Circlestone Labs releases Anima base v1.0, a new open-source diffusion model
A community-driven model with competitive quality and 2.5B parameters...
Circlestone Labs, a small research collective, has launched Anima base v1.0, a new open-source text-to-image diffusion model available on Hugging Face and Civitai. With 2.5B parameters, the model targets the gap between lightweight models (e.g., SD 1.5) and heavyweight closed ones (e.g., Midjourney). It generates 1024x1024 images with enhanced coherence and style diversity, trained on a curated dataset of 500M high-quality image-text pairs. The model is fully open under an Apache 2.0 license, allowing commercial use and fine-tuning.
Early benchmarks show Anima base v1.0 achieving a FID score of 12.3 on COCO, placing it above SDXL 1.0 (12.8) and close to DALL-E 3 (unreleased scores). The community on Civitai has already contributed 50+ fine-tuned variants within 24 hours of release. This model offers a viable alternative for developers and artists who want control without sacrificing quality. Its release could accelerate open-source image generation adoption in professional workflows.
- 2.5B parameter diffusion model supporting 1024x1024 resolution
- FID score of 12.3 outperforms SDXL 1.0 (12.8) on COCO benchmark
- Apache 2.0 license allows commercial use and community fine-tuning
Why It Matters
Anima base v1.0 gives professionals a high-quality, open-source alternative to closed models, reducing costs and enabling customization.