v3.0.0 all-in-one bundle (wildcards + Wildcard Clause Expander)
Adults 18+ only. NSFW photoreal Krea 2.
One zip installs the wildcard pack, three companion nodes, and starter Turbo / Raw workflows. After that you lock a subject and steer the scene inside ComfyUI. No YAML editing. No character-kit rebuild. No ComfyUI relaunch just to change hair, body, wardrobe, or lighting.
What v3.0 actually is
This is a feature release, not a patch. If you only used v2 (or the old split downloads), almost everything below is new.
Custom character and scene, on the node
v2 made you pick a subject and a wardrobe, then live with whatever the YAML rolled. Want a specific adult look? You built a character kit, edited files under custom_wildcards, or ran a twin-YAML compile, then restarted ComfyUI before the next queue.
v3 puts that work on Wildcard Clause Expander. Age, hair, body, skin, face, nationality, wardrobe, room, medium, lighting, treatment (17 override fields) sit on the node. Type a clause, queue again. The wildcard corpus on disk does not change.
Empty fields still roll from the pack (italic ghost text in the UI). Pin copies a ghost into a real override so that slot stays put. Skip drops the clause from populated text. ↻ rerolls one slot without rerolling the rest of the prompt.
That is the point of this version. Keep the same router or template, swap the person and the room from the node, generate.
Three downloads on v3.0.0
| Zip | Use when |
|-----|----------|
| uwModularNSFWWildcardsPhotoreal_v30AllInOne.zip (full) | Embedded <lora:…> in wildcards + photoreal workflows + 358 Krea 2 catalog thumbs. Download companion LoRAs below. |
| uwModularNSFWWildcardsPhotoreal_v30AllInOne_noLora.zip (no embedded LoRA) | Same nodes, YAML, and catalog thumbs; all <lora:…> embeds stripped. Queue without companion LoRA files. |
| uwModularNSFWWildcardsPhotoreal_v30CatalogThumbs.zip (thumbs only) | You already installed YAML from an earlier drop and only need Browse-modal preview PNGs. See placement steps below. |
Install one wildcard YAML variant (do not merge both full and no-Lora YAML sets into user-wildcards/). Extract anywhere, run install.bat at the root of the extracted folder (full or no-Lora zip), paste your ComfyUI path.
That copies:
- Wildcard Clause Expander, PromptAspectSize, PromptLoraExtractor →
custom_nodes/ wildcards/shared+wildcards/krea2→ Impact Packcustom_wildcards/user-wildcards/(merge, does not wipe other packs)- Example graphs →
user/default/workflows/UW-Modular-Wildcards/
Engine 3.1.0 lives in this same download
CivitAI and the zip are v3.0.0. The node reports API engine_version 3.1.0. Same file. There is no second CivitAI version to grab.
3.1.0 is the regenerate button, skip punctuation cleanup (no more , . , or BODY., holes), live preview, and the template catalog. Treat it as the node build inside v3.0, not a separate drop.
What you can do with it
- Lock an adult subject with character atom fields (
age_override,hair_override,body_override, …). There is nosubject_overrideblob anymore. - Steer the shot with scene fields (
wardrobe_override,environment_override,medium_override,lighting_override,treatment_override). - Pick a router (
__uw-krea2-nsfw/router/…__) or a single template (__uw-krea2-nsfw/template/…__) from the catalog, or paste a tag yourself. - Iterate: preview → pin / skip / regenerate one clause → queue. Trailing
[3:4]/[9:16]markers are stripped by PromptAspectSize and used as latent width/height. - Optional: PromptLoraExtractor pulls
<lora:…>tags out of the expanded text into LoRA Manager when that node is in the graph.
Install
Requires ComfyUI with custom_nodes/ and ComfyUI-Impact-Pack already installed.
Windows:
- Extract the zip anywhere (Desktop is fine).
- Double-click
install.batat the top of that folder. - Paste your ComfyUI folder path (you can drag the folder onto the window).
- Confirm replace prompts if you already have any of these nodes.
- Restart ComfyUI, then Impact → Refresh Wildcard.
