OmniForge Krea2 MegaStyles
详情
下载文件 (1)
模型描述
1,500-Style Single Image Studio and Prompt Director for ComfyUI
OmniForge Krea2 MegaStyles is a self-contained ComfyUI custom node, structured prompt generator and vertical single-image workflow created for Krea2 Turbo.
It allows you to write a normal image description and automatically combine it with a selected or randomized visual style, subject lock, genre, scene direction, composition, camera language and organized output metadata.
This resource is not a checkpoint and not a LoRA. It is a custom-node package, database and ready-to-use ComfyUI workflow.
Main Features
1,500 Selectable Styles
The included style catalog contains:
1,500 unique selectable styles
12 balanced visual families
125 styles in every family
285 retained source styles
1,215 additional structured style recipes
The additional recipes were constructed using curated combinations of visual medium, historical era, rendering method and image treatment.
Included Style Families
Anime & Manga
Cartoon & Feature Animation
Comics & Graphic Novels
Handcrafted & Stop Motion
Game, Retro 3D & Stylized Rendering
Printmaking & Decorative Art
Horror, Gothic & Noir
Cinematic & Photographic Illustration
Fantasy & Concept Illustration
Painting & Fine Art
Surreal & Experimental
General Illustration
Large Structured Database
The package contains 17,509 physically stored JSON records.
Major catalogs include:
1,500 styles
73 genres
1,200 locations
1,000 narrative events
1,440 primary actions and locomotion records
800 secondary gestures
800 camera, camera-motion and shot records
240 perspectives
480 blocking templates
1,200 protagonists
800 villains
1,500 creatures
400 allies
400 gods
400 undead characters
600 hybrid characters
400 costumes
400 props
400 powers
300 film worlds
200 palettes
200 weather profiles
200 audio-direction records
160 story arcs
The current Single Image node directly uses the style, family, genre, location, event, action, gesture, camera, perspective and solo-character blocking pools. The remaining structured catalogs are included in the package for legacy functions and future workflow expansion.
Style Selection Modes
Random All Styles
Every style in the complete catalog has an equal chance of being selected.
Balanced Random Families
OmniForge first selects one of the 12 families uniformly and then chooses one style from that family.
This prevents a larger or more common visual category from dominating repeated generations.
Random by Selected Family
Randomizes only within the family selected by the user.
Fixed Style
Uses one exact style selected from the searchable 1,500-style dropdown.
Manual Style Override
A manually entered style takes priority over the dropdown and randomization modes.
Scene Generation Modes
Use Prompt Text
Uses the scene written in the main text editor without adding a database-generated environment.
Fixed Scene
Adds a manually entered scene to the final prompt.
Random Full Scene
Automatically builds a scene using:
Location
Narrative event
Primary action or pose
Secondary gesture
Camera direction
Perspective
Solo-character blocking
The result remains one standalone image rather than a collage, storyboard or multi-panel layout.
Subject and Style Locks
The node includes separate control levels for:
Subject identity
Visual style
Available lock strengths include flexible, strong and absolute modes.
You can use:
A custom subject description
A subject name only
A small built-in profile system for selected known characters
For the most reliable character accuracy, provide a complete custom subject description.
Persistent Anti-Repeat System
OmniForge can remember recently selected database elements and temporarily exclude them from future queues.
Editable repeat windows are available for:
Styles
Families
Locations
Events
Actions
Gestures
Cameras
Perspectives
Blocking templates
The history is stored locally in:
runtime/selection_history_v6.json
Delete this file while ComfyUI is closed to reset the selection history.
Included Workflow
The supplied workflow is configured for a single vertical image at:
1080 × 1920
Default generation settings:
Eight steps
CFG 1
Euler sampler
Simple scheduler
Full denoise
Krea2 Turbo FP8 model
Qwen3-VL 4B text encoder
Qwen Image VAE
The workflow also includes:
rgthree Power LoRA Loader
Multiple optional LoRA slots
Automatic positive-prompt transfer
Dedicated negative prompt
Dynamic folder and filename generation
Collision-safe PNG filenames
Prompt and generation metadata embedded in the PNG
Read-only resolved-prompt preview inside the OmniForge node
All optional LoRA slots are disabled by default. The referenced LoRA files are examples and are not included in this download.
Required Models
The included workflow expects the following files:
krea2_turbo_fp8_scaled.safetensors
qwen3vl_4b_fp8_scaled.safetensors
qwen_image_vae.safetensors
These model files are not included.
The workflow also uses the Power LoRA Loader from rgthree-comfy. Install rgthree-comfy or replace the LoRA stacker with another compatible model and CLIP loader arrangement.
Installation
Close ComfyUI.
Extract the ZIP.
Copy:
ComfyUI_OmniForge_Krea2_MegaStyles_v6
into:
ComfyUI/custom_nodes/
Remove or move older OmniForge installations that contain the same database. Keeping multiple copies may create duplicate nodes and unnecessary scanning.
Restart ComfyUI.
Refresh the ComfyUI browser page with
Ctrl + F5.Load the JSON workflow found inside the included
Workflowfolder.Verify the Krea2 model, text-encoder and VAE paths.
Remove or replace unavailable LoRA names inside the Power LoRA Loader.
No separate pip installation is required specifically for OmniForge.
Recommended Settings
For maximum controlled variety:
style_mode: Balanced Random Familiesscene_mode: Random Full Sceneanti_repeat_enabled: truestyle_repeat_window: 50family_repeat_window: 4scene_repeat_window: 25randomize_each_queue: true
For a specific visual result:
Set
style_modeto Fixed StyleSelect the desired style
Leave
manual_style_overrideemptySet
style_lockto Strong or Absolute
Filename System
The included robust saver supports dynamic filename tokens such as:
%subject%%style%%genre%%family%%scene%%seed%%date%%time%%date:yyyy-MM-dd%%date:HHmmss%
Default output structure:
Krea2/OmniForgeMegaStyles/YYYY-MM-DD/subject-style-time.png
Unsafe path characters and parent-directory traversal are filtered before saving.
Existing files are preserved by automatically adding a numerical suffix.
PNG Metadata
Saved images can contain:
The complete ComfyUI prompt metadata
The exact prompt generated by OmniForge
Selected subject
Selected style
Selected genre
Selected scene
Selection seed
Filename prefix
Structured OmniForge generation report
This makes it easier to inspect, archive and reproduce the prompt construction used for an image.
Important Seed Note
The OmniForge selection seed and the KSampler image seed are currently independent.
The OmniForge seed determines randomized database selections and is stored in the generation report. The actual image-generation seed remains controlled by the KSampler.
For exact image reproduction, preserve both the OmniForge settings and the KSampler seed.
Privacy and Local Operation
The supplied custom-node code contains no network-request or telemetry functions.
It operates locally by:
Reading the included JSON catalogs
Writing a small anti-repeat history file
Generating structured text prompts
Saving PNG files within the ComfyUI output directory
Download Contents
OmniForge custom-node folder
1,500-style catalog
Complete structured database
Local prompt-preview interface extension
Robust PNG saver
Legacy compatibility nodes
Vertical 1080 × 1920 Krea2 workflow
Installation and usage README
Share your generations and mention which style family, scene mode and Krea2 settings you used. Feedback and detailed bug reports are useful for improving future OmniForge releases.

















