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-krea2-two-character-composer

ComfyUI Extension: comfyui-krea2-two-character-composer

Repo Name

comfyui-krea2-two-character-composer

Author
cyberdeliaai (Account age: 0 days)
Nodes
View all nodes(1)
Latest Updated
2026-10-13
Github Stars
0.0K

How to Install comfyui-krea2-two-character-composer

Install this extension via the ComfyUI Manager by searching for comfyui-krea2-two-character-composer
  • 1. Click the Manager button in the main menu
  • 2. Select Custom Nodes Manager button
  • 3. Enter comfyui-krea2-two-character-composer 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-krea2-two-character-composer Description

comfyui-krea2-two-character-composer places two characters with separate prompts and LoRAs in one Krea 2 image, using adjustable regions or masks to control placement and reduce identity blending.

comfyui-krea2-two-character-composer Introduction

<p align="center"> <img src="assets/icon.png" alt="Krea2 Two Character Composer icon" width="200">

comfyui-krea2-two-character-composer is a single ComfyUI node for placing two characters, each with their own character LoRA, in one Krea 2 still image. It gives each character a designated area of the canvas while keeping a shared prompt for the setting, lighting, and overall style.

This is useful when a character’s appearance starts to influence the other character, or when you want clearer control over who belongs on which side of a scene. You can divide the image left and right or top and bottom, arrange two movable boxes, or supply masks of your own. A live preview helps you check the layout before generating.

The regional engine is included in the extension; you do not need to install ComfyUI-Krea2-Regional separately. The composer supports Krea 2 only. It does not support Z-Image Turbo, video, or Krea 2 reference latents.

Example workflow · Version 0.4.8 release

How comfyui-krea2-two-character-composer Works

Think of the canvas as a stage with two marked areas. The Base prompt describes the whole stage: where the characters are, the lighting, and the visual style. Each Character prompt describes one person and is assigned to that person’s area. A character LoRA, when selected, also acts primarily in its assigned area.

The composer turns those choices into regional guidance for Krea 2. It also adds a short placement description to the shared scene prompt when the areas have a clear left/right or top/bottom relationship. It does not add character size instructions, so describe intended sizes and poses yourself. A region occupying 60% of the image does not automatically make its character 60% larger.

For a first image:

  1. Open your working Krea 2 workflow or the example workflow.
  2. Connect the Krea 2 model, krea2 CLIP encoder, and empty latent to the composer. Connect the composer’s model, positive, negative, and latent outputs to the corresponding sampler inputs.
  3. Write the setting in Base prompt, such as Two adults in a garden, daylight, illustration.
  4. Give each character a separate prompt. Select each character’s Krea 2 LoRA and enter its trigger words if needed.
  5. Choose Left / Right, set Character 1 percent to 60, and begin with Balanced separation, LoRA edge softness 0, and Base strength 1.0.
  6. Generate with sampler settings that already work for your Krea 2 model.

In that example, Character 1 is assigned the left 60% and Character 2 the right 40%. The composer does not change your sampler or model-sampling settings. The example workflow starts at 8 steps and CFG 1 for Turbo; for RAW, use the settings from a working RAW workflow.

Krea 2 MODEL ──┐
Krea 2 CLIP ───┼── Two Character Composer ─ model ──── KSampler.model
Krea 2 LATENT ─┘                          ├ positive ─ KSampler.positive
                                          ├ negative ─ KSampler.negative
                                          └ latent ─── KSampler.latent_image

KSampler → VAE Decode → Save Image

Use EmptySD3LatentImage for the latent; despite its name, it produces the 16-channel latent used by Krea 2. Connect the composer’s model output directly to the sampler. Connecting negative is a useful default even if you are working at CFG 1.

comfyui-krea2-two-character-composer Features

Scene and character prompts

Base prompt applies to the whole image. Put shared information here: the location, time of day, lighting, mood, and art style. Character 1 prompt and Character 2 prompt hold each person’s appearance, clothing, and pose. Keeping those descriptions separate gives the composer clearer regional instructions.

Each character has an optional trigger field for the words required by their LoRA. Triggers are placed in that character’s prompt, not in the shared scene prompt. You can select a LoRA from the character’s menu and adjust its strength. Selecting None or setting strength to 0 turns that character’s LoRA off.

You can also add a character-only LoRA inside a trigger or character prompt with <lora:name> or <lora:name:0.8>. The tag is removed before the text is encoded. If you select the same file in the LoRA menu, the menu strength takes precedence. Do not put LoRA tags in the base or negative prompt; a LoRA intended for the whole image should be loaded before the composer.

The Swap characters control exchanges the two characters’ triggers, prompts, LoRA selections, and strengths. In a split layout, it also reverses the percentage so each character keeps their share while changing sides. In a free-box layout, the box coordinates stay where they are, so the characters exchange locations. Connected mask inputs and other input links are not swapped; exchange those connections manually if necessary.

