2 Model Trainers Now Live on WaveSpeedAl: Train Your Own Wan or Flux for LoRA

We are excited to introduce the Wan-Trainer and Flux-Dev-LoRA-Trainer — 2 training tools that can train models with your own dataset.

About Wan-Trainer

Wan is an open source video generation modeling suite developed by Alibaba that supports a variety of tasks, including Text-to-Video, Image-to-Video, Video Editing, Text-to-Image, and Video-to-Audio.

With Wan-Trainer provided by WaveSpeedAI, you can easily upload your own dataset and quickly fine-tune your own Wan-LoRA model in a simple operation! All tasks supported by Wan — from Text-to-Video — can be trained directly on WaveSpeedAI.

Creating a Wan_3D_Pintr_Lora Example:

• Video Dataset Example: Video Dataset
• Dataset Preprocessing: Prepare approximately 10 video clips with consistent features and compress them into a ZIP file.
• Duration: Each video clip should be approximately 5 seconds long.
• Annotation: The Wavespeed platform provides an automatic tagging functionality.
• Parameters:
  • learning_rate: The rate at which the model learns. Higher values may accelerate training but risk overfitting.
  • trigger_phrase: A specific phrase that activates the model to generate an image.
  • number_of_steps: 1000
  • auto_scale_input: If enabled (true), the input video will automatically be scaled to 81 frames at 16fps.

json Example:

{
  "learning_rate": 0.0002,
  "trigger_phrase": "3d",
  "number_of_steps": 1000,
  "auto_scale_input": true,
  "training_data_url": "https://d32s1zkpjdc4b1.cloudfront.net/media/8cb24be7c94d472cb7895e1499bd0ae7/archives/1746773122258761405_gMU3b9hp.zip",
  "model_id": "wavespeed-ai/wan-trainer"
}

About Flux-Dev-LoRA-Trainer

Flux Dev is a non-commercial open source version of the text-to-image model developed by Black Forest Labs. It excels in image quality, detail performance and cue word reproduction. With Flux-Dev-LoRA-Trainer provided by WaveSpeedAI, you can easily upload your own dataset and fine-tune your own Flux-Dev-LoRA model in a simple operation!

Creating a FLUX_Style_Lora Example:

• Image Dataset Example: Image Dataset
• Dataset Preprocessing: Place all training images of the same style in a single folder.
• Resolution: Original images should ideally have a resolution of 1024×1024 or higher. They can be converted to 512×512 later if needed.
• Aspect Ratio: When training on Flux.1, a 1:1 aspect ratio can be used. Additional aspect ratios may be added to enhance diversity.
• Clarity: Use clear, high-quality images whenever possible.
• Diversity: Incorporate variations in angles, lighting conditions, and outfits to improve model adaptability.
• Parameters:
  • trigger_word: Trigger word to be used in captions. If set to None, no trigger word will be used. If captions are not provided, the trigger word will replace captions. If captions exist, the trigger word is ignored.
  • create_masks: If True, segmentation masks are applied to weight training loss. For human subjects, face masks are prioritized when possible.
  • steps: Typically calculated as (number of images) × 100 (e.g., 20 images → 2000 steps).
  • is_style: When True, configures training for style transfer. This disables segmentation and captioning, using the trigger word to define the style.
  • is_input_format_already_preprocessed: Indicates whether input data is preprocessed. Default (False) assumes raw input with matching image/caption filenames (e.g., photo.jpg and photo.txt). Set to True for preprocessed data.
  • data_archive_format: Archive format. If unspecified, inferred from the URL.

json Example:

{
  "create_masks": false,
  "images_data_url": "https://d32s1zkpjdc4b1.cloudfront.net/media/8cb24be7c94d472cb7895e1499bd0ae7/archives/1747100030859337188_6Ddljsqo.zip",
  "is_input_format_already_preprocessed": true,
  "is_style": true,
  "steps": 2000,
  "trigger_word": "test",
  "model_id": "wavespeed-ai/flux-dev-lora-trainer"
}

Try Wan-Trainer and Flux-Dev-LoRA-Trainer Now on WaveSpeedAI

Fine-tune your own AI models with Wan-Trainer and Flux-Dev-LoRA-Trainer on WaveSpeedAI — no complex setup required.

Whether it’s Text-to-Video, Image-to-Video, or even Video Editing, Wan-Trainer lets you upload your own data and customize the powerful Wan model to fit your creative goals.

With Flux-Dev-LoRA-Trainer, you can easily customize the Flux-Dev model—renowned for its image quality and prompt fidelity—by training your own LoRA using your own images.

Whether you’re building expressive video workflows or tailored image generators, WaveSpeedAI gives you full control and fast results. Start fine-tuning your models today!

📲 Follow us on Twitter, LinkedIn and join our Discord channel to stay updated.

© 2025 WaveSpeedAI. All rights reserved.