Character Select SAA

This is a Stand Alone App with AI prompt, Semi-auto Tag Complete and ComfyUI/WebUI(A1111) API support.

Important

Create your own thumb SAA Thumb Generator

Tip

If you find a character that isn't show on the list but can be generated correctly, please don't hesitate to let me know.
The default list v160 Remove invalid and duplicate (5149) is based on waiIllustriousSDXL_v160.
If you are interested in the v120 (5327) list, navigate to HF download v120 files and replace data/wai_characters.csv data/wai_character_thumbs.json

For browser based SAAC information, check README_SAAC.md

For Python based CLI tool for OpenClaw SAA Agent and ClawHub

Item Support	ComfyUI	A1111	Forge Neo
LoRA	Yes	Yes	Yes
BREAK	No	Yes	Yes
Refiner	Yes	Yes	Yes
Image Color Transfer	Yes	No	No
Regional Condition	Yes	Yes	Yes
ControlNet/IPA	Yes	Yes	Yes
ADetailer	Yes	Yes	Yes
API authentication	No	Yes	Yes
MiraITU	Yes	No	No
UNET Model	Yes	No	Yes
Custom VAE for SDXL	Yes	Yes	Yes

Try Online Character Select Simple Advanced App Hugging Face Space

Install and run

Important

Setup your Image API Interface before you start SAA.

For ComfyUI, you need ComfyUI_Mira for SAA.

Clone this repo into your local folder

git clone https://github.com/mirabarukaso/character_select_stand_alone_app.git
cd character_select_stand_alone_app
npm install
npm start

Tip

One-Click package v2.6.5
The full package embeded_env_for_SAA

Update

The One-Click package may not the latest version. If you need to update, please use the GitHub clone version with following command instead.

git fetch
git pull
npm install

Important

Updating version from github will not update the database files danbooru_e621_merged.csv and wai_character_thumbs.json.

Update to the latest version, then manually delete the danbooru_e621_merged.csv and wai_character_thumbs.json file. Restart the app to automatically download the latest thumbnail database from HF.

---

Highlights

Diffusion Models (UNET/CLIP/VAE) Support

Important

These models are significantly difference from SDXL in both prompts and sampling configuration. Please make sure you understand their parameters and settings!
Test and Verified Models: Anima / Qwen Image / Z Image / Flux Supports ComfyUI and Forge neo, NOT support original A1111.

For ComfyUI, check Comfy-Org@HF
GGUF model requires custom node gguf the small-case one.

|---models
|   |---checkpoints
|   |---diffusion_models
|   |---text_encoders
|   |---vae

For Forge Neo, check Download Models@Haoming02
Forge also supports the GGUF model, but the Diffusion models use the same Checkpoint folder. Therefore, manage those models with correct folder yourself.

|---models
|   |---Stable-diffusion
|   |---text_encoder
|   |---VAE

Mira Image Tiled Upscaler

Note

MiraITU A content-aware image super-resolution ComfyUI custom node based on vision models.
Simply drag and drop ANY image into SAA/SAAC to upscale it.
Now supports Checkpoint and Diffuser models.

Supports magnification scales from 1.5x to 8x. By adjusting different tile sizes, it can optimize VRAM usage to the maximum, enabling 8GB graphics cards to achieve image upscaling at 1.5x to 3x (or higher) scales, while also unleashing the full potential of high-end graphics cards with 24GB or more VRAM to accomplish 4x to 8x image upscaling.

Requires ComfyUI_Mira 0.5.6.0 or above
Requires MiraSubPack

And Image Tagger for Mira

Settings	Drag and Drop	Flux.2

Before	6x After (SDXL)	Before	3x After (Flux.2)

The generated results will be affected by different models. The SDXL model requires a tagger to provide more precise content descriptions, and using an upscale model is recommended. Models such as Flux2, which allow a reference latent to connected, can skip the upscale model and stretch the original image directly. Then, simply use the Positive Prompt to configure all tiles consistently.

Upscale Ratio 1024x1360(1536)	Tile Size	VAE 8G	VAE 24G~
x1.5	768	Full	Full
x2	1280	Full/Tiled	Full
x3	1536	Tiled	Full
x4	2048	Tiled	Full
x6	2048/2560	Tiled	Full
x8	2560/3072	Tiled	Full/Tiled

Note: Due to limitations of the SDXL model, exceeding 8x magnification will result in overly fragmented visual slices. For higher magnification requirements, it is recommended to use a dedicated ComfyUI workflow instead.
Comfyui Workflow MiraITU

Image Tagger

Supports WD@SmilingWolf, CL@cella110n and Camie@Camais03 models in ONNX format.

