ANIMA_BOOSTER
🚀 Introducing ANIMA_BOOSTER! Speeding up Anima DiT 2B by 5x (Now on GitHub, Civitai & ComfyUI Manager!)

Hey everyone!
Lately, I’ve been deeply exploring the new Anima DiT 2B model, and honestly, the quality and spatial understanding it delivers are absolutely mind-blowing. However, there was one huge bottleneck that made using it a painful chore: generation speeds were incredibly slow.
I couldn't just stand by and let this amazing model collect dust, so I decided to roll up my sleeves and build a custom solution. Today, I am proud to share ANIMA_BOOSTER with you! It is a high-performance optimization suite that accelerates Anima DiT 2B by 3.5× to 5.0× without any noticeable loss in visual quality.
⚡ What makes ANIMA_BOOSTER a game-changer?
- Integrated Safe JIT Compilation (torch.compile): A stable one-click toggle inside the loaders that compiles transformer blocks on the safe inductor backend, boosting performance by 20-40% without CUDA Graphs crashes.
- SageAttention: Seamless, fast 8-bit attention integration to dramatically cut down VRAM usage and speed up computations.
- Adaptive TeaCache: A smart, dynamic state caching system that automatically adjusts its threshold—keeping early structure-building steps intact for perfect geometry, while aggressively caching late detailing steps.
- BSS Premium UI: I also custom-crafted a high-contrast luxury matte-black and warm gold UI. It is completely clean, responsive, and has all overlapping tooltips suppressed for a premium workspace experience.
🔥 The Ultimate Synergy: ANIMA_BOOSTER + FLS (First Light Sampler)
If you really want to unlock the absolute maximum potential of this model, run it in combination with FLS (First Light Sampler).
Because FLS is unmatched at restoring and enhancing fine details, micro-textures, and overall sharpness, combining it with ANIMA_BOOSTER creates a perfect loop: the booster handles the heavy lifting at lightning speeds, while FLS brings out breathtaking, crystal-clear details. The results are simply unmatched!