Viggle releases LoRA for full character replacement in images

Viggle has released the Viggle/Viggle-Image-Swap LoRA adapter for Qwen-Image-2.1 on Hugging Face, which fully replaces a person in the source frame with a character from a character sheet in 6 steps without CFG, preserving pose, expression, framing, background, and lighting.

Compiled by Sergey KostenchukPublished 2026-11-08Updated 2026-11-08

2026-11-08 Multimodal HuggingFace

🎨 Viggle releases LoRA for full character replacement in images

The Viggle/Viggle-Image-Swap LoRA adapter for Qwen-Image-2.1 fully replaces a person in the source frame—from face to clothing—with a character from a character sheet, preserving pose, expression, framing, background, and lighting. Inference takes 6 steps without CFG; in turbo mode, it uses 2 LoRA steps plus 6 Qwen-Image-2.1-Turbo steps. Output resolution is 1 MP.

🌍 Character swap is reduced to a single pass without fine-tuning for each hero—simplifying pipelines for character generation, merch, and stylization. Downside: The Qwen Research license does not cover commercial use; a separate license is required.

👤 Download the adapter and the inference/swap.py script, assemble a 1536×1024 character sheet (portrait plus three full-body angles), and replace the person in the frame with your hero. Speed is 2.2 seconds with 46 GB of memory—this is on a B200; on weaker GPUs, time and memory requirements will increase, and swap quality is only confirmed by the author.

Source 1: https://huggingface.co/Viggle/Viggle-Image-Swap

Sources