Install-bat-install-example1.png
Agents / non-interactive:
install.bat --yes
or, from the extracted folder:
<ComfyUI>\venv\Scripts\python.exe -B install.py --comfyui "<ComfyUI>" --yes
--skip-workflows skips copying the example graphs. On Linux/macOS, run install.py the same way.
If the installer warns about wildcard_cache_limit_mb, raise that value in Impact Pack’s impact-pack.ini so the YAML corpus stays in full cache mode.
After install, Wildcard Clause Expander is under utils/prompt. Open the example workflows from UW-Modular-Wildcards in the ComfyUI sidebar.
Full copy: INSTALL.txt in the zip.
What’s in the zip
| Piece | What it does |
|-------|----------------|
| wildcards/ | Modular __uw-krea2-nsfw/* and __uw-shared/* YAML |
| wildcards/krea2/*.catalog/ | 358 template thumbnail PNGs (+ catalog.json) for Browse modal previews |
| Wildcard Clause Expander | In-node overrides, live preview, catalog, pin / skip / regenerate |
| PromptAspectSize | Strip trailing [W:H], emit width/height for Empty Latent |
| PromptLoraExtractor | Pull <lora:…> tags out of expanded text |
| workflows/ | krea2-custom-char-turbo.json, krea2-custom-char-raw.json |
Wildcard Clause UI
Graph order in the bundled workflows:
Wildcard Clause → Prompt LoRA Extractor → PromptAspectSizeResolver → CLIP encode (Empty Latent width/height from the resolver).
wildcard-clause-expansion-node-example1.png
Scene group (wardrobe, environment, medium, lighting, treatment). Collapse it when you are only changing the person.
Character group (age, hair, body, face, skin tone, and the rest of the atom fields). This is the replacement for the old subject blob and the old rebuild loop.
Live preview is structural (routers/templates unwrap, terminal pools stay as tags until populate). Populated text is the fully expanded prompt for the current seed.
wildcard-clause-expansion-node-example2.png
Browse templates & routers opens a catalog with preview thumbnails when the matching *.catalog/template/*.png files are installed (bundled in the full / no-Lora zips, or via the separate thumbs-only zip). Picking a card replaces wildcard_text. There is also a starter search and the full Impact-style insert-at-cursor list.
Each override field has Skip, Pin, and ↻. Skip omits the clause. Pin locks the ghost. ↻ draws a new line from that slot’s wildcard pool only.
wildcard-clause-expansion-node-example3.png
First open: load the Turbo workflow, leave the default bedroom router, fill two or three character atoms (age, hair, body), queue. Then hit ↻ on lighting or wardrobe and queue again.
How prompting works
Router (__uw-krea2-nsfw/router/portrait-bedroom-pics__) picks among templates in that room/theme. Single template (__uw-krea2-nsfw/template/bedroom-portrait-standing-mirror__) locks one scene shape. Theme routers such as lingerie and swimsuit keep wardrobe in a tighter pool.
Overrides replace individual clauses in the expanded prose. They do not rewrite the YAML files.
If a template ends with [3:4] or [9:16], PromptAspectSize strips that marker and sets latent size. Do not leave the marker in CLIP text.
Identity is atom fields only. One long “a japanese woman in her twenties with …” string is the v2 subject_override pattern. Split it:
age_override = a woman in her twenties
nationality_override = japanese
hair_override = Her long straight black hair falls past her shoulders
eye_color_override = green
Breaking changes from v2 (re-save workflows)
subject_overrideremoved. Use character atom fields.shot_overrideremoved. Usemedium_overridefor selfie / candid / portrait / illustration openers.- Re-save any graph that still shows a stale widget for those names.
- The old sidecar / twin-YAML character-kit path is not how this pack works anymore. Overrides live on the node (or as Scene / Character JSON presets under Impact
custom_wildcardsif you want named libraries on disk).
Starter workflows and LoRAs
Installer drops graphs into user/default/workflows/UW-Modular-Wildcards/.
- Turbo (
krea2-custom-char-turbo.json) for fast Krea 2 Turbo queues. - Raw (
krea2-custom-char-raw.json) for the higher-quality stack. Raw also expects the extra nodes/VAE listed in the graph Note (rgthree, RES4LYF, LoRA Manager, RealVAE).