Download models with tags from HF, manually rename them according to the following rules, then copy them into models/tagger folder:

• cl_tagger_v2.onnx + cl_tagger_v2_tag_mapping.json
• wd-eva02-large-tagger-v3.onnx + wd-eva02-large-tagger-v3_selected_tags.csv
• wd-v1-4-convnext-tagger.onnx + wd-v1-4-convnext-tagger_selected_tags.csv

SAA
|---models
|   |---tagger
|       |---cl_tagger_1_02.onnx
|       |---cl_tagger_1_02_tag_mapping.json
|       |---wd-eva02-large-tagger-v3.onnx
|       |---wd-eva02-large-tagger-v3_selected_tags.csv
|       |---wd-vit-large-tagger-v3.onnx
|       |---wd-vit-large-tagger-v3_selected_tags.csv
|       |---camie-tagger-v2.onnx
|       |---camie-tagger-v2-metadata.json

Options:
Model Name  >>>  General Threshold(CL/WD/Camie)  >>>  Character Threshold(CL/WD)  >>> Categories(CL/Camie) or mCut(WD)      

The Image Tagger running on Node.JS with onnxruntime-node. It DOES NOT require any backend support But, GPU acceleration seems not working
The Generate Speed is about 3 times slower than Python with onnxruntime in CPU mode, and 12 times slower than onnxruntime-gpu

In other words with my i9-9960x with Titan RTX
The good news is, you can run Image tagger during gegenerate

Device	Avg Tagging Time	Model	Platform	Resolution	Recommend Value
onnxruntime	1.053s	cl_tagger_1_02	Python	448	0.55/0.60
onnxruntime-gpu	0.297s	cl_tagger_1_02	Python	448	0.55/0.60
onnxruntime-node	3.185s	cl_tagger_1_02	Electron(NodeJS@CPU)	448	0.55/0.60
onnxruntime-node	2.917s	wd-eva02-large-tagger-v3	Electron(NodeJS@CPU)	448	0.35/0.85
onnxruntime-node	2.113s	camie-tagger-v2	Electron(NodeJS@CPU)	512	0.50/(NOT USE)

ControlNet / IP Adapter

For ComfyUI
Upgrade your ComfyUI_Mira version to 0.4.9.6 or above

ControlNet requires comfyui_controlnet_aux
Put your ControlNet models in ComfyUI\\models\\controlnet

IP Adapter requires comfyui-art-venture and ComfyUI_IPAdapter_plus
Put your Clip Vision models in ComfyUI\\models\\clip_vision
Put your IP Adapter models in ComfyUI\\models\\ipadapter
I didn't test too much on IPA, but for SDXL/ilXL/NoobXL recommends CLIP-ViT-bigG-14-laion2B-39B-b160k.safetensors with ipa_styleIpadapterFor_NoobAI-XL_v10.safetensors. You may also need CLIP-ViT-H-14-laion2B-s32B-b79K.safetensors
Only the first IP Adapter slot will accept by ComfyUI, others set to On will ignore.

For A1111 and Forge Neo
For ComfyUI and A1111, Post directly feed the processed image to controlnet model without requiring Processor Model preprocessing. They accepts none as preProcessModel.
But Forge based controlnet DOES NOT support none as preProcessModel, it accepts None. Unfortunately, it's impossible to determine whether it's A1111 or Forge from API perspective, so WebUI uses On by default in all cases. Choose the proper none or None for A1111 or Forge yourself.

A1111:
Requires sd-webui-controlnet.
Put your ControlNet and IP Adapter models in stable-diffusion-webui\\extensions\\sd-webui-controlnet\\models, the extension plugin path.
Use none

Forge:
Put your ControlNet and IP Adapter models in models\\ControlNet.
Use None and always On

Usage:

1. Drag and drop (or click Add then Paste) your Image(or openPose image) to SAA/SAAC Image Info, select Pre-processor, Resolution, Post-processor and then click Add ControlNet. After it says Added the previre image will swap to your Pre-processed image, close Image info window, check ControlNet tab for more settings. Hover your mouse over a dropdown/text item to view its feature.
2. In ControlNet tab, you can change Pre-processor by select the a one and click Refresh to generate it and update preview. (Or set Method to On but SAA preview will not update).
3. Because images are too big to save in settings file, so ControlNet settings will not save when SAA close, but also select another settings will not override current ControlNet settings.
4. ControlNet works on normal and regional conditon.
5. If you are new to ControlNet, try Canny or OpenPose first.
6. ComfyUI D-O-E-S N-O-T like submitting the same data, you may receive an Empty response error when submit same data.
7. ComfyUI comfyui-art-venture requires a square image, you may get warnings with NOT square image input.
8. In case of your IPA image is too big, the Resolution selection for IP Adapter will resize your input image to target size, 1024 is enough for most case.
9. The Info button is not working when Pre-Process Model select to IP Adapter

