VELVET VICE — KREA 2 VISION PROMPTER v2.0
Modular Krea 2 Turbo Workflow for ComfyUI
VELVET VICE — KREA 2 VISION PROMPTER v2.0 is a modular ComfyUI workflow built for fast image creation, intelligent prompt assistance and flexible image editing with Krea 2 Turbo.
The workflow provides clearly separated workspaces for CREATE, Native Edit, Custom Target and Classic Img2Img, keeping each generation path independent and easy to control.
Optional refinement, five dedicated detailers, SeedVR2 upscaling and separate preview and output paths provide additional control without forcing everything through one oversized generation chain.
Main Features
Dedicated CREATE workspace
Native Edit workflow branch
Custom Target workflow branch
Classic Img2Img workflow branch
New VELVET VICE Prompt Director v2
Qwen3.5 9B assisted prompting
Manual prompting without LLM dependency
Prompt-First memory architecture
Format & Resolution selector
Manual width and height control
Optional refinement stage
Five dedicated detailers
SeedVR2 final upscaling
Stage-based rendering controls
PowerLoRA integration
Improved interrupt memory handling
Model-neutral public workflow
Fully isolated Emerald + Violet Velvet Vice interface
Direct installation through ComfyUI Manager / Missing Custom Nodes
Prompt Director v2
Version 2.0 introduces the new VELVET VICE Prompt Director v2.
It is designed to transform short ideas, keywords or simple instructions into clean, structured prompts optimized for Krea 2.
The assisted mode uses:
fredrezones55/Qwen3.5-Uncensored-HauhauCS-Aggressive:9b
The new Prompt-First architecture allows the prompt assistant to finish its work and release Ollama before the heavy image-generation models are used, helping reduce unnecessary memory usage.
Manual prompting remains fully available and does not require Ollama or the LLM.
Format & Resolution
The CREATE workspace includes a dedicated Format & Resolution system.
Common aspect ratios and resolution presets can be selected directly inside the workflow while still preserving full manual control over width and height.
This allows quick format changes without removing the flexibility of custom resolutions.
Modular Generation Architecture
The workflow keeps the major generation paths separated:
CREATE
Native Edit
Custom Target
Classic Img2Img
Additional stages such as refinement, detailers and SeedVR2 can be used when required without forcing every image through the complete processing chain.
Stage-based rendering controls also make it possible to work more selectively inside the workflow.
Model Neutral
The public workflow is designed to be model-neutral.
You can select your own compatible Krea 2 model instead of being forced to use a specific creator checkpoint.
Models, LoRAs and third-party custom nodes are not included.
ComfyUI Manager Installation — Recommended
VELVET VICE — KREA is now officially available through the Comfy Registry.
The Velvet Vice custom-node package can now be installed directly from inside ComfyUI.
Simply:
Load the KREA workflow in ComfyUI.
Open Missing Custom Nodes.
Install VELVET VICE — KREA.
Restart ComfyUI.
Reload the workflow.
You can also search directly for:
velvet-vice-krea
inside the ComfyUI Custom Nodes Manager.
Manual / Fallback Installation
The included installer remains available as a manual / fallback installation method.
This can be useful if:
ComfyUI Manager is unavailable
the Registry cannot be reached
you prefer a manual installation
you want to use the included backup / installer system
The Registry installation and the included installer both provide the same VELVET VICE — KREA custom-node package.
One-Click Autoprompt Setup
The Autoprompt model setup is fully automated.
Simply run:
_INSTALL_AUTOPROMPT_MODEL.bat
The installer will automatically:
check if Ollama is installed
install Ollama if required
download the recommended Autoprompt model
store it through Ollama's own model system
Recommended model:
fredrezones55/Qwen3.5-Uncensored-HauhauCS-Aggressive:9b
No manual LLM download or ComfyUI model-folder setup is required.
One click → wait for installation → ready to use.
The Ollama setup remains separate from the ComfyUI Registry installation because the Autoprompt model is managed by Ollama, not by ComfyUI.
Velvet Vice Interface
Version 2.0 uses the isolated Emerald + Violet Velvet Vice design system.
The interface includes refined spacing, clearer workflow sections, improved stage controls and automatic Velvet Vice styling for compatible PowerLoRA nodes.
The KREA interface remains isolated from the Velvet Vice LTX workflows.
Creator Name Update
Previous workflows were published under the name ILLUMINATE AI.
