This is a merge of Axelros' Tiberium Illustrious Mix and xXMadVladXx's Throwing Pasta (MIX) - Spaghetti.
I've been working on this merge for a while. It's nothing too complicated—mostly a combination of two SDXL models and a few LoRAs. The goal was to create a stylized merge capable of western-style art similar to what you’d get from a Pony model, but with the clean lines and colors of Illustrious.
The merge has solid range—it can handle cartoon, anime, realistic, and semi-realistic styles. I wish I could take credit for that flexibility, but it's probably more thanks to the original creators than anything I did. That said, I did tweak the LoRA weights to give the model a more distinct look.
You have some flexibility with settings, and after a lot of hands-on testing with this model, I can confidently recommend it for both text-to-image and image-to-image workflows.
🔧 Recommended Settings:
Euler_A Settings
A well-rounded setup for general SDXL prompting.
Steps: 30
Sampler: euler_ancestral
Scheduler: karras
CFG (Guidance): 5 to 7
DPM++ 2M SDE GPU Settings
Offers excellent prompt adherence and greater stability in my experience.
Steps: 45
Sampler: dpmpp_2m_sde_gpu
Scheduler: sgm_uniform
CFG (Guidance): 2.5
Some of the quality tags I use are:
🏷️ Recommended Quality Tags
These are some of the prompt tags I’ve used consistently with good results:
masterwork, masterpiece, best quality, (masterpiece:1.2), (best quality:1.2), (very aesthetic:1.2), (absurdres:1.2), newest,
For an Anime style I've used:
masterpiece, best quality, anime screencap, anime coloring, very aesthetic, absurdres, amazing quality, scenery,
For Realistic, I've used:
masterpiece, newest, high quality, best quality, very awa, very aesthetic, ultra detailed, realistic, close up, hyperrealistic, depth of field, absurdres, volumetric lighting, tack sharp focus, beautiful composition, intricate detail, professional illustration
or
masterwork, masterpiece, best quality, dynamic lighting , dynamic pose, dynamic angle, (realistic:1.2), (lineless:1.2), detailed,
🔧 Image-to-Image Tips
For upscaling workflows, I've found that a denoise range of 0.4 to 0.5 works best. This preserves structure while still enhancing detail without overbaking the image.
⚠️ ADetailer Note
Be cautious when using ADetailer with YOLO-based models (like bbox/full_eyes_detect_yolo8s.pt or face_yolov8m.pt). If the denoise is set higher than 0.4, ADetailer can over-exaggerate facial features, especially eyes and mouths. Keeping denoise lower helps avoid artifacts or surreal distortions. You can also use Throwing Pasta (MIX) - Spaghetti as the ADetialer's inpainting model since it works well with the style.
📋 ADetailer Settings
Here are the current ADetailer/YOLO settings shown:
steps: 10
cfg: 5.0
sampler_name: dpmpp_2m
scheduler: karras
denoise: 0.45
feather: 10
noise_mask: enabled
force_inpaint: disabled
bbox_threshold: 0.65
bbox_dilation: 1
bbox_crop_factor: 1.5
sam_detection_hint: mask-area
sam_dilation: 4
sam_threshold: 0.93
sam_bbox_expansion: 0
sam_mask_hint_threshold: 0.70
sam_mask_hint_use_negative: False
drop_size: 5
Maybe I have my settings wrong and if you know a fix let me know. I only pretend to know what I'm doing.