All Pre-processor models are managed by comfyui_controlnet_aux(ComfyUI) and sd-webui-controlnet(A1111), most models will download from Hugging Face.
All Post-processor models, aka the Apply ControlNet Model you need download by yourself from ComfyUI Model Manager or Hugging Face.

LoRA Slot

WebUI(A1111) supports it's default LoRA prompt style <lora:xxxxx:1.0>.
ComfyUI supports more detailed configuration of LoRA, for more information please refer to LoRA from Text.
Also support check LoRA info by click the 'i' button in LoRA Slot. And, if there's a same named PNG file with LoRA, the image will show in LoRA info page.

To use LoRA in ComfyUI API, you need update your ComfyUI_Mira node to at least 0.4.9.2

ADetailer

Caution

Check before you download any .pt file from unknown/untrusted site! ltdrdata/ComfyUI-Impact-Pack#843

Note

If the settings are too confusing, just remember to adjust the gold parameter in the bottom-right corner (Denoise).

For ComfyUI
ADetailer requires Impact Pack and Impact Subpack
Put your ADetailer models in ComfyUI\\models\\ultralytics\\bbox
Put your SAM models in ComfyUI\\models\\sams

For A1111/Forge Neo
ADetailer requires ADetailer Plugin
Upscaler, Control Processor, ADetailer lists have to be read from the API.
The default ADetailer model list will be updated after the first generation. Simply start generating an image as normal.
Put your ADetailer models in sd-webui-forge-neo\\models\\adetailer or stable-diffusion-webui\\models\\adetailer

Queue Manager

The queue management system enables you to submit multiple generation tasks, each with their own distinct parameters. Once submitted, the queue automatically begins processing them and removes completed tasks.
If an error occurs, or if you manually unchecked Enable Generation, the queue will pause after the current task has finished and will remain preserved.
You can delete or view details of tasks within the queue. Deleting the first task in the queue cancels the current generation process.
Recommended that the length of the queue should not exceed 10,000.

JSON/CSV List

Support *.json and *.csv files, just drag and drop (or click Add then Paste) those file info Image Info window. File format please refer to wai_characters.csv and wai_tag_assist.json, try drag them into SAA.
__Random__, randomly selects an item from the list without a seed bound, works for Single and Batch (Random) generate mode.
__Enumerate__, enumerates every item one by one and only works in Batch (Random) mode, in Single it downgrade to __Random__

Wildcards

Supports *.txt wildcard files, copy your wildcards into resources\app\data\wildcards
By default, wildcards are randomly selected using the current seed. If wildcard random seed is Checked, a new random seed will be generated for every selection every time.
Subfolder is not supported

In case you didn't like wildcards file or json/csv wildcard, try the following in your prompts:

{ standing | sitting | on stomach | on back }
{ red | green | blue | blonde } { {long | short} hair | eyes}

Regional Condition

Try SAA Regional Condition with only 3 steps:

1. Click the Regional Condition Checkbox
2. Choose listed character or your OC
3. Start common prompt with duo, masterpiece, best quality, amazing quality(Don't forget quality words like me), have fun!

For ComfyUI
Get tired of complex workflow?

For A1111/Forge Neo
Regional Confition requires sd-webui-regional-prompter

Note

Image Left/Right Ratio is the only working parameter for A1111
The generated results may differ significantly from those using ComfyUI

Semi-Auto Tag Complete

Tags credits from a1111-sd-webui-tagcomplete

Entering the first few characters will automatically search for matching tags, starting with * will search for tags with a specific ending, and *tag* will search for tags that match in the middle.
Use mouse to select the prompt word, but you can also use the keyboard up and down with Enter or Tab to select, press Esc to close the prompt box.
ctrl + up and ctrl + down arrow to adjust the weight of the current tag, or adjust multiple tags by selecting a fragment, the usage is similar to comfyui and webui, but some details of the logic may be different.

Supports English and Chinese(translate file required) tag search.
Special thanks to Kiratian(天痕) for helping to translate the tags into Chinese version.