Full zip: both graphs prepend five photoreal <lora:…> tags in wildcard_text. PromptLoraExtractor sends them to LoRA Text Loader — missing files error.
No-Lora zip: workflows start with the router tag only. No companion LoRA files required.
Wildcard YAML (full zip only): solo fingering / toy / masturbating templates embed two sex-act LoRAs (table below). The no-Lora zip removes those embeds.
Companion LoRA download table (full zip)
Download from CivitAI, then rename to the filename in Place as (paths relative to your ComfyUI models/loras/ or Stability Matrix Data/Models/Lora/).
| Used in | <lora:…> path | Strength | Trigger | Place as | CivitAI model | Version |
|---------|-----------------|----------|---------|----------|---------------|---------|
| Workflow photoreal stack | krea2/nsfw/snofs_krea_v1_3D | 0.6 | — | krea2/nsfw/snofs_krea_v1_3D.safetensors | SNOFS | Krea 2 v1.3D |
| Workflow photoreal stack | krea2/styles/photoreal/style-photoreal__Realism_Engine_Ideogram_4_Krea_2 | 0.5 | — | krea2/styles/photoreal/style-photoreal__Realism_Engine_Ideogram_4_Krea_2.safetensors | Realism Engine Ideogram 4 + Krea 2 | Krea2 v3.0 |
| Workflow photoreal stack | krea2/sliders/slider-photo__Photo_Detail_Slider_for_SNOFS_Krea | 0.5 | — | krea2/sliders/slider-photo__Photo_Detail_Slider_for_SNOFS_Krea.safetensors | Photo/Detail Slider for SNOFS Krea | v1.0 |
| Workflow photoreal stack | krea2/styles/photoreal/style-photoreal__Mrpopo_Photoreal_Enhanced | 0.6 | — | krea2/styles/photoreal/style-photoreal__Mrpopo_Photoreal_Enhanced.safetensors | Mrpopo's Krea2 photoreal enhanced | Mrpopo photoreal ultimate |
| Workflow photoreal stack | krea2/face/face-expression__Snapchat_Face_Helper | 0.5 | snapchat_style_pose (prompt body) | krea2/face/face-expression__Snapchat_Face_Helper.safetensors | Snapchat Face Helper | v1.0 |
| Wildcard solo fingering / masturbating | krea2/nsfw/acts/nsfw-act-masturbation__krea2_sex | 0.75 | zitmast | krea2/nsfw/acts/nsfw-act-masturbation__krea2_sex.safetensors | krea2 sex 淫模 | mast |
| Wildcard solo toy | krea2/nsfw/acts/nsfw-act__krea2_sex_淫模_krea2_dildo_in_pussy | 0.75 | kreadildoinpussy | krea2/nsfw/acts/nsfw-act__krea2_sex_淫模_krea2_dildo_in_pussy.safetensors | same model ↑ | ldo |
| Casting router (manual load) | (not embedded) | ~0.8 | — | krea2/detail/detail-casting__Casting_Photo_Character_Sheet.safetensors | Casting Photo / Character Sheet | v3 |
CivitAI download filenames differ (e.g. krea2 masturbation.safetensors, snofs_photoSlider_000000200.safetensors). Rename after download so the <lora:path> in the pack matches the file on disk.
Templates that pull embedded act LoRAs (full zip): action-solo-fingering, action-solo-fingering-mirror-selfie, fantasy-action-solo-fingering, fantasy-action-solo-masturbating (masturbation LoRA); action-solo-toy, fantasy-action-solo-toy (dildo LoRA). Routers such as portrait-intimate-action-pics can hit these.
If a LoRA is missing: LoRA Text Loader errors on the path. Fix: install + rename, use the no-Lora zip, avoid those templates, or delete the <lora:…> lines from wildcard_text (workflows only).
Catalog thumbnails (optional separate download)
Full and no-Lora all-in-one zips now ship 358 pre-generated Krea 2 template thumbs (~228 MB) under wildcards/krea2/*.catalog/. install.bat merges them next to each *-templates.yaml so Wildcard Clause Browse shows preview cards instead of striped placeholders.
