Update Notes
2026-07-09: Added more Background Effect presets from local WebP assets, including travel streets, transport interiors, arcades, capsule hotel, pachinko/slot interiors, sauna, vending machine corner, data center, showroom, planetarium, skate park, recording booth, and desert highway. Background thumbnails now use the updated cache version. Replaced the recommended test workflow with workflow/Krea2-BBOX-Node-Test.json.
Krea2 BBOX Prompter Suite
ComfyUI custom node suite for manually drawing BBOX regions on a canvas and converting that layout into a structured prompt for Krea2.
You specify where each subject, object, or visible text region should appear, write prompts for each box, and export a JSON-style prompt that can be passed to Krea2.
BBOX regions are layout guidance for Krea2. They are not strict masks. Final placement still depends on the model, prompt wording, and the overall scene context.
Recommended environment: ComfyUI 0.26.0 or later.
<Important Note>
For full setup details, examples, updates, and troubleshooting,
please see the GitHub repository linked below.
ukr8b3g-cmyk/Krea2-BBOX-Prompter
Quick Installation
For a new install:
cd C:\path\to\ComfyUI\custom_nodes
git clone https://github.com/ukr8b3g-cmyk/Krea2-BBOX-Prompter.git Krea2-BBOX-Prompter-SuiteIf the folder already exists, update it instead:
cd C:\path\to\ComfyUI\custom_nodes\Krea2-BBOX-Prompter-Suite
git pullThen restart ComfyUI and hard refresh the browser if the old UI is still cached:
Ctrl + F5What This Node Can Do
Draw BBOX regions manually to guide placement of people, objects, text, and other elements.
Enter Object / Text prompts per colored slot.
Object: for visual elements such as people, objects, or scene parts.
Text: for visible text that should actually appear inside the image.
Use
SCENE / BACKGROUNDto describe the overall image intent and environment.Add per-slot framing hints such as
Cowboy shot,Full body, andUpper body.Add per-slot camera angle hints such as
Front view,Low angle, andTop-down view.Generate a Krea2-oriented JSON prompt with the Export node.
Connect
width/heighttoEmptyLatentImageto keep the canvas size and generation size in sync.Add photographic, film, portrait, flash, art, lighting, and mood effects with the Prompt Effect node.
Choose style presets from a thumbnail-card UI using local WebP thumbnails.
Node: The Krea2 diffusion_model that I merged it, aiming for a clear, soft-core style that works well for both SFW and sensitive content; I plan to release it at a later date.
Description
FAQ
Comments (11)
I sure hope future models catch the regional prompting bug from ideogram, as a native alternative method.
Thank you for the comment.
This node does use a JSON-style structured prompt, but it is different from Ideogram 4’s native regional prompting system.
In this node, the JSON is used as a Krea2-oriented structured prompt. It contains the overall scene, background, BBOX coordinates, and per-box descriptions. The BBOX data is used as layout guidance, not as a strict mask or a guaranteed region lock.
So it does not mean that random text, symbols, or lines will appear by default. In normal use, the goal is simply to guide Krea2’s composition more clearly. However, because the final image is still generated by Krea2’s model behavior, I cannot guarantee 100% perfect regional control in every case.
The node includes measures to reduce layout confusion and prompt leakage, such as separated object/text regions, BBOX position data, and structured scene/background information. Still, depending on prompt wording, seed, overlapping boxes, and scene complexity, unexpected details can sometimes happen.
For the exact intended behavior and limitations, please also refer to the official README in the repository.
thank you for sharing such an excellent node. COOL~
Thank you for the question.
This node is mainly designed for Krea2 workflows and exports a Krea2-oriented JSON-style structured prompt. I have not officially tested or guaranteed compatibility with Z IMAGE TURBO. If Z IMAGE TURBO accepts the exported JSON-style prompt as normal text, it may work experimentally, but the BBOX behavior depends on that model’s prompt interpretation.
For layering / hierarchy, this node is not a true layer stack, mask system, or depth-control system. The colored BBOX regions are composition guidance, not strict masks or guaranteed region locks.
For “Son Goku holding a Dragon Ball,” I recommend describing the relationship directly rather than relying on overlapping boxes:
- Blue Object: Son Goku, upper body, holding a glowing orange Dragon Ball with both hands in front of his chest.
- Red Object: a glowing orange Dragon Ball, centered in front of the chest, held between both hands.
The ID-photo inset example is a separate framed/inset object, so it is easier than a physically interacting object. For held objects, prompt wording and scene context are more important than box hierarchy.
You can try it, but it is not guaranteed. Avoid large overlaps, keep the Dragon Ball box small and centered, and repeat the “holding with both hands” relationship in the scene and character prompts.
can apply a lora only on a specific region?
Not directly.
This node can assign prompts to BBOX regions, but it does not apply a LoRA only inside one specific region. LoRA is usually applied to the whole model/conditioning, so its influence may affect the entire image.
You can reduce the effect by lowering LoRA strength and describing the LoRA-related subject only in the target region, but this is not a true regional LoRA mask. For strict regional LoRA control, you would need a separate masked/regional LoRA workflow.
ig you could try a 2nd (i2i) pass with a different prompt+bbox location with the lora active?
@dow185 This node assigns prompts to individual BBOX regions, but it does not support applying a LoRA exclusively inside one specific region.
As a possible workaround, you can find and use a suitable i2i or inpainting workflow separately:
1. Generate a base image without the LoRA.
2. Load the result into an i2i or inpainting workflow of your choice.
3. Enable the LoRA only during the second pass.
4. Mask the target area and describe the LoRA-related subject there.
5. Start with a low LoRA strength and adjust it gradually.
This may help keep the LoRA's influence closer to the selected area. However, the LoRA still affects the overall conditioning, so this method cannot guarantee completely isolated regional LoRA control. The required inpainting workflow is separate from this node and must be selected or prepared by the user.
I'm getting a missing node error for Krea2BBOXPoseGestureInjectorV1 when loading the workflow. I couldn't find this node in the GitHub repository. Could you please tell me where I can download it?
Thank you for reporting this, and sorry for the confusion.
Krea2BBOXPoseGestureInjectorV1 was an older standalone development node. Its pose and gesture functions are now included directly in the current Krea2 BBOX Prompter node.
Therefore, the old standalone node no longer needs to be installed separately. The missing node error appears because the workflow was created with an older development version.
Please reinstall the latest version of the Krea2 BBOX Prompter Suite using the following steps.
Close ComfyUI.
Open your ComfyUI custom_nodes folder.
Example:
ComfyUI\custom_nodes\
Delete the existing old folder:
Krea2-BBOX-Prompter
Download the latest version from the official GitHub repository:
https://github.com/ukr8b3g-cmyk/Krea2-BBOX-Prompter
You can install it using either of the following methods.
Method 1: Download ZIP
Open the GitHub page.
Click Code.
Select Download ZIP.
Extract the ZIP file.
Place the extracted Krea2-BBOX-Prompter folder inside:
ComfyUI\custom_nodes\
Make sure the folder is not nested twice, for example:
Incorrect:
custom_nodes\Krea2-BBOX-Prompter-main\Krea2-BBOX-Prompter\
Correct:
custom_nodes\Krea2-BBOX-Prompter\
Method 2: Git clone
Open PowerShell or Command Prompt inside the custom_nodes folder and run:
git clone https://github.com/ukr8b3g-cmyk/Krea2-BBOX-Prompter.git
Restart ComfyUI.
Reload the workflow and replace the old missing Krea2BBOXPoseGestureInjectorV1 node with the current Krea2 BBOX Prompter node.
The pose and gesture functions are already included in the current Prompter node, so no additional node download is required.
Official GitHub repository:
https://github.com/ukr8b3g-cmyk/Krea2-BBOX-Prompter
Civitai project page:
https://civitai.red/models/2742034/krea2-bbox-prompter-suite
Thank you again for bringing this issue to my attention.
Thank you