Details
The on-site LoRA Trainer is a web wizard that guides users through uploading a dataset of images, captioning them (manually or via an AI auto-captioning feature powered by the JoyCaption model), and setting training parameters. Civitai's cloud infrastructure runs the training process, producing a LoRA file the user can then download or use directly in the on-site generator. Supported base models include SD 1.5, SDXL, and Flux. Training costs Buzz, with pricing scaled by model type and training complexity. The trainer is open to all registered users. A natural language auto-captioning feature, which automatically describes dataset images using a vision-language model, was added in September 2024.
Have evidence about Civitai's AI practices? Submit a report.
Report a Sighting →