Edit: to get the same results as my example images use SNOFS @ 0.6 + Realism Engine @ 0.6 + Realism Slider @ 0.6 + the standard Bypass Lora and Turbo Lora that is already built into the Turbo and Raw workflows respectively.
VERSION 2 — UPDATED WIRING. SAME ONE-CLICK IDEA
Paste one wildcard router tag. Hit Queue. Get a new full photoreal scene every run.
v2 replaces the old ImpactWildcardProcessor front-end with my WildcardClauseExpander + PromptLoraExtractor. Same simplicity — open the graph, paste a tag, go. Optionally you can override the subject within the node to lock in a custom subject description or lock in a certain outfit.
See this post for more info on how to use the Custom Subject Override if needed -> https://civarchive.com/models/2807882/custom-character-uw-expansion-pack-nsfw-modular-natural-language-wildcards
TLDR (how to use it)
1. Install the required nodes and wildcard pack (see below).
2. Open the Turbo or Raw workflow JSON in ComfyUI.
3. The wildcard tag is already filled in — or paste your own router tag.
4. Optional: fill Subject or Wardrobe on the Wildcard Clause Expander node.
5. Queue. New scene every run with populate + randomize on.
Default starter tag:
uw-krea2-nsfw/router/portrait-bedroom-pics
WHAT YOU GET IN THIS ZIP
krea2-custom-char-turbo.json Fast 8-step Turbo previews
krea2-custom-char-raw.json Raw UNET + Turbo LoRA quality path
WORKFLOWS.md Tag cheat sheet + full requirements
README.md Quick start
AGENT-INSTALL.md Step-by-step for AI agents / Cursor
This zip is workflows only. No wildcard YAML, no node source code, no character sidecar files.
TURBO VS RAW
Turbo — use for fast iteration and previews (8-step turbo path)
Raw — use for higher-quality final stills (Raw UNET + 2-pass sampler)
Both graphs are pre-wired the same way. Pick one, open it, queue.
HOW THE GRAPH WORKS (simple)
Wildcard Clause Expander
expands your tag and optional subject/wardrobe overrides
→ Prompt LoRA Extractor
pulls <lora:...> tags out of the text if you use them
→ Prompt Aspect Size Resolver
sets width/height from the prompt
→ CLIP encode + Empty Latent → sample → VAE → Save Image
LoRA tags in your prompt automatically route to LoRA Text Loader (LoraManager).
WHAT CHANGED FROM V1
v1 (older download on this page):
- Used ImpactWildcardProcessor for wildcard expansion
- No LoRA Extractor in the chain
v2 (this upload — recommended):
- WildcardClauseExpander + PromptLoraExtractor
- Optional subject_override / wardrobe_override for custom characters
- LoRA tags in prompts work out of the box
Character Expansion v1 YAML sidecars are a separate older artifact. v2 uses the Subject and Wardrobe fields on the expander node instead.
REQUIRED DOWNLOADS
Base uw-* wildcards (required)
Wildcard Clause Expander — Character Expansion v2 (required)
Prompt LoRA Extractor (required for v2 graphs)
https://civarchive.com/models/2818995/lora-extractor-comfyui-node-used-with-uw-wildcards-pack
Prompt Aspect Size resolver (required)
ComfyUI-Impact-Pack (required — wildcard backend)
https://github.com/ltdrdata/ComfyUI-Impact-Pack
comfyui-lora-manager (required — LoRA Text Loader node)
Krea 2 models — see the Note node inside each graph for exact filenames
Raw workflow also needs: rgthree-comfy, RES4LYF, VAE Utils
INSTALL
1. Install all required nodes above into ComfyUI/custom_nodes/
2. Install base wildcards into Impact Pack's custom_wildcards/ folder
3. Restart ComfyUI → Impact → Refresh Wildcard
4. Workflow → Open → pick krea2-custom-char-turbo.json or krea2-custom-char-raw.json
5. Queue once to confirm an image saves
AI agents: read AGENT-INSTALL.md inside this zip before guessing install order.
STARTER ROUTER TAGS
uw-krea2-nsfw/router/portrait-bedroom-pics
uw-krea2-nsfw/router/portrait-swimsuit-pics
uw-krea2-nsfw/router/portrait-intimate-action-pics
uw-krea2-nsfw/router/portrait-cheerleader-pics
Raw also supports portrait-intimate-action-partner-pics.
OPTIONAL — LOCK A CUSTOM CHARACTER
On the Wildcard Clause Expander node:
subject_override paste a natural-language identity (hair, face, skin tone, body description)
wardrobe_override paste a costume or outfit description
Leave both empty to use random subjects from the wildcard pack.
Description
v2.0.0 — Updated wiring for Wildcard Clause + LoRA Extractor
- Replaces ImpactWildcardProcessor with WildcardClauseExpander → PromptLoraExtractor
- LoRA Text Loader path for inline <lora:...> tags in prompts
- Optional subject_override / wardrobe_override on expander (Character Expansion v2)
- Ships AGENT-INSTALL.md for AI-assisted setup
- Turbo + Raw graphs updated; WORKFLOWS.md + README.md included
- Requires v2 companion node zips; no YAML or characters/ in this workflow zip
FAQ
Comments (9)
Hi. I'm enjoying this workflow, but could you kindly explain where a complete list of terms I could place in the Wildcard Router Tag node is available? The 3-4 samples that you give work fine, but I'm sure that there must be a lot others that I'm missing. Sorry if I misunderstood something, but this is my first experience with using wildcards in ComfyUI, so an explanation like I'm an 8-year-old might help! Many thanks.
I attached a text file with the available routers you can use
I assume using the existing power lora loader (which is "leave as-is" is fine, too, right? Instead of using <tags> where I find it easier to just load the lora there.
EDIT: I have a character lora, test trained with figzig. First thing I immediately notice is the hair color changed with a completely different hair style. No caption. But also want to add some sliders because it is not trained on figure.
Could you give an example which part of a wildcard prompt get's overridden by "subject_override"? I would like to simply change the hair color, and maybe nationality.
Hey! Thanks so much for the support. Yeah so the power lora loader is just meant for the Loras that in my mind you never remove (the bypass filter + the turbo Lora in the case of the raw mode).
You should add an extra Lora Manager node right after that to add in your own extra Loras.
I need to release a new version of my Workflow + my Wildcard node.
The current version of my custom wildcard node you have is the old outdated version... I didnt realize when I did it but there was a slight bug in the custom wildcard node that was making it rebuild the entire set of wildcards you have installed which can take a long time if you have multiple wildcard packs.
I already have that fixed.
I just got distracted by Minimax H3 coming out that I sort of got completely derailed the last 3 days
(btw I also am going to release a minimax H3 wildcard templates pack. It mainly is built around Image to Video, so you are expected to generate an image through Krea2 first and then pass that image along to H3... Literally all the videos on my page are generated using that flow.
Lastly my new custom wildcard node actually already has the things you asked for now. I added more specific overrides for various 'sub fields' that people commonly want to override, like the Hair, body type, age gender descr, etc...
So I will add the specific override also for Hair color and try to get it released tonight.
possible to add reference node so I can input a face and wardrobe ref?
Reference? Do you mean actual image references influencing the image output?
If you mean something like being able to take in an image, use an LLM to write a verbal description of that image, then inject those words into the node thats possible but.... Not something Im interested in lol that sounds like a lot.
V3 of my node is going to allow you to add in custom text overrides for all the different fields like the characters wardrobe for example.