Wan 2.2 Text To Image LoRA

Wan 2.2 Text To Image LoRA

Playground

WAN 2.2 generates super-detailed images from text prompts and supports custom LoRAs for fine-grained style and subject control. Ready-to-use REST inference API, best performance, no coldstarts, affordable pricing.

Features

Wan-2.2-LoRA (Text-to-Image)

Wan-2.2-LoRA builds upon the acclaimed Wan 2.2 text-to-image model by introducing full LoRA (Low-Rank Adaptation) compatibility — empowering creators to fine-tune visuals with personalized styles, characters, or aesthetics. It combines Wan’s signature cinematic rendering and world-class detail synthesis with the flexibility of custom-trained LoRAs.

Why it looks great

• LoRA-ready architecture – Import .safetensors LoRA weights directly from Civitai, Hugging Face.
• Cinematic lighting engine – Advanced diffusion backbone that simulates depth, tone, and atmosphere with film-grade realism.
• Text rendering excellence – Handles both English and Chinese typography seamlessly within the image, not as overlays.
• Cross-style adaptability – From photorealism to anime, oil painting, 3D CG, or minimalism — one prompt can shift universes.
• Consistent composition – Retains character identity and spatial coherence across multi-prompt workflows.

Limits and Performance

• Max resolution per job: up to 1536 × 1536 pixels
• LoRA path: supports <owner>/<model-name> or direct .safetensors URLs
• LoRA scale: adjustable from 0.1 – 1.5 (default = 1.0)
• Output formats: JPEG / PNG / WEBP
• Processing speed: ~6–9 seconds per image
• Prompt input: multi-line, bilingual, descriptive prompts supported

Pricing

• $0.025 per image Each image is billed individually.

How to Use

1. Write a detailed prompt (in English or Chinese).
2. Set size — width and height (up to 1024×1024).
3. Add LoRA(s) – paste LoRA path or URL; adjust scale for blending strength.
4. (Optional) Set a seed for reproducibility (-1 = random).
5. Choose output format (JPEG / PNG / WEBP).
6. Run → preview result → iterate with different LoRAs or scales.

Pro tips for best quality

• Mix multiple LoRAs for hybrid aesthetics (e.g., cyberpunk + watercolor).
• Use 0.6–0.9 scale for realistic subtle blending.
• Lock seed to maintain consistent faces or characters across styles.
• Start from simple prompts; layer complexity gradually for control.

Reference

• Use your LoRA
• Train your LoRA

Note

• LoRAs from Civitai or Hugging Face are also supported if exported in .safetensors format.
• For multi-LoRA blending, ensure each LoRA file is stylistically aligned for optimal results.

# Submit the task
curl --location --request POST "https://api.wavespeed.ai/api/v3/wavespeed-ai/wan-2.2/text-to-image-lora" \
--header "Content-Type: application/json" \
--header "Authorization: Bearer ${WAVESPEED_API_KEY}" \
--data-raw '{
    "size": "1024*1024",
    "seed": -1,
    "output_format": "jpeg",
    "enable_sync_mode": false,
    "enable_base64_output": false
}'

# Get the result
curl --location --request GET "https://api.wavespeed.ai/api/v3/predictions/${requestId}/result" \
--header "Authorization: Bearer ${WAVESPEED_API_KEY}"