VELVET VICE is the same creator under a new name and visual identity.
This is a rebranding, not a change of ownership or developer.
Feedback and detailed bug reports are welcome.
Created by VELVET VICE
Previously known as ILLUMINATE AI
Description
V2.0 – Major Workflow Update
VELVET VICE · KREA 2 VISION PROMPTER v2.0 is a major update focused on prompt intelligence, workflow usability, memory handling and a cleaner public release.
What changed:
New KREA Prompt Director v2
Upgraded to the same Qwen3.5 9B LLM used in the Velvet Vice LTX workflows.
Improved keyword-to-prompt expansion specifically tuned for Krea 2.
Better preservation of subjects, actions, camera instructions and other important user constraints.
Adaptive prompt expansion avoids unnecessarily bloated prompts.
Improved handling of multi-subject scenes, composition, lighting and scene structure.
Local output validation and cleanup.
MANUALmode still completely bypasses Ollama.ASSISTEDmode uses a single LLM call.
Prompt-First Memory Gate
The LLM finishes its work before the heavy Krea render models are allowed to load.
Ollama is released after prompt generation to reduce unnecessary memory usage.
Designed to keep the prompt assistant and image-generation models from occupying resources at the same time whenever possible.
New Format & Resolution Selector
Added directly to the CREATE / Text-to-Image branch.
Choose common aspect ratios instead of entering width and height manually every time.
Includes portrait, landscape, square, widescreen, vertical and ultrawide formats.
PRESETandMANUALmodes are both available.Manual width/height control is fully preserved.
Render This Stage
Added partial-execution controls for selected workflow stages.
Allows easier testing of Base, Refinement, Detailer and SeedVR2 / Final stages without manually navigating the graph.
Improved PowerLoRA integration
Newly created, duplicated or pasted PowerLoRA nodes can automatically inherit the Velvet Vice Krea visual style.
Krea theme detection remains isolated from the separate Velvet Vice LTX workflows.
Memory / Interrupt improvements
Improved cleanup behavior after manually interrupted generations.
Avoids aggressive model unloading methods that can unnecessarily move large model allocations into system RAM.
Emerald + Violet UI update
Updated Krea-specific Velvet Vice color palette.
Existing Velvet Vice animated node behavior is preserved.
Theme ownership is isolated so Krea styling does not overwrite LTX 2.3 / 2.5 workflows.
Layout cleanup
Full workflow layout audit performed.
Increased spacing in several tightly packed areas.
Improved spacing around Detailers, Preview/Save sections, Image Edit controls and other dynamically sized nodes.
Model-neutral public workflow
Removed the previous hard dependency on a specific personal Krea checkpoint.
The workflow is intended to work with the user's own compatible Krea 2 model installation.
General
Existing independent CREATE, Native Edit and Classic Img2Img branches remain intact.
Existing SeedVR2, Detailer, Sampler and Bypass architecture has been preserved.
Workflow title and release branding updated to v2.0.
This version is focused on making Velvet Vice Krea 2 smarter, easier to use, more memory-conscious and much cleaner for public distribution.
FAQ
Comments (6)
This seems pretty darn awesome, I’m enjoying trying it out. I may have missed it, but do you have any readme docs on the i2i/image edit lane?
@semipro211 Hi, thanks for your curiosity. Yes, further down you will find the Classic Image-to-Image and the Edit. Custom Target Format. Right there are the descriptions.
@semipro211
# KREA 2 EDIT — CUSTOM TARGET FORMAT
This optional branch lets you render the edited image in a new target resolution or aspect ratio while still using the original source image as the grounded visual reference.
## HOW TO USE
1. Keep the standard Original Target branch for preserving the source format.
2. Enable this Custom Target Format group only when you want a different output size or aspect ratio.
3. Set the custom width and height in KREA 2 EDIT — CUSTOM TARGET LATENT.
4. Use the same Krea 2 Edit prompt and source image as usual.
5. Test the base result first, then optionally enable SeedVR2 for the final upscale.
## RECOMMENDED PRACTICE
- Use moderate format changes first.
- Extreme aspect-ratio changes require the model to invent more background content.
- Grounding 768 remains a strong general starting point.
- If the subject changes too much, try higher grounding values such as 896 or 1024.
