| --- |
| license: other |
| tags: |
| - comfyui |
| - quantization |
| - int4 |
| - convrot |
| - diffusion |
| - image-generation |
| - video-generation |
| - upscaling |
| - krea |
| - ltx |
| - seedvr2 |
| --- |
| |
| # INT4 ConvRot Comfy Models — Winnougan |
|
|
|  |
|
|
| A collection of INT4 ConvRot-quantized diffusion, video, and upscaling models for ComfyUI, built to run comfortably on 8GB-class GPUs (developed and tested on an RTX 3070 Ti) without gutting output quality. |
|
|
| ## What's in this repo |
|
|
| | Model | Type | Notes | Quant | |
| |---|---|---|---| |
| | **Krea 2 Raw** | Image diffusion | Base Krea 2 checkpoint, unmodified pipeline | INT4 convrot | |
| | **Krea 2 Turbo** | Image diffusion | Distilled/turbo variant, fewer steps | INT4 convrot | |
| | **LTX-2.3 1.1 Distilled** | Video diffusion | Distilled LTX-2.3 build | INT4 convrot | |
| | **Sulphur 2 Base** | Video diffusion | Base checkpoint built off of LTX-2.3 | INT4 convrot | |
| | **SeedVR2 (7B)** | Image and Video Upscaler | Full 7B variant | INT4 convrot | |
|
|
| All models are quantized to **INT4** using **Starnodes Model Converter** (https://github.com/Starnodes2024/comfyui-starnodes-modelconverter) |
|
|
| ## Why ConvRot INT4 |
|
|
| Standard INT8/INT4 row-wise quantization throws away a lot of precision on the weight matrices that matter most for visual fidelity. ConvRot groups weights along their largest power-of-4-compatible dimension before quantizing, which keeps much more of the original model's detail and reduces the artifacting you'd normally see from a naive INT4 cast. The trade-off is VRAM and disk savings big enough to run models like SeedVR2 7B and full video diffusion checkpoints on 8GB cards. |
|
|
| ## Requirements |
|
|
| - ComfyUI (nighlty build) |
| - If you're getting chronic errors update your Conda environment (I'm running Pytorch 2.12, cu132, Python 3.12, Flashattention/Sageattention and Triton 3.8) |
|
|
| ## Installation |
|
|
| 1. Install `ComfyUI-INT4-Fast` into `ComfyUI/custom_nodes/` |
| 2. Download the model(s) you want from this repo into the matching `ComfyUI/models/diffusion_models/` (or appropriate folder for video/upscale models) |
| 3. Load with the INT4 loader node from ComfyUI-INT4-Fast — do not use the standard checkpoint/UNETLoader nodes, they will not decode these correctly |
| 4. See the `Samples and Workflow` folder in this repo for ready-to-use ComfyUI workflow JSONs and sample outputs |
|
|
| ## Quantization pipeline |
|
|
| Built with Starnodes power: |
|
|
| Grab the Starnodes model converter and do it yourself if you wish. It supports INT8 and INT4 convrot: |
| [Starnodes](https://github.com/Starnodes2024/comfyui-starnodes-modelconverter) |
|
|
|
|
| ## Links |
|
|
| - 🎥 YouTube: tutorials and walkthroughs for this collection |
| - 💬 Discord: community, support, and early access |
| - 🩷 Patreon / ☕ Ko-fi: support ongoing quantization work |
| - 🤗 More models: [huggingface.co/Winnougan](https://huggingface.co/Winnougan) |
|
|
| ## License |
|
|
| Inherits the license terms of each respective base model (Krea 2, LTX-2.3, Sulphur 2, SeedVR2). Check each upstream model's license before commercial use. |
|
|
| --- |
|
|
| *Part of the âš¡ Winnougan quantization series.* |