Layouts and masks

  • Left / Right: Character 1 occupies the left portion and Character 2 the right.
  • Top / Bottom: Character 1 occupies the top portion and Character 2 the bottom.
  • Free boxes: Move and resize two regions independently in the preview, or enter their coordinates in Boxes.
  • Connected masks: Supply a mask for either character to replace that character’s layout-defined area during generation.

For either split, Character 1 percent can be set from 1 to 99; Character 2 receives the remainder. This sets regional area, not the apparent size of either person. Region boundaries align to Krea 2’s 16-pixel token grid, so very thin regions may not be usable.

In Free boxes, drag a box to move it or drag a corner to resize it. Coordinates are percentages in the order x, y, width, height, with x measured from the left and y from the top. For example:

5, 10, 40, 55; 35, 5, 60, 85

This makes a smaller first box toward the left and a larger second box toward the right. Keep each box within the image: x + width and y + height must each be no greater than 100. Clearing Boxes restores the default boxes when you choose Free boxes or load the workflow.

Boxes can overlap. In an overlap, Character 1 has prompt-region priority; that does not guarantee Character 1 will appear in front. Describe foreground and background relationships in your prompts. Areas outside both boxes follow only the base prompt.

A connected character_1_mask or character_2_mask replaces that character’s area, regardless of the selected layout. For example, you can connect a mask made with a Load Image node’s mask editor. The mask is resized to the image. A character’s prompt applies where its mask is more than half white; grey also reduces that character LoRA’s influence there. If a batch of masks is supplied, the first mask is used.

Separation and edge controls

Separation determines how strongly the character regions remain distinct:

  • Balanced: The starting choice for a coherent shared scene.
  • Strong: Also blocks image attention between the regions during the first half of sampling. Try it if identities mix, while watching for a less coherent shared scene.
  • Soft: An experimental option that keeps prompts separated while the layout forms, then allows the whole image to read both character prompts during the last 40% of sampling. Character LoRAs remain regional.

LoRA edge softness changes how character LoRA influence fades near a boundary. At 0, the edge is hard, which is a good starting point for distinct identities. If a boundary looks abrupt, try 32 and increase cautiously; more softness can also allow appearance details to blend. This setting does not move boxes or change prompt-region priority.

Base strength is experimental. Its default, 1.0, leaves the balance unchanged. Below 1.0, character prompts have relatively more influence in their areas; above 1.0, the shared scene and style have more. Values from 0.5 to 1.5 are suggested starting points for experimentation.

Live layout preview

The preview shows where the two regions are before you generate. Preview ratio can be Auto or a fixed preset: 1:1, 4:3, 3:4, 3:2, 2:3, 16:9, 9:16, 21:9, 9:21, 5:4, or 4:5. For example, select 4:5 when working at 1152×1440.

Auto can read visible width and height settings, CyberKrea resolution labels, and simple latent passthroughs. You can also connect both optional width and height inputs from a resolution node. If dimensions are calculated only when the workflow runs, Auto cannot predict them and shows a square preview marked “size unavailable.” Select a fixed ratio in that case.

The preview ratio does not resize your image. Generation and masks use the dimensions of the connected latent. Keep the preview ratio consistent with that latent so the on-screen layout is representative.

Reusable prompt text

Common prompts let you define a phrase once and reuse it across fields:

Base prompt:   A classroom at noon, {outfit: navy school uniform, red tie}.
Character 1:   a tall boy with black hair, {outfit}
Character 2:   a girl with a blonde ponytail, {outfit}

{name: text} defines the text and leaves it in that location. {name} repeats it, even if the definition appears later. Names are not case-sensitive. Defining the same name twice with different text stops the node with a message. An undefined name remains as text and is listed in the prompts output. Choice syntax such as {a|b} and weights such as {word:1.2} are left alone.

Optional negative prompts

At CFG 1, a conventional negative prompt does not work as it normally would. With NegPiP Prompt 2.1 or newer installed, the composer can use negatives at CFG 1:

  • Enter a whole-image exclusion in Negative prompt, such as blurry, (text, watermark:0.5).
  • Put a negative weight in one character’s prompt, such as a monk, (beard:-0.6), to target that character’s region.

The composer uses the installed NegPiP pack itself; do not add a separate NegPiP Prompt node before it. NegPiP is optional when you do not use negative features. It also enables weight syntax such as (red hair:1.3) in prompts that it processes; the stock Krea 2 encoder otherwise reads such weights as plain text.

Inspection and image metadata

The prompts output shows what was encoded. Connect it to Preview Any while refining descriptions. The optional region_masks output can be viewed through Mask To Image → Preview Image to inspect both masks as a batch.

