Save 4 hours! We auto-setup your workflow! Free!

Drop your workflow.json — we handle every dependency, custom node, and model. Just open the link and run.

Auto-Setup Workflow Json (Free) Now!
ComfyUI > Nodes > ComfyUI Krea Reference

ComfyUI Extension: ComfyUI Krea Reference

Repo Name

krea-reference

Author
Kevin Gilper (Account age: 4646 days)
Nodes
View all nodes(1)
Latest Updated
2026-10-29
Github Stars
0.02K

How to Install ComfyUI Krea Reference

Install this extension via the ComfyUI Manager by searching for ComfyUI Krea Reference
  • 1. Click the Manager button in the main menu
  • 2. Select Custom Nodes Manager button
  • 3. Enter ComfyUI Krea Reference in the search bar
After installation, click the Restart button to restart ComfyUI. Then, manually refresh your browser to clear the cache and access the updated list of nodes.

Visit ComfyUI Online for ready-to-use ComfyUI environment

  • Free trial available
  • 16GB VRAM to 80GB VRAM GPU machines
  • 400+ preloaded models/nodes
  • Freedom to upload custom models/nodes
  • 200+ ready-to-run workflows
  • 100% private workspace with up to 200GB storage
  • Dedicated Support

Run ComfyUI Online

ComfyUI Krea Reference Description

ComfyUI Krea Reference guides Krea 2 image generation with role-based reference images and text-defined Concept Sliders, letting you adjust visual attributes and save controls in reusable workflows.

krea-reference Introduction

krea-reference is a set of ComfyUI tools for guiding Krea 2 image generation. It gives you two main ways to steer an image: use reference pictures to show what you want, or use text-defined sliders to adjust a visual attribute. You can save these controls in a ComfyUI workflow and return to them when you want to try another variation.

With reference cards, you can give each input image a job—for example, “use this for the subject” and “use that for the visual style.” With Concept Sliders, you can describe an attribute and adjust its strength without training a slider model or adding separate slider weights.

V11 is the recommended starting point. It is available in package version 0.5.0 or newer. Existing V9, V10, and Concept Slider V1 workflows keep their original behavior. Experimental V12 and V13 features require current GitHub source and additional compatible host dependencies; they are not included in registry package 0.5.0.

V11 reference-guided teapot example

This V11 example uses separate synthetic product and graphic references to guide subject and style. The illustrated gallery shows its inputs, settings, and limitations.

How krea-reference Works

Think of a prompt as your written brief and a reference card as a picture you attach to that brief. A card tells the workflow which part of the picture to pay attention to: its subject, style, lighting, layout, or even something you want to use as a counter-example. The V11 reference stack brings those cards together before image generation.

Load Image → Image Guide Card V11 → Reference Stack Encoder V11 → sampler

A Concept Slider works differently. Instead of supplying an image, you describe an attribute using text and paired endpoints. The slider card turns that description into a control you can move toward one end or the other. For example, a brightness slider can make a small adjustment near neutral or push more strongly toward a brighter result.

Concept Slider Card V11 → Concept Slider Stack V11 → sampler

V11 reference and slider workflows use separate encoders. Start with the example workflow for the kind of control you want rather than assuming that their cards connect to the same stack. These controls influence image generation; they do not guarantee that only one part of the image will change.

krea-reference Features

Reference cards: give images distinct roles

V11 accepts up to 12 reference cards in its reference stack. You can assign roles such as subject, visual style, lighting, or layout. This is useful when one picture has the object you want but another has the look you prefer.

Signed balancing limits how far reference influences can be pushed in either direction. The stack also uses caching that accounts for content and revisions, helping it avoid repeating some work while you tune a workflow. Established recipes, including the balanced recipe, retain their settings.

For a practical starting point, open the Native V11 Reference starter workflow. You will need to supply its input images.

Concept Sliders: explore variations without slider training

V11 accepts up to 8 Concept Slider cards. You define an attribute in text, then adjust its value to explore different results. No separately trained slider weights are required.

The slider stack offers:

  • Automatic or manual combined budgeting: controls how the influence of several cards is managed together. The current showcase workflow uses automatic reach and budget, where card values determine the effective controls up to a combined limit. This automatic scaling requires a GitHub source update newer than registry package 0.5.0; see the automatic scaling guide before using or updating that showcase.
  • Duplicate and opposite cancellation: helps manage cards whose descriptions repeat or counteract one another.
  • Per-card timing: lets you set when an individual card applies during generation.
  • Height and crowd presets: provide starting configurations for those attributes.
  • A smoother transition near zero: makes very small slider values stay closer to the neutral result.

The comparison below shows why slider strength matters:

Plain promptTiny V11 adjustmentStronger V11 adjustment
Plain prompt baselineBrightness plus 0.05 in V11Brightness plus 3 in V11

These images use the same prompt and seed, with brightness values of 0, +0.05, and +3. The tiny change stays closer to neutral. A stronger value can also affect shape or framing, so treat slider numbers as creative controls—not measured brightness units.

