April 18, 2025 — The open-source project AI-Toolkit, developed by ostris, continues to evolve as a powerful command-line solution for AI-powered workflows. After proving highly effective for Flux.1 LoRA training, the toolkit has now introduced support for HiDream-I1 LoRA training, marking another major step forward for creators focused on photorealistic image generation.
The update enables fine-tuned control over HiDream-I1, a model known for generating ultra-realistic and cinematic outputs. Early tests with Flux.1 LoRA training have already shown impressive results using the toolkit’s streamlined workflow system. With the HiDream-I1 integration, users can begin experimenting with personal styles, character consistency, or dataset-specific adjustments using lightweight LoRA methods.
Currently, HiDream-I1 LoRA training is only supported on GPUs with 48GB of VRAM, but support for 24GB VRAM hardware is expected in future updates. This limitation is due to the model’s size and memory requirements, but the roadmap suggests greater accessibility is on the horizon.
With continued improvements, including support for emerging models and training formats, AI-Toolkit is positioning itself as an essential tool for AI enthusiasts, researchers, and developers.
Explore the project on GitHub.
Leave a Reply