lora-support
Wan 2.1 i2v 480p Ultra-Fast generates unlimited image-to-video content at 480p, supporting custom LoRAs for style personalization. Ready-to-use REST inference API, best performance, no coldstarts, affordable pricing.
Input
Idle
您的請求將花費 $0.125 每次運行。
使用 $10 您可以運行此模型大約 80 次。
示例查看全部
README
Wan 2.1 I2V 480p LoRA Ultra Fast — wavespeed-ai/wan-2.1/i2v-480p-lora-ultra-fast
Wan 2.1 I2V 480p LoRA Ultra Fast is a fast, cost-efficient image-to-video model that animates a single reference image into a short clip guided by your prompt. It supports LoRA injection so you can enforce a specific motion/style/character look while keeping generation lightweight—great for rapid iteration, previews, and high-volume variants.
Key capabilities
- Image-to-video (I2V) generation at 480p
- Prompt-driven motion with coherent, image-anchored composition
- LoRA support for consistent style, motion patterns, or character look
- Tunable quality/speed via inference steps
- Great for quick drafts, storyboards, and large-scale A/B testing
Use cases
- Quick animation passes for key art, avatars, and character cards
- Motion experiments (e.g., 360° rotation, walk cycles, subtle facial motion) using LoRAs
- Fast concept validation for ads and social loops
- Batch generation of multiple motion variants from the same image
- Previews before re-running at higher resolution (720p/1080p)
Pricing
| Resolution | Duration | Price per run | Effective price per second |
|---|---|---|---|
| 480p | 5s | $0.125 | $0.025/s |
| 480p | 10s | $0.188 | $0.019/s |
Inputs
- image (required): reference image that anchors subject and composition
- prompt (required): action + camera + motion intent
- loras (optional): LoRA items to apply (path + scale)
Parameters
- duration: clip length (commonly 5s or 10s)
- num_inference_steps: sampling steps (higher can improve motion coherence)
- guidance_scale: prompt adherence strength
- flow_shift: motion behavior tuning
- negative_prompt: what to avoid (blur, jitter, artifacts, low quality)
- seed: set for reproducible results (-1 for random)
- size: output size preset (e.g., 832×480)
- loras[].path: LoRA identifier (e.g., creator/name) or a safetensors URL
- loras[].scale: LoRA strength (start around 0.8–1.0)
Prompting guide (I2V)
Keep the prompt like a director note:
- Subject: what’s on screen
- Motion: what changes over time (spin, turn, step forward, hair movement)
- Camera: locked-off / slow push-in / orbit / handheld, etc.
- Constraints: “smooth motion”, “no camera shake”, “keep character consistent”
Example prompt
Felt doll standing on a grassy field. The doll slowly rotates for a full 360-degree turn with smooth motion, stable framing, no jitter. Soft daylight, shallow depth of field, gentle background bokeh.