Thumbs-only zip (uwModularNSFWWildcardsPhotoreal_v30CatalogThumbs.zip) — if you already installed YAML and only need images:
- Extract the thumbs zip.
- Merge each
krea2/*.catalog/folder into Impact Packcustom_wildcards/user-wildcards/krea2/(sibling to the matching YAML). - Restart ComfyUI (or refresh the catalog index).
- Open Browse — cards should show images.
Example:
From zip: krea2/krea2-nsfw-templates.catalog/template/bedroom-portrait-standing-mirror.png
On disk: .../user-wildcards/krea2/krea2-nsfw-templates.catalog/template/bedroom-portrait-standing-mirror.png
Sibling: .../user-wildcards/krea2/krea2-nsfw-templates.yaml
Router cover PNGs are not in this release; template thumbs only. Missing PNG → striped placeholder for that card.
To regenerate thumbs yourself (FinePorn v4 on ComfyUI-Clean), see ComfyUI-WildcardClause/README.md → Generating thumbnails.
Starter tags
Routers (mixed templates per room):
__uw-krea2-nsfw/router/portrait-bedroom-pics__
__uw-krea2-nsfw/router/portrait-intimate-action-pics__
Theme routers (wardrobe locked to lingerie / swimsuit):
__uw-krea2-nsfw/router/portrait-lingerie-pics__
__uw-krea2-nsfw/router/portrait-swimsuit-pics__
One template:
__uw-krea2-nsfw/template/bedroom-portrait-standing-mirror__
Paste into wildcard_text on Wildcard Clause, or pick the same names from the catalog.
Showcase images
The two images on this version are FinePorn v4 Turbo, in-node character overrides (catalog-girl atoms, including the chest tattoo), and light skin-beauty post. One is a balcony railing selfie, the other a boudoir vanity portrait. PNG metadata is the expanded prompt plus a short generation footer, not a dumped ComfyUI workflow graph.
Requirements
- ComfyUI
- ComfyUI-Impact-Pack
- Krea 2 Turbo or Raw models for the bundled graphs
- Full zip: companion LoRAs in the table above (or use the no-Lora zip)
- Windows
install.bat, orinstall.py --comfyui … --yeson any OS - Adults 18+ only
Files: uwModularNSFWWildcardsPhotoreal_v30AllInOne.zip (full) · uwModularNSFWWildcardsPhotoreal_v30AllInOne_noLora.zip (no embedded LoRA) · uwModularNSFWWildcardsPhotoreal_v30CatalogThumbs.zip (thumbs only)
Questions and breakage reports in the comments.
Description
v1.0.0 — Initial public release
Framework: modular L1→L2→L3 natural-language wildcards for Krea 2 photoreal · Impact Pack · agent-ready docs · Basic + Advanced (+ optional aspect node)
Included
Shared atoms — common / SFW / NSFW / fantasy pools
SFW portraits — lifestyle / selfie routers
NSFW rooms — bedroom, bath, hotel, apartment, living room, closet, kitchen, balcony, poolside, boudoir
Themes — lingerie, swimsuit, cheerleader, plaid uniform, cosplay, nature, librarian
Intimate action — solo, partner, bus, landscape
Group action — multi-person scenes
Aesthetic pilot — editorial / soft-glam
Fantasy core — forest / castle / tavern / ruins (+ dark vignettes)
Fantasy selfie — phone fantasy portraits
Fantasy action — intimate fantasy + landscape
Fantasy monster / epic / group — size-diff, scenic nude, multi-partner
Dual-image — before→after side-by-side
Casting sheet — multi-pose ultrawide sheets
Also
Advanced aspect markers
[W:H]· companion PromptAspectSize node · contribution-friendly new-YAML extensions
FAQ
Comments (20)
This is my first ever post on Civit! So feel free to message me if I need to change anything. I generated all of these images in like an hour. Almost everyone of the images generated was a great image because of how great the prompts are + the good workflow I am using, but theres no Post processing.
Edit.