You can explore the controls in the Native V11 Concept Slider showcase workflow. Unlike the reference starter, it does not need input images.

What to expect from the controls

Reference borders, fine-texture transfer, and attributes that are closely tied to other visual features remain limitations. If a change affects more than you intended, compare it with a neutral result and try a smaller slider value or a simpler set of references. The nine worked examples and V11 gallery provide concrete settings and visible results.

krea-reference Models

krea-reference does not include model weights. Its V11 workflows use your installed Krea 2 diffusion model, Krea Qwen3-VL encoder, and Qwen image VAE files. The extension’s reference cards and Concept Sliders are ways of guiding that setup; they are not a collection of separately trained slider models.

The V13 illumination study below used a specific combination—Krea2 Turbo INT8 ConvRot with General Edit V2—for its demonstrated result. That comparison does not establish that other model combinations or editing tasks will behave the same way. For workflow file selection and connections, use the V11 user guide; for the experimental native example, consult the V13 requirements and setup.

What's New with krea-reference

  • V11 — recommended for everyday starting workflows: introduces the current reference-card and text-defined Concept Slider workflows. Reference cards can have distinct jobs, while sliders add combined budgeting, cancellation of duplicate or opposite controls, per-card timing, presets, and finer behavior near zero. V11 is available in package 0.5.0 or newer.
  • V11 automatic scaling update — current GitHub source: the slider showcase now uses automatic reach and budget, with card values setting effective controls up to a combined limit. This particular feature requires a source update after registry 0.5.0. The V11 guide’s migration examples explain the change.
  • V12 — experimental native studies: adds native context and consumer work used by the later V13 example. It also offers an optional pinned Krea2T enhancer integration, OFF by default. The enhancer was on in the V13 demonstration, but its standalone quality benefit and usefulness across other tasks have not been established.
  • V13 — verified native illumination comparison: demonstrates both dimming and brightening of one synthetic scene while preserving its objects and arrangement. This is an opt-in native editing study, not the same mechanism as V11’s conditioning-based sliders. It is available from current GitHub source, not a new registry release. Other concepts, intermediate strengths, and combinations of controls have not been generally calibrated.
  • Earlier versions: V9, V10, and Concept Slider V1 workflows retain their original behavior. Saved workflows are not automatically migrated. Their guides and historical examples remain in the prior-version archive.

V13 illumination example

Dim: coefficient −1Neutral: 0Bright: coefficient +1
V13 dim illuminationNative neutral illuminationV13 bright illumination

These are the actual rendered image pixels from one full-frame comparison; private PNG workflow metadata was removed. The images share a source, preservation prompt, and seed. The documented run used Krea2 Turbo INT8 ConvRot, General Edit V2 at strength 1, reference boost 4, enhancer on at strength 1, eight Euler steps, a simple schedule, and CFG 1. The effect was visible both before and after aspect restoration. For the precise setup, instruction endpoints, and limits of the claim, see the verified signed illumination comparison.

Troubleshooting krea-reference

A starter workflow opens, but its images or models are missing. What should you check?

  1. For the reference starter, supply your own input images. The slider showcase does not require reference images.
  2. Select the Krea 2 diffusion, Krea Qwen3-VL encoder, and Qwen image VAE files available in your ComfyUI setup. The extension does not ship those weights.
  3. Follow the workflow model and image setup instructions and the V11 user guide.

A V11 slider result changes more than the intended attribute. Is something wrong?
Not necessarily. Stronger values can alter shape or framing as well as the named attribute. Compare against a neutral result, then try a smaller value. The brightness comparison above shows how a tiny adjustment differs from a working-range adjustment. Sliders are creative guides, not physically calibrated measurements.

The automatic reach and budget controls do not match the slider showcase. Why?
That showcase uses an update newer than registry package 0.5.0. Check the automatic reach and budget migration guidance and make sure the workflow and source version you are using belong together.

A V12 or V13 native workflow is unavailable or rejects the current setup. What should you do?

  1. Confirm that you are using current GitHub source; registry package 0.5.0 does not include V12/V13.
  2. Check the native host requirements. Active native use checks specific dependency source revisions and rejects unsupported combinations.
  3. Follow the V13 setup and first-comparison instructions. Do not assume that the verified illumination result extends to other editing concepts or model combinations.

Do you need Photo Studio or trained slider files?
No. The linked V11 starters are native ComfyUI workflows, and V11 Concept Sliders do not need extra trained slider weights.

Will an older saved workflow change to V11 automatically?
No. Older workflows keep their original behavior and are not migrated automatically. Use the prior-version guides when working with them.

Learn More about krea-reference

ComfyUI Krea Reference Related Nodes

RunComfy
Copyright 2025 RunComfy. All Rights Reserved.

RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Models, enabling artists to harness the latest AI tools to create incredible art.

ComfyUI Krea Reference detailed guide | ComfyUI