Upload README.md with huggingface_hub

README.md

@@ -8,19 +8,16 @@ tags:
 - controlnet
 - tensorrt
 - rtx
-- convrot
-- int8
-- quantized
 - comfyui
 pipeline_tag: image-to-image
 ---
 
-# CCSR: ConvRot INT8 & TensorRT RTX Acceleration
+# CCSR: TensorRT RTX Acceleration Engine
 
-High-fidelity **ConvRot INT8** quantized weights and ultra-fast **TensorRT RTX** execution engines for **CCSR (Creative Content Super-Resolution)**, designed for low-VRAM and real-time generative image upscaling in **ComfyUI**.
+Ultra-fast **TensorRT RTX** execution engine and auxiliary modules for **CCSR (Creative Content Super-Resolution)**, designed for real-time generative image upscaling in **ComfyUI**.
 
 > 📦 **ComfyUI Loader & Upscaler Extension:**
-> All nodes supporting both ConvRot INT8 and TensorRT execution are hosted in:
+> All nodes supporting TensorRT engine execution are available in:
 > 👉 **[https://github.com/ussoewwin/ComfyUI-NunchakuFluxLoraStacker](https://github.com/ussoewwin/ComfyUI-NunchakuFluxLoraStacker)**
 
 ---
@@ -29,28 +26,24 @@ High-fidelity **ConvRot INT8** quantized weights and ultra-fast **TensorRT RTX**
 
 Creative Content Super-Resolution (CCSR) is a diffusion-based super-resolution framework leveraging a Controlled UNet and ControlNet structure to synthesize rich photorealistic textures and fine details.
 
-This repository provides two high-performance deployment options for CCSR:
+This repository provides an optimized **NVIDIA TensorRT RTX Engine** implementation for CCSR:
 
-1. **ConvRot INT8 Checkpoint (`real-world_ccsr_convrot_int8.safetensors`)**:
-   - Applies orthogonal Hadamard rotation matrices to UNet linear layers to eliminate activation outlier spikes prior to per-channel INT8 quantization.
-   - Quantizes ControlNet convolutional layers to INT8 Conv2d.
-   - Preserves high-precision FP16 weights for VAE autoencoders and condition encoders.
-   - Cuts checkpoint size from **3.2 GB to ~2.0 GB** and saves **~1.1 GiB VRAM** while maintaining full super-resolution quality.
-
-2. **TensorRT RTX Engine (`ccsr_apply_f16io.rtxplan` + `ccsr_trt_aux.safetensors`)**:
-   - Fuses the ControlNet and Controlled UNet denoising computation into an optimized NVIDIA TensorRT RTX engine.
-   - Fixed 512px tile resolution (64×64 latent tile) executing at ~**24 ms/step** (~**4.7× speedup** over PyTorch FP16 at ~113 ms/step on modern RTX GPUs).
-   - Engine-only operation via companion auxiliary weights (`ccsr_trt_aux.safetensors`, VAE + condition encoder), requiring no full checkpoint download.
+- **Fused Denoising Engine (`ccsr_apply_f16io.rtxplan`)**:
+  - Fuses the ControlNet and Controlled UNet denoising computation into a single compiled TensorRT engine.
+  - Fixed 512px tile resolution (64×64 latent tile) executing at ~**24 ms/step** (~**4.7× speedup** over PyTorch FP16 at ~113 ms/step on modern RTX GPUs).
+  - Synchronized stream execution on current PyTorch CUDA streams to eliminate race conditions and deadlocks.
+- **Engine-Only Deployment (`ccsr_trt_aux.safetensors`)**:
+  - Contains only the essential companion modules: FP16 AutoencoderKL (VAE encoder/decoder) and condition encoder.
+  - Automatically loaded alongside the engine, eliminating the need to download large full checkpoints (~3.2 GB saved).
 
