Krea 2 Character LoRA One-Click Training in ComfyUI: Zero Code, Fully Localized WebUI
Training a character LoRA for Krea 2 used to mean configuration files and command lines. This episode uses a community one-click trainer with a fully localized WebUI: upload a training set, set a trigger word, and train. The notes below come from the episode subtitles and the published mirror page; no workflow JSON was shipped or parsed, so no node-level parameters are claimed beyond what the episode states.
What a character LoRA is
A LoRA is a small external model that teaches the base model to draw a specific character or style. The episode shows its training set on the left and images generated with the trained LoRA on the right; the author calls the result a success.
The workflow in the WebUI
- Choose the training engine. The default is AI Toolkit, which the author calls the most common tool and mature for Krea 2 Turbo LoRA training.
- Create a project and name it — the author uses the character name.
- Set the trigger word. This is effectively the character's dedicated name and matters a lot.
- Upload the training set and start training.
The author emphasizes there is no code and no configuration: with a training set, someone with no training background can run the one-click flow.
Hardware and deployment
- Recommended training GPU: a 24 GB 4090, which the author calls the best value in time and cost.
- The mirror is on 优云智算: https://www.compshare.cn/images/yhJutoBxdzPr?referral_code=A7fTQUYY9lCDFJwA6MMvoq
- After deployment, open the WebUI and follow the steps above.
Links
- Mirror (with referral code): https://www.compshare.cn/images/yhJutoBxdzPr?referral_code=A7fTQUYY9lCDFJwA6MMvoq
- Video tutorial (YouTube): https://www.youtube.com/watch?v=7EJqgHv2YaU
- Chinese original (Bilibili): https://www.bilibili.com/video/BV1mqgr6sEXS
Limits
- Training results depend on the dataset size, variety and captioning; the episode's success is not a guarantee for every character.
- AI Toolkit and the trainer are community tools; their behaviour may change between versions.
- The mirror is a deployment convenience; third-party tools keep their own licenses.
Top comments (0)