AiARTiST middle-aged and elderly women's portrait photography XL LoRA

AiARTiST Middle-Aged Women's Portrait Photography XL LoRA

File Name: AiARTiST_NVZH_DIM128_pruned.safetensors

Supporting Model: AiARTiST-CADS2.0 XL Commercial Advertisement Assistance System (Online Generation)

https://www.liblib.art/modelinfo/a56ebacdba7d4e30b97bb124bc3fc28f  <<Click to Direct

Model List: https://www.liblib.art/search?keyword=AiARTiST

----------------------------------------------------------------------

Model Usage Statement:

1. You are not allowed to host this model and its derivative versions (such as fusion model versions) on websites/applications intended to earn revenue or donations.

2. You are not allowed to directly sell this model and its derivative versions (such as fusion model versions) unless you have made sufficient manual modifications to this model so that it can be fully determined as your personal work in a legal sense. If you violate this clause, all legal consequences will be borne by you personally. I bear no responsibility.

3. You cannot intentionally use this model to create or share illegal or harmful content for distribution and output. Please abide by public morality and use this model for positive purposes.

----------------------------------------------------------------------

Trigger Word: asian female

Generated Parameters Reference:

Parameters

AiARTiST, 45NVZH, asian female, 45 years old, smile, A woman dressed in luxurious attire, modern time, <lora:AiARTiST_45NVZH_DIM128_pruned:0.8>
Steps: 30, Sampler: DPM++ 3M SDE Exponential, CFG scale: 7, Seed: 3850399008, Size: 1080x1920, Model hash: a9dce34f98, Model: AiARTiST_CADS_justice_alliance_a, VAE hash: 737a0c5124, VAE: sdxl-vae-fp16-fix.safetensors, Denoising strength: 0.4, Clip skip: 2, ADetailer model: face_yolov8n.pt, ADetailer confidence: 0.3, ADetailer dilate erode: 4, ADetailer mask blur: 4, ADetailer denoising strength: 0.4, ADetailer inpaint only masked: True, ADetailer inpaint padding: 32, ADetailer version: 23.11.1, Lora hashes: "AiARTiST_45NVZH_DIM128_pruned: 85b61b23e34d", Mask blur: 4, Inpaint area: Only masked, Masked area padding: 32, Version: f0.0.17v1.8.0rc-latest-269-gef35383b

"Recommend using sample image prompts directly and then adding modifying details as needed."

Recommended LoRA Weight: 0.6-1.2, suggested value 0.9

Recommended Sampling Steps: 25-30   CFG scale: 5-8, suggested value 7

Recommended Sampling Methods: DPM++ 2M Karras, DPM++ 3M SDE Exponential

Recommended Resolution: Any resolution

Final images below 1920x1080 or 1920x960 may not fully reflect the texture accuracy of the base model

Can be done in two steps, document-generated image with long side of 1024-1280, Forge branch can generate high-definition images afterwards

Other front-end processes can be sent to generate images with a redrawing magnification of 0.4-0.5 and an enlargement of 1.2-1.5 times

If image size is used for publishing, it can be sent to "Tiled Diffusion" or "post-processing" for enlargement

Post-processing settings for high definition: 8x_NMKD-Superscale_150000_G can be overlaid

                                         4x-UltraSharp or R-ESRGAN 4x+ Anime6B

----------------------------------------------------------------------

Closing Words

As of 2024, many still do not fully understand XL, thinking online generation is slow and requires tagging for output.

Actually, these are outdated concepts from tutorials. Today, I tested Lib's online generation, and XL and 1.5 have similar output speeds.

Furthermore, thanks to XL's larger parameter volume and dual text layer design, it can actually generate the majority of drawing styles based on the base model's characteristics.

Even with some concepts locked in by LoRA, you can still change the desired visual elements through prompts with automatic translation tools on Lib's online platform.

SDXL is a form between commercial and community models, requiring fewer LoRA combinations, less complex prompts, and negative keywords. All you need is your creativity and aesthetic taste. Even re-painting through image generation can yield excellent results.

Additionally, there is often talk about XL's ControlNet models being ineffective.

As of April 1, 2024, the SDXL ecosystem already has over 20 mature CN models, and the number continues to grow.

Soft and hard edges, Openpose, depth and normal maps, sketches, straight lines, doodles, various face-swapping plugins, and high-definition re-painting are all available.

To address issues like ghosting and image quality degradation in XL's CN controllers, you can optimize by reducing CFG appropriately and increasing the number of steps. If control is not strict, my optimization solution is to first generate a strict image similar to the base model with CN using 1.5, then send it for image generation, switch to larger models like AiARTiST UNIT/CADS in XL for high-definition enlargement and redraw, with a magnification of 0.4-0.6.
One major advantage of Lib's online generation is the ease of switching models without frequently unloading large models from memory, but the downside is that sending XL models to the GPU is a bit slow...

Lib recently launched XL Layer Diffusion transparent background images, Tile Diffusion redraw enlargement, image to animation conversion, and a series of popular tools

Away from noise, reduce disk usage, member acceleration – this is why we always encourage everyone to become members and try online creativity.

----------------------------------------------------------------------

The base model used for training: AiARTiST-CADS2.0 XL Commercial Advertisement Assistance System(Enterprise Customized)

Direct Link: https://www.liblib.art/modelinfo/a56ebacdba7d4e30b97bb124bc3fc28f

Model List: https://www.liblib.art/search?keyword=AiARTiST

————————————————————————————————————————

For test queries, leave a message. Business cooperation visit personal homepage +V Zyuan980

AiARTiST | Metaverse China · Shandong Digital Person/AiGC/Metaverse Direction

Being a tool for artists, contact information:https://afdian.net/a/AiARTiST-lib