 ---
 
 ## 📦 Available Files
 
-| Filename | Description | Quantization / Engine | File Size | Recommended Location | License |
+| Filename | Description | Architecture / Components | File Size | Recommended Location | License |
 | :--- | :--- | :--- | :--- | :--- | :--- |
-| `real-world_ccsr_convrot_int8.safetensors` | Full CCSR Checkpoint | ConvRot INT8 (UNet/ControlNet) + FP16 (VAE/Encoder) | ~2.0 GB | `ComfyUI/models/checkpoints/` or `models/CCSR/` | Apache-2.0 |
-| `ccsr_apply_f16io.rtxplan` | TensorRT Fused Denoising Engine | TensorRT RTX Engine (Tile 512px / Latent 64×64) | ~1.4 GB | `custom_nodes/.../nodes/CCSR/trt_engines/` | Apache-2.0 |
-| `ccsr_trt_aux.safetensors` | Auxiliary Weights for TRT Engine | FP16 VAE AutoencoderKL + Condition Encoder | ~450 MB | `custom_nodes/.../nodes/CCSR/trt_engines/` | Apache-2.0 |
+| `ccsr_apply_f16io.rtxplan` | TensorRT Fused Denoising Engine | ControlNet + UNet fused RTX Engine (Tile 512px / Latent 64×64) | ~1.4 GB | `custom_nodes/.../nodes/CCSR/trt_engines/` | Apache-2.0 |
+| `ccsr_trt_aux.safetensors` | TRT Auxiliary Weights | FP16 VAE AutoencoderKL + Condition Encoder | ~450 MB | `custom_nodes/.../nodes/CCSR/trt_engines/` | Apache-2.0 |
 
 ---
 
@@ -58,53 +51,48 @@ This repository provides two high-performance deployment options for CCSR:
 
 Measurements conducted on an NVIDIA RTX 4090 / RTX 5090 environment:
 
-| Execution Mode | Checkpoint / Engine Size | VRAM Overhead (Denoising) | Step Latency (Tile 512) | Notes |
+| Execution Mode | Files Required | VRAM Overhead (Denoising) | Step Latency (Tile 512) | Speedup |
 | :--- | :--- | :--- | :--- | :--- |
-| **Stock CCSR (FP16 PyTorch)** | ~3.2 GB | ~3.8 GiB | ~113 ms / step | Baseline PyTorch execution |
-| **CCSR ConvRot INT8** | ~2.0 GB | ~2.7 GiB (**~1.1 GiB saved**) | ~105 ms / step | Quantized weights kept in VRAM, full fidelity |
-| **CCSR TensorRT RTX** | ~1.4 GB (+ 0.45 GB aux) | ~2.2 GiB | ~**24 ms / step** (**~4.7× faster**) | ControlNet + UNet fused TRT engine |
+| **Stock CCSR (FP16 PyTorch)** | Full Checkpoint (~3.2 GB) | ~3.8 GiB | ~113 ms / step | 1.0× (Baseline) |
+| **CCSR TensorRT RTX** | Engine + Aux (~1.85 GB total) | ~2.2 GiB | ~**24 ms / step** | **~4.7× faster** |
 
 ---
 
 ## 🚀 Usage in ComfyUI
 
-Support for both **ConvRot INT8** and **TensorRT Engine** execution is integrated natively into the dedicated loader extension:
-
-* **ComfyUI Extension / Loader Repository:** [https://github.com/ussoewwin/ComfyUI-NunchakuFluxLoraStacker](https://github.com/ussoewwin/ComfyUI-NunchakuFluxLoraStacker)
-
-### Installation
-
-Clone the extension repository into your `ComfyUI/custom_nodes` directory:
-
-```bash
-cd ComfyUI/custom_nodes
-git clone https://github.com/ussoewwin/ComfyUI-NunchakuFluxLoraStacker.git
-```
-
-### Method 1: ConvRot INT8 (PyTorch Execution)
-
-<p align="center">
-  <img src="https://huggingface.co/ussoewwin/CCSR-ConvRot-INT8-and-TensorRT-Engine/resolve/main/png/ccsr.png" alt="CCSR PyTorch / ConvRot INT8 Workflow in ComfyUI" width="800">
-</p>
-
-1. Place `real-world_ccsr_convrot_int8.safetensors` in:
-   - `ComfyUI/models/checkpoints/` (for `CCSR_Model_Select`) or
-   - `ComfyUI/models/CCSR/` (for `DownloadAndLoadCCSRModel`)
-2. In ComfyUI, add **`CCSR_Model_Select`** or **`DownloadAndLoad CCSRModel`**. The node automatically detects the `comfy_quant` INT8 markers and activates quantized ops (`QuantConv2d` & `mixed_precision_ops`).
-3. Connect the output `ccsr_model` to **`CCSR_Upscale`**.
-4. Configure upscale parameters (`scale_by`, `tile_size=512`, `tile_stride=256`, color correction `adain`/`wavelet`).
+TensorRT engine execution for CCSR is integrated natively into the **[ComfyUI-NunchakuFluxLoraStacker](https://github.com/ussoewwin/ComfyUI-NunchakuFluxLoraStacker)** custom-node pack.
 
