Core trigger words: collarbone, bare shoulders, white gloves, see-through, lace-trimmed dress, white wedding dress, white footwear, high heels, <lora: wedding dress.no.02:0.6> (Note: white footwear, high heels may be added or removed depending on the situation)

Advantages: moderate output rate, capable of producing good images, interchangeable heads, compatibility, low weight of 0.6.

But the downside is: requires a large number of draws to produce good images, prone to various breakdowns.

Parameters: Steps: 20 (note: 30 or more also acceptable), Sampler: DDIM, CFG scale: 7.0, Size: 768x1024, Model: majicMIX realistic safetensors, Lora Weight 2: 0.6-0.7.

(Note: When changing faces, it is advisable to use a recent photo or a seated posture image, preferably with a resolution of 768x1024 or 512x1024. If facial features are not good, facial repair should be enabled. When used alone, weight of 0.7 is recommended. Depending on the situation, whether to use a veil or not, you can choose to add (((headwear))) prompts on the front or back. Parentheses may be added or removed accordingly.

(Different loras have different compatibility and feedback with weights fluctuating around 0.6-0.8.)

Red text emphasizes:

1. Steps: 30, enable high-definition restoration, restoration weight 0.33 or 0.7. This can help control leg breakdowns, increase the output rate of white silk high heels, recommended for solo use. Increase the lora weight to 0.7 or decrease to 0.6, ideally avoiding facial loras or using highly compatible facial lora models. There have been significant improvements seen in the results, which were not observed with other models.

2. When changing faces, it is best to use a seated posture, remove leg prompts to bring the face closer to the frame for clearer and more distinctive features. The effect of facial lora will be more effective and less prone to breakdowns. High probability may not require high repair restoration or facial repair to avoid interference with the facial lora.

Three examples: (These are just examples for reference, you can freely interpret, but not following the basic logic of AI drawing will likely result in failure.)

Example 1: Close-up face change image (leg parts removed to use seated posture, socks and high heels have been removed)

Prompt: 1girl, solo, perfect face, look at viewer, sitting,

<lora:xxxxx:0.8>, (Note: this is the facial lora weight, adjust the value based on feedback),

collarbone, bare shoulders, white gloves, see-through, lace-trimmed dress, white wedding dress, <lora: wedding dress.no.02:0.6>

masterpiece, best quality, highly detailed,

Negative prompt: unclear faces, (((censored))), (Multiple Girls), (bad hands), (bad feet), (bad legs), (bad anatomy), virtual artifacts, low-quality,

Steps: 30

Sampler: DDIM

CFG scale: 7.0

Width: 768

Height: 1024

Model Name: majicMIX_realistic_safetensors

Example 2: Seated image with white silk high heels

Prompt: 1girl, solo, perfect face, look at viewer, sitting,

<lora:xxxxx:0.8>, (Note: this is the facial lora weight, adjust the value accordingly),

collarbone, bare shoulders, white gloves, see-through, lace-trimmed dress, white wedding dress, white footwear, high heels, <lora: wedding dress.no.02:0.6>

masterpiece, best quality, highly detailed,

Negative prompt: unclear faces, (((censored))), (Multiple Girls), (bad hands), (bad feet), (bad legs), (bad anatomy), virtual artifacts, low-quality,

Steps: 30

Sampler: DDIM

CFG scale: 7.0

Width: 768

Height: 1024

Model Name: majicMIX_realistic_safetensors

Distant view of white silk high heels standing posture image

Prompt: 1girl, solo, perfect face, look at viewer, sitting (or standing),

(Note: Since full-body standing images often have smaller faces, the effect of facial lora is almost zero, it is suggested not to add facial lora, enabling facial restoration may prevent facial lora from being disturbed.),

collarbone, bare shoulders, white gloves, see-through, lace-trimmed dress, white wedding dress, white footwear, high heels, <lora: wedding dress.no.02:0.6>

masterpiece, best quality, highly detailed,

Negative prompt: unclear faces, (((censored))), (Multiple Girls), (bad hands), (bad feet), (bad legs), (bad anatomy), virtual artifacts, low-quality,

Steps: 30

Sampler: DDIM

CFG scale: 7.0

Width: 512

Height: 1024

Model Name: majicMIX_realistic_safetensors