Artist Search for Anima and others Activate it using the @ symbol. The result will be filtered by groups 1 and 8.
To apply the Artist tag in Anima Model, you need to add an @ symbol at the start of the tag. E.g. mira -> @mira

Mark	ID	Category	Group
[G]	0	General	Danbooru
[A]	1	Artist	Danbooru
[©]	3	Copyright	Danbooru
[C]	4	Character	Danbooru
[M]	5	Meta	Danbooru
<G>	7	General	E621
<A>	8	Artist	E621
<©>	10	Copyright	E621
<C>	11	Character	E621
<S>	12	Species	E621
<M>	14	Meta	E621
<L>	15	Lore	E621
Wildcards	255	Wildcards	SAA

Image info

Drag and drop your image into SAA window, supports Png/Jpeg/Webp.
Works both for WebUI(A1111) and ComfyUI(with image save node from ComfyUI_Mira).
Double click the image to close.
The Send button will override Common Prompt, Negative Prompt, Width & Height, CFG, Setp and Seed.
LoRA in Common Prompt also works if you have the same one. If you don't like LoRA in prompts, try Send LoRA to Slot.

Realtime Character preview and Search

The Character List supports keywords search in both Chinese and English.

Right Click Menu

I just noticed that the electron app doesn't have a right-click menu, so I made one.

Spell Check (English)
Right-click on a word that has a spell check error (a wavy line drawn at the bottom) to see a hint for the corresponding word.

AI prompt generate test
Right click on AI prompt to get AI promot without generate.
Once got result from Remote/Local AI, an information overlay will show in screen, switch AI rule to Last to keep the result in later generate.

Copy Image/Metadata
Right click on Gallery to copy current image or copy the metadata to clipboard.
ComfyUI with Image Saver node will output an a1111 like metadata.
Copy image based on convert base64 data back to png, but metadata trimed by chromium core, it's impossible to put them back with chromium API, a C based lib could solve that problem, but it's not worth to do. If you do need the original image, check from the relevant (ComfyUI/WebUI) output folder.
For SAAC: Drag and drop image from browser to local folder or save as from browser right click.

Send LoRA to Slot
Right click on Common and Positive to send text form LoRA to LoRA Slot.

The Top buttons
From Left to right: Save Settings, Reload Model List, Refresh page, Right to Left, Theme Switch.

---

AI prompt

Remote

1. Follow the setup guide to setup your Remote AI url , Remote AI model and API Key
2. Put something in AI Prompt e.g. make character furry, and I want a detailed portrait

Local

1. Make sure you know how to build Llama.cpp yourself, or download them from trusted source
2. Download Model from HuggingFace, recommend GGUF like oh-dcft-v3.1-claude-3-5-sonnet-20241022.Q8_0 (Here)
3. Recommend Server args llama-server.exe -c 16384 -ngl 40 --port <your local LLM port> --no-mmap -m "<your GGUF model here>
4. Set AI Prompt Generator to Local
5. Set Local Llama.cpp server to your Local AI address and port
6. (Optional) You may need to check API settings for any other Local AI service
7. Put something e.g. make character furry, and I want a detailed portrait in AI Prompt

---

Image API Interface

For ComfyUI

Important

A bug has just been found in the ComfyUI front end that prevents the image workflow from being recognised correctly.
Check Comfy-Org/ComfyUI_frontend#6988
If you encounter issues with workflows failing to load correctly, try using 2025-05-03-022732_1775747588.json instead.
Fixed in 1.35.5 and above: pip install -U comfyui-frontend-package>=1.35.5

1. Enable DEV mode in ComfyUI Settings, and load examples\2025-05-03-022732_1775747588.png into your ComfyUI, make sure you have install ComfyUI_Mira v0.4.9.2 or above from ComfyUI Custom Node Manager.
   1.1. You might need install opencv-python by ComfyUI->Manager->Install PIP packages-> opencv-python
2. Select Image API Interface to ComfyUI
3. Make sure Image Interface IP Address:Port same as your ComfyUI page
4. Have fun

In case you have preview issue in SAA, try add --preview-method latent2rgb into your startup BAT file
My setup:

py ComfyUI\main.py --fast --use-sage-attention --cuda-malloc --windows-standalone-build --listen 0.0.0.0 --port 58188 --preview-method latent2rgb

For WebUI(A111/Forge neo)

1. Enable API mode by add --api in COMMANDLINE_ARGS (webui-user.bat)
2. Start WebUI
3. Select Image API Interface to WebUI
4. Make sure Image Interface IP Address:Port same as your WerUI page
5. Have fun

Custom path for 3rd party package

Warning

Enable custom path will override your model path settings.
Not recommend for official WebUI(A1111/Forge) and ComfyUI API.
To enable a custom path, modify the file data/custom_path.yaml.

