Supported base models
Flux.1 Dev LoRA
Photoreal characters, products, styles
The general-purpose pick. Train a face, a product, or a visual style and use the weights anywhere Flux runs.
Qwen Image LoRA
High-fidelity Asian character LoRAs
Strong on stylised illustration and East-Asian character likeness. Same dataset format as Flux — drop in a zip and go.
Wan 2.2 Image LoRA
Stylized image generation
High-resolution image LoRA on the Wan 2.2 base. Returns dual high/low-noise weights for advanced workflows.
Wan 2.2 I2V LoRA
Image-to-video motion control
Train on short video clips to lock in motion, camera, or character animation. The flagship for custom video.
Credits are held when you start, fully refunded if training fails. Output is a .safetensors file you can download and use anywhere.
How it works
Upload dataset
Drop a zip of images (or short clips for I2V). 10–50 samples is the sweet spot.
Pick base model + trigger word
Choose Flux, Qwen, or Wan 2.2 and set a short trigger like sks-style or ohwx.
Click Start
Credits are held up front, refunded on failure. Training typically finishes in 10–30 minutes.
Download .safetensors when done
Direct download of the LoRA weights. Use them on Hypereal or anywhere else that loads them.
Use cases
Character LoRAs
"You" in five styles. Train on 15 selfies and generate consistent likeness across any prompt or product mockup.
Product LoRAs
Lock in your brand asset — a shoe, a bottle, a logo — and place it in any scene without re-prompting from scratch.
Motion LoRAs
Custom video animations. Train on 5–10 short clips of a signature movement and reuse it across new images.