<h1 id="anima_booster-for-comfyui">ANIMA_BOOSTER for ComfyUI ⚡</h1>The ultimate high-performance optimization suite designed to maximize inference speed and optimize the performance of the Anima DiT 2B model.Delivers a massive 3.5× to 5.0× speedup with virtually no loss in visual quality!&gt; 💡 Ultimate Quality &amp; Speed Combo: &gt; We highly recommend pairing this optimization suite with our companion node <a rel="ugc" href="https://github.com/BlackSnowSkill/ComfyUI-BSS_FLSampler">FLSampler (BSS)</a> to perfectly restore any lost micro-details at lightning-fast speeds!🔗 GitHub Repository (Full Guide &amp; Docs): <a target="_blank" rel="ugc" href="https://github.com/BlackSnowSkill/ANIMA_BOOSTER">https://github.com/BlackSnowSkill/ANIMA_BOOSTER</a><hr /><h3 id="key-features-(v1.3.0)">🌟 Key Features (v1.3.0)</h3> 📥 Anima Checkpoint &amp; UNET Loaders: Tailored loaders supporting standalone UNETs and full Checkpoints with built-in bfloat16 auto-detection. ⚡ Safe One-Click torch.compile: Stable JIT compilation toggle boosting speed by 20% to 40%. 🚀 SageAttention: Integrated accelerated 8-bit attention tailored for DiT. 🧠 Adaptive TeaCache: Timestep-aware latent caching that skips redundant calculations, protecting early structural steps. 🎨 BSS Premium UI: Elegant, high-contrast dark-matte interface with clean controls and zero intrusive tooltips.<hr /><h3 id="quick-installation">📂 Quick Installation</h3>1. Open ComfyUI Manager -&gt; Install via Git URL.2. Paste this URL: <code>https://github.com/BlackSnowSkill/ANIMA_BOOSTER</code>3. Click Install and restart ComfyUI.For detailed parameters, portable environment wheels, and manual installation, please refer to the <a rel="ugc" href="https://github.com/BlackSnowSkill/ANIMA_BOOSTER">GitHub README</a><hr />☕ Support &amp; LicenseSupport Development: Support me and get exclusive models on <a rel="ugc" href="https://boosty.to/blacksnowskill">Boosty</a>!License &amp; Usage: © 2026 blacksnowskill (BSS). All rights reserved. This project is protected by copyright. Copying, distribution, merging, or use on other websites/repositories without explicit written permission is strictly prohibited.

### What's New:1. *<code>v1 (Legacy Fast)</code>** (Default) — Restores the aggressive caching behavior that gives you an instant 2.0x speed boost on SDE/stochastic samplers (like <code>er_sde</code>, <code>sde gpu</code>) right out of the box, without needing to touch any other settings!2. *<code>v2 (Standard Precise)</code>** — The mathematically precise, dynamic mode for <code>euler_a</code> protecting the early steps of your image structure.