Hey everyone, for Advanced mode to work you need my custom Prompt Aspect Size node. https://civitai.red/models/2789943/uw-custom-node-prompt-aspect-size-resolver-wildcard-wh-latent-wh-comfyui?modelVersionId=3143923
The idea is that all of the prompts have their Aspect Ratio embedded into the end of the prompt. So the special node I have will extract that AR from the prompt and then output the image Width + Height based on the MegaPixel size you choose!
Thank you!
Could you also add templates without subject modifying parts? So that it can be used with character loras.
Oh interesting!
I thought it through and I just came up with a pretty neat solution I think to handling the problem. Let me know what you think.
The main issue is that according to kreas prompting guidelines, you are supposed to put the subject phrase within the prompt, not the very beginning or end. So it makes it hard if you are trying to do something like have a hard coded string describing your subject then be clumped together with the description of their actions and setting and photo type, etc etc.
My solution is going to be a template character file. So you will take the template character file and then fill in the section at the top for the subject description. Then you can use that template as a new wildcard that will be able to pull from all of the other references in the wildcards shared data pack.
Does that sound like something that fits what you are needing?
@Vesperra Yes that could work. It should be a simple description though like trigger word + hair + make-up like things that can change about a character. Don't need to randomize hair and make-up, user can change the template if they want a different variety.
@hadeng
Hey yup thats what Im working on now! I tried it out with a "Starfire" character description. So the idea is that the "Subject clause" will be a custom field as apart of the template.
The subject clause defines the characters general look not including clothes (face, age, gender, body descr).
So this new Subject file should be compatible with all the other existing photo templates in my pack. It would take the subject and then put them into the photo with a randomized environment, clothes, and 'action'
If you want a custom Subject + Clothing Description that is going to be released as a separate file. The reason is because not all of the Templates are compatible with a custom Clothing description: like if your character is wearing a full outfit, then that will be incompatible with a prompt that is trying to have the character do something that requires them to be nude.
But yeah I am going to try to release those two things within the next day or maybe few hours. I just need to finish testing it.
@Vesperra Great! Though, if the subject clause includes attributes like face, race, age, gender, and body, does that mean users would need to describe the character for every LoRA when using a template? It might be better to make those optional, so users can keep things simple by just adding the trigger word.
Looking forward to it!
@hadeng Hey I implemented my solution to this. To answer your question from before, it is up to you. Lol the prompt and Loras just push your output in a certain direction. like if I have a Starfire character Lora, then the only subject description I may need is just "a photograph of a woman dressed as Starfire".
But if you dont have a Lora and you want an exact character look then you will probably want to give a very detailed description.
Anyways, hopefully this can help.
@Vesperra Thank you! It took an effort to generate files but it worked great after that.
I had to edit the build file as I use the portable version.
$Python = Find-UpwardPath -StartDir $CompanionRoot -RelativePath "venv\Scripts\python.exe" if (-not $Python) { $Python = Find-UpwardPath -StartDir $CompanionRoot -RelativePath "python_embeded\python.exe" } if (-not $Python) { throw "Python not found (looked for venv\Scripts\python.exe and python_embeded\python.exe)." }
@hadeng Ahhh I see. So the problem is that it had a failure related to not being able to use python cause it was relying on an incorrect .venv? Thats poor design on my part, so I will fix that, thanks! That is good info to know though, Ill make some changes to the build file. Please post your gens :) I love to see the girls people can make!
Cheers
Ok, so, this is pretty cool. A LOT of work, I'm sure. Anyways, I don't want to take my frustration out on you, so, consider this a cry for help. This thing is VERY complex. It's almost impossible to try and deal with manually. So, I've thrown the biggest model i can manage at it, a qwen 34b model, and it's still not enough. It just takes a big-ole dump anytime I try and add custom files into the mix. It's just too much, and I'm not going to risk my claude or openai accounts on nsfw content. What sort of model are you using with this? It must be at least something 100b or more. Any suggestions before i just rm -f the whole thing?
Thanks man! So are you not able to run it normally just using the built in templates? Im not sure exactly what you mean by 'adding in custom files'. (I assume you mean you are trying to add in your own custom template file that will then utilize the existing wildcard package).
Also I am going to reply back now with a bunch of different things so you can reply to them separately and make the flow of conversation easier.
Also can I give a different suggestion?