-### Method 2: TensorRT RTX Engine (Ultra-Fast Execution)
+### Workflow Example
 
 <p align="center">
-  <img src="https://huggingface.co/ussoewwin/CCSR-ConvRot-INT8-and-TensorRT-Engine/resolve/main/png/ccsrtensor.png" alt="CCSR TensorRT RTX Engine Workflow in ComfyUI" width="800">
+  <img src="https://huggingface.co/ussoewwin/CCSR-TensorRT-Engine/resolve/main/png/ccsrtensor.png" alt="CCSR TensorRT RTX Engine Workflow in ComfyUI" width="800">
 </p>
 
-1. Place both `ccsr_apply_f16io.rtxplan` and `ccsr_trt_aux.safetensors` into:
-   - `ComfyUI/custom_nodes/ComfyUI-NunchakuFluxLoraStacker/nodes/CCSR/trt_engines/`
-2. In ComfyUI, add **`Load CCSR Model (TensorRT)`** (`LoadCCSRModelTensorRT`). The node populates the dropdown with available `.rtxplan` engines and automatically loads the companion `ccsr_trt_aux.safetensors`.
-3. Connect the `ccsr_model` output to **`CCSR Upscale (TRT)`** (`CCSR_Upscale_TRT`).
-4. Execute upscaling at tile size 512. The engine processes tiles at ~24 ms/step with stream-synchronized CUDA execution.
+### Installation & Setup
+
+1. **Install the Custom Node Pack**:
+   ```bash
+   cd ComfyUI/custom_nodes
+   git clone https://github.com/ussoewwin/ComfyUI-NunchakuFluxLoraStacker.git
+   ```
+
+2. **Place Engine & Aux Files**:
+   Download both `ccsr_apply_f16io.rtxplan` and `ccsr_trt_aux.safetensors` and place them directly into the engine directory:
+   ```
+   ComfyUI/custom_nodes/ComfyUI-NunchakuFluxLoraStacker/nodes/CCSR/trt_engines/
+   ├── ccsr_apply_f16io.rtxplan
+   └── ccsr_trt_aux.safetensors
+   ```
+
+3. **In ComfyUI**:
+   - Add **`Load CCSR Model (TensorRT)`** (`LoadCCSRModelTensorRT`). The node automatically discovers `.rtxplan` files in `trt_engines/` and loads the companion `ccsr_trt_aux.safetensors`.
+   - Connect the `ccsr_model` output to **`CCSR Upscale (TRT)`** (`CCSR_Upscale_TRT`).
+   - Connect an input image to `image`.
+   - Configure upscale parameters:
+     - `tile_size`: 512 (fixed to match the compiled static engine shape)
+     - `tile_stride`: 256 (recommended for seamless blending)
+     - `color_fix_type`: `adain` (or `wavelet` / `none`)
+     - `steps`: Effective diffusion step count (densified schedule guarantees exact execution of requested step count)
 
 ---
 
@@ -113,5 +101,5 @@ git clone https://github.com/ussoewwin/ComfyUI-NunchakuFluxLoraStacker.git
 - **Original CCSR Implementation & Weights**: [csslc/CCSR](https://github.com/csslc/CCSR) (Apache-2.0 License)
 - **Research Paper**: *"Creative Content Super-Resolution with Pre-trained Diffusion Model"* by Liang et al.
 - **ComfyUI CCSR Node Foundation**: [kijai/ComfyUI-CCSR](https://github.com/kijai/ComfyUI-CCSR) & [Kijai/ccsr-safetensors](https://huggingface.co/Kijai/ccsr-safetensors) (Apache-2.0 License)
-- **ComfyUI Loader & TRT Integration**: [ussoewwin/ComfyUI-NunchakuFluxLoraStacker](https://github.com/ussoewwin/ComfyUI-NunchakuFluxLoraStacker) (CCSR nodes located under `nodes/CCSR/`)
+- **TensorRT Engine & ComfyUI Loader**: [ussoewwin/ComfyUI-NunchakuFluxLoraStacker](https://github.com/ussoewwin/ComfyUI-NunchakuFluxLoraStacker) (CCSR nodes located under `nodes/CCSR/`)
 - **License**: **Apache-2.0**