
One image in, endless motion out. A hundred clips for the price of a cocktail — start batching.