Use Cursor. I did thorough research using OpenAI to research the license requirements for Cursor before I used it with this and they have nothing that limits you from NSFW content. They are allowed to ban for 'obscene' content I think, but in comparison OpenAI and Claude have actual explicit wording in their license that bans NSFW... So based on that if Cursor didnt want you to do anything NSFW then I believe they would have it in their wording.
This entire thing was developed using Composer-2.5 and recently Cursor-Grok-4.5 (new model) through Cursor.
I already use Cursor for my normal life for daily coding + other work too, its top tier.
but the entire architecture of this thing is made using to be Agentic and work with Agents. Its why I have AGENTS.md files in each of the folders. Its very easy for the AI to understand the structure + understand how Krea2 is required to be prompted and then construct really nice Natural language wildcards from it.
Also. For open source LLMs I actually have used them also to make changes.
Qwen 3.6 35B MoE ablitterated
and theres a newer one that I would probably recommend:
https://huggingface.co/dealignai/Bonsai-27b-Ternary-CRACK-GGUF
Bonsai took the Qwen 35B model and they did some AI engineering magic and they were able to Quantize the original Qwen 3.5 Model so that the Q2 quant has similar intelligence to the Q8 quant.
So the point is: Its now like a 5GB model that has 95% the intelligence of the original model but at 1/8th the size. If you had problems with context window size then using something like that maybe much more helpful.
+ The CRACK is a form of Ablitteration, so the model will do NSFW stuff
I used the Qwen 3.6 35B MoE model already to make changes to my wildcards before. I had it read in the context of the Wildcards Shared Data files and then make changes to my existing NSFW templates to essentially expand on them and make them 'sexier'. So it definitely is possible to do, I just dont know what hardware you are running.
Hope some of that helps! Feel free to ask if you still need help.
@Vesperra For clairity, yes, I'm attempting to add files to the structure via agent, specifically pi agent via local LLM, you've made good llm suggestions here, both cursor and bonsai, i'll poke at that. thanks for the response.
@gurilagardnr Hey I just released a character expansion pack system that may help with what you need! https://civitai.red/models/2807882/custom-character-uw-expansion-pack-nsfw-modular-natural-language-wildcards?modelVersionId=3166214
Thanks, man. This one came highly recommended, installed it, and it's working great. The pics are of great quality, but I am struggling to understand how to use it correctly. I have seen the paths in yaml files and can use them but is there a tutorial somewhere, that will help? Thanks again.
Hey thank you for the support! If you dont already I also recommend the Advanced mode so the prompt will pass along the aspect ratio from the prompt to the ComfyUI workflow!
Can you explain a bit more how you are trying to use it / what you are trying to do? Im guessing you are trying to make modifications to add in new Templates?
I would say this: there are two things I assume you would want to do.
1. Either add in new templates yourself
or
2. Edit the existing shared data.
So for 2 for example... All of the templates pull from existing Shared data groups. So whenever it creates a "subject" clause for example, it will pull in a term for Age + Body Type + Ethnicity + Hair.
So if you want to change how the subject looks, you just need to go to these shared Data term groups and modify them (so you could add in more hair styles there, more age descriptions, body types, etc).
Bro, I got it. I was using the wildcards and somehow the custom wildcards (these ones) that I pasted in the folder were not showing up in the node, so I was under the impression that i will need to type the whole thing manually and that will be hard to remember - with the complete syntax and then needing to know what's available. But after a couple of sessions and comfy relaunches, I noticed that they are all available in the dropdown and I can even search it from the wildcard node itself. Works flawless and man, the variety you have here is amazing!!!
I thank you for your hard work for compiling these and making it available to the community, I'll be sure to spread the word around. :)
@Pitiful_Season42942867
Thanks! Make sure you are using my new V2 version for the Custom Character feature I released too. It lets you customize the subject within the workflow so you still get the randomized poses + envs + outfits, but it will have the same Suject prompt locked in. It is wayyyyyyy easier compared to the v1 that I released for it.
Im glad you like it! If you find any full prompts that do not seem to have an equivalent in the wildcards feel free to message me or post a comment here and Ill add the general prompt into the template.