LewDakaExplicitIllustraIL

This checkpoint has its own style, so avoid using other artstyle Loras with it, as they may not work properly, however, you can use negative embeddings and other Loras, just not artstyle Loras.

Reminder: for the characters, this was my older model for V1, since it’s mostly based on older character data. For newer characters, you’ll still need a character LoRA to help it out a bit. However, you can find information on which characters it supports here: https://huggingface.co/datasets/flagrantia/character_select_stand_alone_app/tree/main

Download: current_character_list_v160.txt

You can use lighting, character, and detail Loras alongside it. If you’re using a character Lora, remember to never include the copyright tag. Using a copyright tag often tries to force the image to look like the original, which usually results in odd outputs.

⚠️ Caution: Be careful when using this model in ComfyUI, as it was specifically designed for Stable Diffusion Forge and Neo Forge. It is recommended to follow the provided settings or refer to the reference image for correct configuration. Deviating from these settings may lead to unexpected results.

Avoid merging it with other models, as it is not intended for such modifications. Doing so may significantly degrade image quality, potentially resulting in low-contrast or corrupted outputs.

OH Remember: This was specifically made for Stable Diffusion WebUI:
https://github.com/Haoming02/sd-webui-forge-classic/tree/neo
using it on generation sites like PixAI:

or

TensorArt:
Using sites settings may not work properly there, and the images might end up looking very strange or inconsistent. OH: (These weren’t shared by me, by the way, so I didn’t share them to the site, someone else did).

Also, using art style LoRAs with it may not work very well, and like i mention since it already has its own built-in art style. Try to avoid using art style LoRAs. However, anything that does not involve an art style.

Recommended Prompts:

Positive Prompt:

masterpiece, ultra‑HD, high detail, depth of field, best quality, 8k, best quality, ergonomic, (anime coloring, anime screencap:1.5), official art,

Negative Prompt:

Not Needed - BUT you can add: bad quality, worst quality, worst detail, sketch, censor,

Recommended structure for your character Lora and tags:

<Character Lora here> [Character name tags here]
[Character Appearance tags here]
[Emotions and Poses tags here]
[Lighting tags here]
[Background tags here]

This structure is especially helpful if you’re writing your own prompt structure, but anything following it will still work in a random order.

BUT there are two ways of doing the layout structure. You can do it like this:

OR you can do it like this, which I recommend:


So it’s more like upside down compared to right-side up. Either way works, but try the upside-down layout first, often doing them a bit differently for the layout makes the images a bit different. It’s a bit odd, often the eyes, clothing, and background change slightly.

Oh, and one last thing: try not to rely on overly short tags, since I explained that here.
https://www.deviantart.com/lew-daka/art/Short-Tagging-vs-Long-Tagging-Info-Below-1323452292
So be mindful of avoiding very short tags. Character tags are fine, but for more detailed tagging, well, you can refer to what I mentioned on the post I shared about using more tags for quality and using a bit more descriptive tags.

Recommended settings for Forge or NeoForge:

Steps: 30

Sampler: Euler a

Schedule type: Simple

CFG scale: 5

Size: 1024x1408 - Use -->1024x1408 for original dimensions

Hires.fix settings:

Hires upscale: 2 or --> 1.5 can be used as an optional upscale to speed up generation. Using 2 often takes a bit longer, so try 1.5 when you want faster results.

Hires steps: 10

Denoising strength: 0.3

Hires Module: 1 --> Use same choices

Hires CFG Scale: 5

Hires upscaler: 4x_foolhardy_Remacri or 4x-UltraSharp

Diffusion in Low Bits: Automatic (fp16 LoRA)

Clip skip:2

ADetailer model: face_yolov8n
VAE not needed. - But you can use this VAE: https://civitai.red/models/1623082/forges-built-in-vae. It often helps.

eps_scaling_factor: 1.05 - Forge Neo --> Settings/(Search): Epsilon Scaling - here's an example of what it does and why I recommend it:
https://www.deviantart.com/lew-daka/art/SD-Neo-Forge-PR-Testing-Disabled-vs-Enabled-1317802635

SD-Neo Forge: GPU/CPU/NV --> info on what you should try. There’s a post I made about that here:
https://www.deviantart.com/lew-daka/art/SD-Neo-Forge-GPU-CPU-NV-Testing-Info-Below-1325527711

You can refer to the images for the settings if needed:

all Images are generated using Stable Diffusion WebUI Forge - Neo.

Also, to fix the double eyelid issue, I recommend using Reduce Wrinkles for Illustrious:
https://civitai.com/models/1126192/for-illustrious-and-noob-reduce-wrinkleslines-above-eyes
Try setting the weight value to 1.5 for a more aggressive effect, or 0.8–0.9 for a moderate fix. I usually stick with 1.5 or 1, but I recommend trying 1.5 first.

If you’re having the finger or toe issues: I recommend using these 2 lora files:
https://civitai.com/models/1496218/toes-adjuster-for-illustrious-and-pony

https://civitai.com/models/1515981/fingers-adjusters-for-illustrious

For the fingers LoRA, use version 2.0.
For the toes LoRA, use the Illustrious version.

Set the weights as follows:

Fingers weight value: 0.8

Toes weight value: 0.3

Also, add the tag hands, to the positive prompt as well. This helps fix it too.