If you use Save Image With Metadata (Cyberdelia) from comfyui-cyberdelia-metadata 2.2.2 or newer, saved metadata can retain the scene, both character prompts with triggers, the original global negative prompt, and the regional LoRAs and strengths. No extra connection from prompts is required. This metadata extension is optional and is not needed to generate images.

comfyui-krea2-two-character-composer Models

The composer does not include a choice of separate generation models. It works with a Krea 2 model supplied by your workflow and supports Krea 2 character LoRAs selected separately for the two characters. A LoRA loaded globally before the composer affects the whole image instead of just one character region.

The included example workflow uses Turbo sampler settings; if you use RAW, retain the settings of your working RAW workflow. The composer does not alter those settings. Z-Image Turbo and other non-Krea 2 models are not supported and produce an error rather than a composed image.

What’s New with comfyui-krea2-two-character-composer

  • Version 0.4.8: Registry packages omit tests, development tools, and CI workflow files. Generation behavior is the same as version 0.4.7.
  • Version 0.4.7: The composer reuses model layer objects when it re-executes, such as after you edit a prompt or LoRA strength. It also releases converted LoRA weights and runtime masks at sampling cleanup or when another composer cache becomes active, rebuilding them when needed. This helps avoid retaining those GPU tensors in older cached outputs. Actual speed and memory differences depend on your hardware and workflow.
  • Version 0.4.2: Character LoRA calculations skip tokens where that character’s mask is zero, reducing unnecessary adapter work. Region boundaries, strengths, grey-mask weighting, and edge softness retain their behavior. Generation still runs the base model and regional attention, so no fixed overall speedup is promised; the same seed may not produce pixel-identical results compared with version 0.4.1.

Saved workflows from the 0.2, 0.3, and 0.4 series are supported with their values restored to the appropriate fields and newer options starting at their defaults. Older prompts that contain inline triggers or LoRA tags can continue to work.

Troubleshooting comfyui-krea2-two-character-composer

The node is missing in ComfyUI.

  1. Restart ComfyUI and refresh your browser.
  2. Check that the extension folder is not inside an extra nested folder.
  3. Read ComfyUI’s startup log for an error.
  4. Check that your ComfyUI version has native Krea 2 support.

A character LoRA does not appear in the menu.

  1. Check that it is a Krea 2 LoRA in ComfyUI/models/loras.
  2. Refresh ComfyUI’s model list or restart ComfyUI.
  3. Reopen the node’s LoRA menu.

The two characters’ appearances mix.

  1. Keep each character’s description and trigger in that character’s own fields.
  2. Set LoRA edge softness to 0.
  3. Try Strong separation and compare the result with Balanced.
  4. Check whether a character LoRA was loaded before the composer; globally loaded LoRAs can affect both regions.

Free boxes do not behave as expected.

  1. Clear Boxes to restore the default boxes, or enter two valid groups of four numbers separated by a semicolon.
  2. Make sure each box remains within the image and each character has usable space.
  3. Check for connected character masks: those replace the corresponding box areas during generation.

The preview shape is wrong or says “size unavailable.”

  1. Choose a fixed Preview ratio matching your intended image dimensions.
  2. If using Auto, check whether the source dimensions are visible before the workflow runs.
  3. Check the connected latent. Its actual dimensions, not the preview setting, determine the generated image.

A negative prompt asks for NegPiP.

  1. Check that comfyui-negpip-zimage version 2.1 or newer is available.
  2. Restart ComfyUI.
  3. Leave out a separate NegPiP Prompt node; the composer calls the pack when needed.

The node reports an empty, thin, faint, or fully covered character area.

  1. Enlarge or brighten the affected mask, or make its box wider.
  2. Move overlapping boxes so each character retains an area of their own.
  3. Remember that boundaries use a 16-pixel grid and Character 1 has priority in overlaps.

Prompts are missing from saved image metadata.

  1. Use Save Image With Metadata (Cyberdelia) from comfyui-cyberdelia-metadata version 2.2.2 or newer.
  2. You do not need to connect the composer’s prompts output to that save node.

Frequently asked questions

  • Does a 60/40 split make one character larger? No. It divides the image into prompt regions. Describe each character’s intended size and position in the prompts.
  • Does an overlapping box put Character 1 in front? No. Character 1 has prompt priority in the overlap, but you should describe who is in front.
  • Can I use my own masks? Yes. Connect a mask to either character’s mask input; it replaces that character’s layout-defined region.
  • Can I generate video or use another model family? No. This composer is for Krea 2 still images, including batches.
  • Will the regions guarantee perfect placement? No. Overlapping people, hands crossing a boundary, and very narrow regions can still be difficult for the model.

Learn More about comfyui-krea2-two-character-composer

When reporting a problem to the project, include the composer and ComfyUI versions, the exact error message, and a small workflow that reproduces it. Remove private prompts and file paths before sharing.

comfyui-krea2-two-character-composer 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-krea2-two-character-composer detailed guide | ComfyUI