v1.3

v1.2.1

v1.2.0 - Stability &amp; Performance Update.This major update focuses on robust stability, codebase refactoring, and fixing scaling bugs for stochastic/SDE samplers. We have eliminated unstable components and made the suite bulletproof for everyday generation.🆕 What's New in v1.2.0:1. 🐛 TeaCache Fixed for SDE/Stochastic Samplers (e.g., <code>er_sde</code>, <code>sde gpu</code>)The Issue: Stochastic samplers working on a sigma scale (like <code>[14.6 .. 0.0]</code>) previously confused TeaCache's fixed threshold. This triggered aggressive caching on the very first step, resulting in fast generations but heavily distorted images covered in artifacts.The Solution: We implemented dynamic timestep scale auto-detection <code>st.max_t</code>. TeaCache now mathematically adapts to any sampler and scheduler (sigmas, 1000..0, or 1..0). Early structural steps are fully protected, while late-stage detailing is safely cached. Enjoy perfect image quality with SDE samplers!2. 💎 Safe One-Click JIT Compilation <code>torch.compile</code>)Unstable <code>AnimaTorchCompile</code> node removed: The complex external compilation node was prone to PyTorch crashes (CUDA Graphs tensor overwrite errors) when handling dynamic latent dimensions.Integrated JIT Toggle: We integrated a safe, one-click <code>torch_compile</code> toggle directly into Anima Booster Loader and Checkpoint Loade. It runs on the stable <code>inductor</code> backend (default mode) without CUDA Graphs. Enjoy the same **+20% to +40% speed boost** with **100% stability**!3. 🗑️ Codebase Cleanup*Removed <code>AnimaSparseAttention</code>: Local sparse attention on blocks trained on Full Attention destroyed global image geometry and caused structural artifacts. Removed <code>AnimaTorchCompile</code>: Replaced by the native, stable JIT toggle in the model loaders.The package is now cleaner, lighter, and completely safe.4. 📦 Graceful Degradation &amp; Portable Windows Support All high-performance modules (like SageAttention) are now fully optional. If not installed, the loader will seamlessly fall back to PyTorch's native SDPA without throwing import errors.Windows/Portable Tip: We recommend installing the ComfyUI-Sage-EasyInstall node via ComfyUI Manager to easily fetch precompiled Triton and SageAttention binary wheels.🎛️ Recommended Settings for Maximum Speed &amp; Quality:Anima Booster Loader: Set <code>sage_attention</code> to <code>auto</code> and enable <code>torch_compile</code>. (Note: The first 2-3 generations will have a warm-up phase while PyTorch compiles the blocks).Anima TeaCache: Set <code>threshold</code> to <code>0.15</code> and keep <code>adaptive</code> <code>ON</code>.For SDE Samplers (like <code>er_sde</code>): Now fully compatible and artifact-free! If you want to push the speed further while maintaining great quality, try raising the TeaCache <code>threshold</code> to <code>0.22 - 0.25</code>.

v1.2

What's New🧹 Streamlined Interface Removed non-functional options that caused crashes with Anima DiT models:<ul><li>Removed <code>weight_dtype</code> selector — Anima models are natively bfloat16. Forcing FP8 caused <code>LayerNormKernelImpl not implemented for Float8_e4m3fn</code> crashes because PyTorch does not support FP8 in RMSNorm operations used by DiT architectures. The loader now always uses the optimal native dtype.</li><li>Removed <code>compile_mode</code> selector — Modes <code>reduce-overhead</code> and <code>max-autotune</code> use CUDA Graphs internally, which conflict with Anima's in-place tensor operations, causing <code>CUDAGraphs tensor overwritten</code> errors. The loader now always uses the stable <code>default</code> compilation mode.</li></ul>🔬 SageAttention Fix Fixed a critical bug where selecting specific quantization kernels (<code>sageattn_qk_int8_pv_fp16_cuda</code> / <code>fp8</code>) silently failed due to API changes in <code>sageattention &gt;= 1.0.6</code>. The library migrated from C++ extensions to Triton kernels and removed those functions from the public API. The node now correctly uses the <code>auto</code> mode which delegates to ComfyUI's built-in SageAttention integration — this automatically selects the optimal quantization strategy for your GPU.📦 New: Checkpoint Loader Added <code>Anima Booster Checkpoint Loader</code> — a new node for loading bundled Anima checkpoints that contain Model + CLIP + VAE in a single file. Outputs all three components while applying the same SageAttention and torch.compile optimizations as the standard Anima Booster Loader.⚡ Clean Node Design Both loaders now feature a minimal, crash-proof interface:<ul><li>Model selector</li><li>SageAttention toggle (auto / disabled)</li><li>torch.compile toggle (on / off)</li></ul>No unnecessary options that could lead to errors. Just plug in and generate.

[BSS] - Anima Booster (Custom node)

ANIMA_BOOSTER for ComfyUI ⚡

🌟 Key Features (v1.3.0)

📂 Quick Installation