The V11 files below are native ComfyUI canvas workflows, with nodes, links, widget values and canvas layout. They are not Photo Studio imports and do not need Photo Studio to run.

Install Krea Reference 0.5.0 or newer, restart ComfyUI and refresh the page.
Download V11 Reference starter — raw JSON or V11 Concept Sliders — raw JSON. On a GitHub file page, use Download raw file; do not save the HTML page as JSON.
Drag the downloaded JSON onto the ComfyUI canvas, or use ComfyUI's workflow Open command.
Select installed Krea model files in the diffusion-model, text-encoder and VAE loaders. The saved examples use krea2/krea2_turbo_nvfp4.safetensors, krea2/qwen3vl_4b_fp8_scaled.safetensors and krea2/qwen_image_vae.safetensors. Folder separators vary by host. A compatible alternative model must be selected explicitly if those files are absent.
For the reference starter, copy the repo's example_assets/krea-reference-examples folder into ComfyUI/input/krea-reference-examples, or choose your own files in all three Load Image nodes. The assets are available in the GitHub repository; registry packages may not include them. Even the zero-strength card's connected Load Image needs a valid file. The slider showcase requires no input images.
Queue the workflow. Defaults are 1024×1536, eight Euler/simple steps and CFG1. The slider showcase also renders a plain comparison branch, so it runs two samplers. View the image previews and the stack reports.

Every Krea guide card, reference stack, concept card and concept stack in these V11 files is a V11 class, including negative/comparison branches. Their JSON version: 0.4 denotes the ComfyUI canvas format, not the Krea node version. Standard Comfy nodes and shared socket types keep their usual names.

V11 opt-in examples

Reference starter: the established reference topology with V11 classes; balance starts off. Replace the sample image filenames and select your installed models.
Concept Slider showcase: six cards, combined budget 2, per-card presets/timing and a same-seed plain comparison. Start with one nonzero card. Near-zero values blend toward the plain prompt; larger changes can still reframe the subject.

See the V11 guide. Existing example files remain available unchanged.

This folder contains the Concept Slider, V10, and V9 starter workflows. The Load Image nodes use the included synthetic example filenames; copy ../example_assets/krea-reference-examples/ into your ComfyUI input folder, or replace the Load Image nodes with your own test images before queueing. (The Concept Slider workflow needs no images at all.)

The examples intentionally avoid LoRA and model-enhancer plumbing so the graph stays focused on Krea plus the reference and slider nodes.

Concept Slider V1 Showcase

File: krea-slider-v1-showcase-workflow.json

Training-free attribute sliders in one graph, with a walkthrough note on the canvas. Six slider cards feed the main stack:

brightness at +3 and fog density at +2 - auto poles: the stack derives more-vs-less from the attribute name alone.
height at -4 with custom poles - the render-proven concrete wording for a weak direction.
realism at 0 with custom style poles, ready to drag: negative goes cartoon, positive goes photoreal.
age and color saturation parked at 0 - a zero slider is skipped entirely and costs nothing.

Plus the feedback and comparison wiring:

slider_report -> Preview Any: each slider's exact pole sentences, its computed push, and everything skipped and why.
A second stack on the same prompt and seed with zero cards, so one queue renders the WITH-vs-WITHOUT pair side by side.
The negative prompt runs through a third slider stack with no cards - a cardless slider stack behaves exactly like CLIP Text Encode.

Queue once, compare the pair, then drag values; with study reuse on, re-runs only re-encode the sliders you changed. The full manual - the dial table, the ten-slider audit, and the slider-writing cookbook - is the Concept Slider guide.

V10 Full Showcase

File: krea-v10-full-showcase-workflow.json

The best workflow for seeing the V10 controls together. Six cards, plus gentle balance, study reuse, and both feedback outputs wired to preview nodes:

Content anchor: keep the same subject, strength 0.80.
Visual style: suggest the visual style, strength 0.55.
Background/setting (new in V10): use the background/setting, strength 0.35.
Color palette only (new in V10): suggest the color palette, strength 0.40.
Camera framing (new in V10): copy the camera framing, strength 0.30, with When this card guides set to early layout only.
Text/logo guard: avoid copying text/logos, strength 0.03.

Queue once, read the stack report (Preview Any), check the prepared-references contact sheet (Preview Image), then tweak strengths - with Reuse image studies on, re-runs skip every encoder pass.

V10 Counter-Example

File: krea-v10-counter-example-workflow.json

Demonstrates the new Guide direction. Keep the subject, push a style out:

Reference 1: keep the same subject, toward this image, strength 0.80.
Reference 2: suggest the visual style, away from this image, strength 0.25.

Start away cards low (0.10 to 0.30) and always keep something positive - a written prompt or a toward card - saying what you do want.

V10 Starter Reference Stack

File: krea-v10-reference-stack-workflow.json

The compact V10 starter graph:

Load Image -> KG Krea 2 Image Guide Card V10 -> KG Krea 2 Reference Stack Encoder V10 -> sampler positive input

Plus the two feedback wires: stack_report -> Preview Any and prepared_references -> Preview Image.

V10 Starter Recipe (custom recipes in action)

File: krea-v10-starter-recipe-workflow.json

One card running cinematic color grade - one of the three custom recipes that ship enabled in custom_recipes/starter-pack.yaml (no extra install needed):

Reference 1: cinematic color grade, strength 0.70, on the style reference asset - the scene takes the reference's grade like a film LUT while the prompt keeps the subject.

Swap the dropdown to borrow the weather or borrow the clothing style (with a suitable reference) to try the other starter recipes, or open the Recipe Builder to make your own - the stack report names the recipe on its card line either way.

V9 Full Showcase

File: krea-v9-full-showcase-workflow.json

This is the best workflow for seeing the V9 controls together. It uses five cards:

Content anchor: keep the same subject, strength 0.80.
Visual style: suggest the visual style, strength 0.65.
Material/texture: suggest material or texture, strength 0.35.
Manual lighting/mood: manual tuning with lighting and shadows, strength 0.45.
Text/logo guard: avoid copying text/logos, strength 0.03.

V9 No-Prompt Style Transfer

File: krea-v9-no-prompt-style-transfer-workflow.json

Use this when you want image 2's visual style applied to image 1 without typing a prompt. It starts with:

Reference 1: keep the same subject, strength 0.80.
Reference 2: suggest the visual style, strength 0.65.
Final image prompt: blank.
Written prompt strength: 0.0.
When images guide: smart per-card timing.

For product/object content with a person-style reference, lower Reference 2 toward 0.45.

V9 Starter Reference Stack

File: krea-v9-reference-stack-workflow.json

This workflow demonstrates the V9 Krea reference route:

Load Image -> KG Krea 2 Image Guide Card V9 -> KG Krea 2 Reference Stack Encoder V9 -> sampler positive input

Workflow Dependencies

The workflows require Krea 2 compatible model, CLIP, and VAE files in your ComfyUI model folders.

If your model filenames differ, update the loader widgets in the workflow after loading it.

The V10 workflows additionally use the core Preview Any and Preview Image nodes for the stack report and prepared-references outputs; both ship with ComfyUI.

V11 worked examples and troubleshooting covers product/style guidance, counter-examples, brightness, custom poles, presets, timing, budgets, caching and migration.

README.md Unescape Escape

Example Workflows

Open directly in ComfyUI