1. Set use_custom_path to true
2. Set enable to false to disable whole category
3. Comment out to disable any single custom path you don't need to override
4. Supports both single string and multi-line string for path lists
5. Supports absolute and relative paths (relative to base_path)

Folder path issue for Remote usage

SAA needs to search your ComfyUI/WebUI checkpoints folder to retrieve models, LoRAs and other items. If you use a remote back-end address instead of 127.0.0.1, the folder search will fail and SAA will run in Default mode. In this mode, you cannot change the models or set the LoRAs by LoRA slot.
There are two ways to solve this problem:

1. Mirror folder - copy your remote models folder to local, then setup SAA with the local folder. This is simple, but you need more space to mirror the entire models folder to local folder.
2. Symbolic link or Shared folder - Create a Symbolic link or simply setup your remote models folder as shared folder (read-only recommended), then setup SAA with that folder.

Advanced security settings (API authentication)

Warning

DO NOT forward any UNSECURED local port to public internet
WebUI(A1111) ONLY, DO NOT forward Comfyui API to public internet until they create a proper and secured way

Check more WebUI(A1111) command args at Command-Line-Arguments-and-Settings

Copy your webui-user.bat to webui-user-api.bat then edit it with following args.
Replace user:pass to your Username:Password
--api --api-auth Enable API and API authentication.
--nowebui means you don't need the WebUI browser interface.
--port 58189 API Call port to 58189

set COMMANDLINE_ARGS= --xformers --no-half-vae --api --api-auth user:pass --nowebui --port 58189

Start your A1111 with new webui-user-api.bat
Copy and paste your Username:Password to SAA->Settings->WebUI API Auth, then set Enable to ON

---

Hires Fix and Image Color Transfer

Please refer to Image Color Transfer for more details about Image Color Transfer.
Due to lack of generate rule and missing openCV, Color transfer no longer support in WebUI

Makesure all upscaler models located in upscale_models folder.

Notes for ComfyUI:
Comfyui needs to download upscale models by yourself. Select Manager->Model Manager and filter with upscale, then download them.

Upscaler notes for A1111/Forge:
A1111 uses a name-based upscaler model list. The static upscaler list should work, and will update to API list after the first generate.

Forge uses a file-based upscaler model list. But it's messy! IMPORTANT: If the upscale_models folder is NOT exist, SAA will use static upscaler list as A1111
If you're confused about how to do it, don't panic—just run generate once, and HiFix model list will update properly.
The solution:

1. Create a folder called upscale_models inside the models folder and put all your upscaler models in it.
2. Create a symbolic link named after the upscaler model folder, e.g. ESRGAN, which points to upscale_models.
3. Restart your Forge. The Hires Fix model should now work and will update to API list after the first generate.

---

Chinese Translate and Character Verification

Many thanks to the following people for their selfless contributions, who gave up their valuable time to provide Chinese translation and character data verification. They are listed in no particular order.
Silence, 燦夜, 镜流の粉丝, 樱小路朝日, 满开之萤, and two more who wish to remain anonymous.

Character List Special thanks

lanner0403 WAI-NSFW-illustrious-character-select 2000+ Verified Character List.
Cell1310 Illustrious XL (v0.1) Recognized Characters List more than 100+ Verified Character List.
mobedoor #23
UdinXProgrammer #62
Nurimtod #75 #83
atmogenic #84 #85

---

FAQ

Double clicked saa.exe but nothing happen?

1. It might caused by files download issue or missing files.
2. Try run it in console, by input cmd in your Explorer's address bar to open a console.
3. Type saa then enter in console, check backend logs for more information.

I messed up with setup wizard...

1. Close the App
2. Delete settings.json in resources/app/settings
3. Try it again

ERR_CONNECTION_REFUSED

1. In most cases, it's the wrong address for the (ComfyUI/WebUI) back-end API.

A Browser based SAA?

1. YES
2. Check Advanced security settings (API authentication) for more information.

Error HTTP 400 ...... Cannot execute because node StepAndCfg does not exist ......

1. Install ComfyUI_Mira
2. Restart your ComfyUI

Upscale Model list is None (ComfyUI)

1. Have you modified the default directory configuration?
2. A non-official version?
3. Check #58

ComfyUI/WebUI is busy, cannot run new generation, please try again later.
Refer to 5, 6 in README_SAAC.md