The only LoRA available so far for Qwen Image 2.1, meant to improve character consistency in anime-style editing. Creator loads it into the edit workflow and generates a character reference sheet.
Qwen 2.1 Anime Consistency LoRA
Experimental LoRA for Qwen-Image-2.1 that improves character consistency in anime-style image editing.
Updated EU status checked Last shown
Qwen 2.1 Anime Consistency LoRA at a glance
- What is Qwen 2.1 Anime Consistency LoRA?
- Qwen 2.1 Anime Consistency LoRA is a model from WarmBloodAban. Experimental LoRA for Qwen-Image-2.1 that improves character consistency in anime-style image editing.
- Who makes Qwen 2.1 Anime Consistency LoRA?
- Qwen 2.1 Anime Consistency LoRA is made by WarmBloodAban.
- Does Qwen 2.1 Anime Consistency LoRA work in the EU?
- Available in the EU. Downloadable open weights under Apache-2.0 with no regional restrictions in the license, so usable in the EU/EEA when run locally or on self-managed infrastructure. Last verified 14 Nov 2026.
- How much does Qwen 2.1 Anime Consistency LoRA cost?
- Open source. Free to download from Hugging Face; requires the Qwen-Image-2.1 base model and local or self-hosted compute. Licence: Apache-2.0.
- Where was Qwen 2.1 Anime Consistency LoRA shown?
- In 1 creator video by AI Search. Most recently by AI Search on 22 Sept 2026, at 13:17 in “Finally! New best local AI image editor is here”.
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
Qwen2.1_Anime_consistency is a LoRA adapter for Qwen-Image 2.1's image-editing mode, published by WarmBloodAban in the repository Qwen-Image-2.1-LoRAs on Hugging Face. It is intended to keep characters consistent across anime-style edits. It was fine-tuned mainly on character model sheets (4-view references) and a range of facial-expression edits.
The author labels it experimental. It was mainly built to test and benchmark training parameters for Qwen Image 2.1, so editing results and output stability are not guaranteed. The recommended LoRA weight is 0.6 to 0.8, and sampling steps and CFG should follow the base model's official settings. A YouTube creator (AI Search) described it as the only LoRA available for Qwen Image 2.1 at the time, and showed it generating a character reference sheet in an edit workflow.
The same repository also lists a second LoRA, Qwen2.1_Anything2RealCharacters, which converts images from any art style into realistic human characters. The weights are free to download under Apache-2.0 and load through diffusers or in ComfyUI-style workflows. They require the Qwen/Qwen-Image-2.1 base model.
- LoRA adapter for Qwen-Image-2.1 image editing
- Targets character consistency in anime-style edits
- Trained on 4-view character model sheets and facial expression edits
- Recommended LoRA weight 0.6–0.8
- Apache-2.0 licensed, free to download
- Marked experimental by the author
As seen on YouTube
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Similar AI tools
Closest neighbours by meaning, not just by category.
Krea 2 LoRAs
App · Krea AI
LoRA training on Krea's platform to teach Krea 2 a custom style, character, face or product from a few images.
Qwen Video Edit
Model · Alibaba
Research method for instruction-based video editing by repurposing Qwen-Image-Edit with Wan 2.1 video latents.
MOSS Image Sumi-ink LoRA
Model
Community LoRA for the MOSS Image model that applies an ink-splatter, sumi-ink art style to generated images.
Bespoke Nimble 9B
Model · Bespoke Labs
Apache-2.0 LoRA adapter on Qwen3.5-9B that scores answer choices, booleans and rubric levels against a given context.
Stable Layers
Research · Stability AI
Stability AI research method that fine-tunes Qwen-Image-Layered for image layer decomposition using VLM-scored RL.
Qwen Image 2.1
Model · Alibaba
Alibaba Qwen's image generation and editing model, accessible through Qwen Chat and used as an open editing baseline.