Analysis updated 2026-05-18
Suppress specific visual artifacts like blurriness or low quality directly inside a Krea 2 prompt without using a separate negative prompt field
Fine-tune which transformer blocks inside the Krea 2 model are affected by the negative weighting for more targeted image control
Integrate NegPiP-style prompt control into an existing ComfyUI Krea 2 workflow with minimal node changes
| blue-pen5805/comfyui-krea2-negpip | 0xh4ku/manga-pdf-to-epub | ayyouboss0011/sherlockmaps | |
|---|---|---|---|
| Stars | 61 | 60 | 60 |
| Language | Python | Python | Python |
| Setup difficulty | moderate | moderate | moderate |
| Complexity | 2/5 | 2/5 | 3/5 |
| Audience | general | general | data |
Figures from each repo's GitHub metadata at analysis time.
Requires ComfyUI already installed with a Krea 2 model and a krea2-type CLIP loaded via CLIPLoader.
ComfyUI is a popular node-based interface for running AI image generation models locally. This repository is a small custom extension (called a "custom node") for ComfyUI that adds a specific capability called NegPiP-style negative prompt weighting to workflows that use the Krea 2 image model. In most AI image generators, you can write a "negative prompt" to tell the model what you do not want in the image. NegPiP takes this further: instead of a separate negative prompt box, you embed negative weights directly inside your regular prompt text. For example, writing (blurry:-1.0) inside a positive prompt tells the model to suppress blurriness while still describing what you do want. This gives you finer control over the final image without needing a separate negative prompt field. The extension adds a single new node called Apply Krea2 NegPiP to ComfyUI's visual workflow editor. You place this node in your graph between the model and text encoder steps. It intercepts the Krea 2 model and its CLIP text encoder, applies the negative-weight patches internally, and returns modified versions that the rest of your workflow uses as normal. Installation is straightforward: clone the repository into ComfyUI's custom_nodes folder and restart the application. No extra configuration is needed beyond that. The node exposes several tuning parameters: you can adjust the overall strength of the negative effect, choose which internal transformer blocks inside the model to patch (with start, end, and stride controls), and enable an optional stronger effect path. The README notes that this node only works with Krea 2 model layouts and may not correctly apply negative weights to tokens that follow image embedding placeholders in a prompt. The project is quite new (created June 2026) and borrows its core idea from the original NegPiP plugin for Stable Diffusion web UI.
A ComfyUI custom node that lets you write negative weights like (blurry:-1.0) directly inside your prompt text when using Krea 2 image generation models, giving fine-grained suppression control without a separate negative prompt box.
Mainly Python. The stack also includes Python, ComfyUI, PyTorch.
Setup difficulty is rated moderate, with roughly 5min to a first successful run.
Mainly general.
This repo across BitVibe Labs
Verify against the repo before relying on details.