From 077c0201add71cdd54d06ec41a3109d554fa3e98 Mon Sep 17 00:00:00 2001 From: SOLRICKS Date: Thu, 12 Nov 2026 01:19:55 +0300 Subject: [PATCH] add GB10-Attn Grace Blackwell attention optimization Updated project title and added GB10-Attn section with features, usage, and compatibility details. --- README.md | 35 +++++++++++++++++++++++++++++++++-- 1 file changed, 33 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index 30e2d18..d70cf8e 100644 --- a/README.md +++ b/README.md @@ -1,4 +1,4 @@ -# ComfyUI Video Anti-Aliasing & MiniMax H3 Upscaler +# SOLRICKS — ComfyUI Video & AI Optimization Tools

ComfyUI Manager Verified Release @@ -10,7 +10,7 @@ > Available in **ComfyUI Manager**. Search for **SOLRICKS**. -Anti-aliasing, detail refinement, cinematic effects, and MiniMax H3 upscaling for ComfyUI image and video workflows. VideoTAADLAA combines temporal anti-aliasing, jittered sampling, and DLAA-inspired refinement for cleaner, more stable edges. VideoDetailRefiner adds a lightweight detail and texture refinement pass for image and video outputs. VideoAdaptiveAA provides a lightweight edge-focused cleanup pass for aliasing-prone regions. +Anti-aliasing, detail refinement, cinematic effects, and MiniMax H3 upscaling for ComfyUI image and video workflows. VideoTAADLAA combines temporal anti-aliasing, jittered sampling, and DLAA-inspired refinement for cleaner, more stable edges. VideoDetailRefiner adds a lightweight detail and texture refinement pass for image and video outputs. --- @@ -83,6 +83,37 @@ Temporal chunking and spatial tiling limit per-pass model work. The full source --- +## GB10-Attn + +**GB10-Attn** is a lightweight ComfyUI attention optimization node designed for NVIDIA GB10 Grace Blackwell systems, including Dell Pro Max workstations powered by the GB10 Superchip. + +It integrates with ComfyUI's native Sol-Attn sparse attention backend to reduce attention computation overhead in compatible AI image and video generation models. + +### Features + +- **Grace Blackwell optimization** — Designed for NVIDIA GB10 GPUs (SM121). +- **Native sparse attention** — Uses ComfyUI's existing optimized attention backend. +- **Performance profiles** — Quality, Balanced, Speed, and Dense Reference. +- **Automatic GPU detection** — Identifies compatible hardware and supported attention backends. +- **Safe fallback** — Preserves the original model when optimization is unavailable during setup. +- **General model support** — Designed for compatible ComfyUI transformer architectures, including MiniMax H3. + +### Usage + +Insert GB10-Attn into the MODEL connection before sampling: + +`Load Diffusion Model → GB10-Attn → LoRA / Sampler` + +**Recommended configuration:** Balanced profile, GB10 only (SM121), Strict disabled. + +Avoid combining GB10-Attn with other attention replacement patches on the same model. + +**Compatibility:** Requires a recent ComfyUI version with native Sol-Attn support and a compatible CUDA environment. + +**Development status:** Experimental. Performance improvements depend on the model, sequence length, and hardware. GB10-specific benchmarks and visual quality validation are still pending. + +--- + ### Anti-Aliasing Models The pack includes two 1x refinement models: