ComfyUI Plugin Krea Prompt Generator: Build Prompts by Attribute

A Prompt Tool Built for the Inspiration-Challenged

Writing high-quality prompts is often the most mentally taxing part of any AI image generation workflow. One Reddit user admitted to being "lacking in imagination" and decided to build a set of custom nodes for ComfyUI — called Krea Prompt Generator — that lets users assemble complete prompts by selecting attributes from dropdowns or feeding in a reference image.

The tool focuses specifically on female character image generation. Under the hood, it leverages ComfyUI's text generation capabilities and text encoder. The author openly acknowledges that the attribute dictionary was "borrowed" from the existing comfyui-character-prompt-builder project, which already offers comprehensive coverage of various attributes. The project also candidly admits to some "vibe-coding" — after running into repeated issues, the author enlisted Claude to help fix parts of the code, then did their own testing and validation.

ComfyUI is an open-source node graph-based interface for Stable Diffusion. Rather than following a fixed linear UI, users drag and connect functional nodes to build their image generation pipelines. This design allows community developers to write "custom nodes" — essentially Python modules that, once registered, can be used on the canvas just like built-in components. The text encoder in this pipeline converts prompt strings into vector representations that the model can read (commonly referred to as conditioning), serving as the core bridge between text descriptions and image generation. CLIP, introduced by OpenAI and widely used across Stable Diffusion models, maps images and text into a shared semantic space via contrastive learning.

How the Six Nodes Divide the Work

The toolkit's core consists of nodes split by function. Each node handles one dimension of the prompt, and they're chained together before feeding into the generator node.

Node Overview

• Krea Prompt - Female Person: Basic character attributes including age, nationality, body type, face shape, eyes, hairstyle, skin tone, tattoos, and more
• Krea Prompt - Female Fashion: Clothing items with colors and materials, plus shoes, accessories, jewelry, nails, and other details
• Krea Prompt - Female Actions: Poses such as standing, sitting, kneeling, and lying down, as well as hand, leg, and head movements and props
• Krea Prompt - Scene: Style, setting, time of day, season, weather, lighting, and camera angle
• Krea Prompt Generator: Receives CLIP, all attributes, and keywords; outputs conditioning, prompt, and llm_request
• Krea Prompt - Save Unique Prompt: Appends each newly generated prompt to a text file, automatically skipping duplicates

This modular breakdown lets users fine-tune each dimension rather than dealing with one large, opaque black box.

Three-State Dropdowns: Flexible Randomness Control

Every dropdown offers three selection modes — one of the more thoughtful design choices in the whole system:

• none: The attribute is excluded entirely from the request sent to the LLM
• random: The generator randomly picks a value based on the seed — the same seed always produces the same result
• Specific value: Passed through as-is

The seed-based reproducible randomness means users get the thrill of surprise while still being able to precisely reproduce any satisfying combination they discover. Each attribute node also includes a free-text field for adding extra detail within that category, balancing structure with flexibility.

A "seed" here refers to the initial value for a pseudorandom number generator. Given the same seed, the program produces an identical random sequence every time — and therefore the same attribute combination. This mechanism is widely used in AI image generation for reproducibility: when users stumble upon a result they like during random exploration, they simply note the seed value to recreate it exactly without memorizing every parameter. "Vibe-coding" is an informal term that's become popular in the community, referring to a development style driven by intuition, trial-and-error, and AI assistance rather than strict engineering discipline — emphasizing rapid iteration over code quality, and common in personal tools and experimental projects.

Reference Image Support and Wiring

Beyond attribute selection, the generator node also supports reference image and mask inputs. Users chain their attribute nodes into the generator's attributes input, then connect the reference image and mask to the generator node as well. There's a setting inside the generator that controls what information gets extracted from the image. The author's own assessment of this feature: "works ok" — functional, but not perfect.

The top of the workflow also includes quick bypass switches to skip the attribute nodes, image reference nodes, or toggle mask inversion with a single click — handy for debugging and rapid iteration.

Installation and Current Status

The plugin has not yet been submitted to the ComfyUI Registry, so users need to install it manually via git clone. The author notes that as long as your ComfyUI is updated to a version from late February this year or later, no additional dependencies are required. The workflow file is included in the repository and is also embedded in the workflow image in the original post.

The author says they'll invest time in formally submitting it to the Registry if there's enough community interest. They also acknowledge that the prompt-saving-to-text-file functionality is still "a bit janky," and welcome feedback on anything that's missing.

Summary: A Pragmatism-First Community Tool

The value of tools like this isn't in technical breakthroughs — it's in lowering the barrier to entry. It transforms the scattered mental effort of composing prompts into a set of clickable, reproducible, and extensible structured nodes. For ComfyUI users who freeze up in front of a blank prompt box, this is a worthwhile starting point. That said, as a semi-vibe-coded personal project that hasn't made it into the official registry yet, it does require a degree of hands-on willingness, and stability still has room to improve with community input.

Project repository: comfyui-krea-prompt-generator