## TYPICAL CUSTOM TARGETS
- 1344 × 768 (16:9)
- 768 × 1344 (9:16)
- 1536 × 1024 (3:2)
- 1024 × 1536 (2:3)
- 1536 × 1536 (square)
For best efficiency, keep only the Krea 2 Edit output mode you actually need enabled.
@semipro211 11 CLASSIC IMG2IMG — CREATIVE IMAGE-TO-IMAGE
This workflow includes a standalone Classic Img2Img branch.
Unlike Krea 2 Edit, this branch uses a more traditional denoise-based image-to-image process. It is generally more flexible, more creative and allows stronger reinterpretation of the original image.
BEST USE CASES
Use Classic Img2Img when you want to:
transform the source image more freely
create stronger stylistic changes
push the image further away from the original
use a normal denoise-based editing workflow
make use of a traditional optional negative prompt
get more creative reinterpretation rather than strict source preservation
Classic Img2Img is usually less strict than Krea 2 Edit when it comes to identity, pose, composition and overall scene preservation.
QUICK START
Enable the CLASSIC IMG2IMG group.
Keep the Krea 2 Edit branch bypassed.
Load your source image into CLASSIC IMG2IMG — SOURCE IMAGE.
Write your transformation prompt in the Classic Img2Img Positive Prompt.
Optionally use the Classic Img2Img Negative Prompt.
Start with the recommended settings below.
Test the base result first.
Enable SeedVR2 only after the Classic Img2Img result already looks good.
HOW TO PROMPT
Classic Img2Img behaves more like a creative reinterpretation workflow.
You can prompt it more openly than Krea 2 Edit, but it still helps to be clear about:
what should change
what style or look you want
what should stay recognizable if needed
Example:
Transform the image into a cinematic portrait with dramatic lighting, richer shadows, sharper clothing details and a more refined facial appearance while preserving the overall subject and composition.
If you want stronger transformation, you can describe the desired result more boldly.
Example:
Reimagine the source image as a high-fashion editorial portrait with luxurious styling, elegant makeup, premium clothing textures and a dramatic studio atmosphere.
DENOISE
Denoise is the main control for change strength in this branch.
General behavior:
lower denoise = closer to the source image
higher denoise = stronger transformation and more reinterpretation
Suggested ranges:
0.25–0.40 = subtle changes
0.40–0.60 = balanced image-to-image editing
0.60–0.80 = stronger transformation
0.80+ = highly creative, but less predictable
Recommended starting value:
0.50
NEGATIVE PROMPT
Unlike Krea 2 Edit, this branch supports a more normal optional negative prompt workflow.
Use the negative prompt to suppress unwanted artifacts or tendencies, for example:
blurry
low detail
distorted anatomy
extra fingers
overexposed
text artifacts
Keep the negative prompt relatively short and focused. Avoid giant keyword walls.
RECOMMENDED DEFAULT SETTINGS
Denoise: 0.50
Steps: 8
CFG: 1.0 or your preferred Classic Img2Img baseline
Sampler: Euler
Scheduler: Beta
Optional Negative Prompt: enabled only if needed
SeedVR2: Bypass during initial testing
OPTIONAL SEEDVR2 FINAL UPSCALE
SeedVR2 is used only as the final stage.
Recommended order:
Classic Img2Img
→ Base Preview
→ Optional SeedVR2
→ Final Preview
→ Save
SeedVR2 may improve final presentation quality and resolution, but it can also reinterpret small details. Always compare the base result and the upscaled result.
Use Bypass, not Mute or Never, when disabling the SeedVR2 group.
WHEN TO USE KREA 2 EDIT INSTEAD
Use Krea 2 Edit instead if you want:
more precise editing
stronger identity preservation
better pose and composition retention
more controlled, instruction-based changes
source-faithful editing
SUMMARY
Classic Img2Img
Creative, flexible, denoise-based and best for stronger reinterpretation.
Krea 2 Edit
Precise, controlled, identity-preserving and best for targeted edits.
@VelvetVice Thank you immensely for the detailed reply. I am more at the beginner/intermediate level, and the sheer volume of available workflows and the complexity of said workflows can be a bit overwhelming.
This really helped, thank you again
@semipro211 Thank you for choosing my workflow, it still needs improvement... everything works wonderfully... but I noticed that when using Autonomvision for a long time, it becomes sluggish... I still need to see what might be causing it. But after a restart or generating directly in manual mode, the system does not slow down...
Anyway... at the moment I'm finishing up Minimax H3 Vision... after that, I'll get to the aforementioned thing.