You're looking at a specific version of this model. Jump to the model overview.
nightmareai /majesty-diffusion:ff97d797
Input schema
The fields you can use to run this model with an API. If you don’t give a value for a field its default value will be used.
| Field | Type | Default value | Description |
|---|---|---|---|
| clip_prompt |
string
|
The portrait of a Majestic Princess, trending on artstation
|
Prompt for CLIP guidance
|
| latent_prompt |
string
|
The portrait of a Majestic Princess, trending on artstation
|
Prompt for latent diffusion
|
| height |
integer
(enum)
|
256
Options: 128, 192, 256, 320, 384 |
Output height (output will be scaled up 1.5x with default settings)
|
| width |
integer
(enum)
|
256
Options: 128, 192, 256, 320, 384 |
Output width (output will be scaled up 1.5x with default settings)
|
| init_image |
string
|
Initial image
|
|
| init_mask |
string
|
A mask same width and height as the original image with the color black indicating where to inpaint
|
|
| init_scale |
integer
|
1000
|
Controls how much the init image should influence the final result. Experiment with values around 1000
|
| init_brightness |
number
|
0
|
Init image brightness
|
| latent_scale |
integer
|
12
|
The `latent_diffusion_guidance_scale` will determine how much the `latent_prompts` affect the image. Lower help with text interpretation, higher help with composition. Try values between 0-15. If you see too much text, lower it
|
| clip_perceptors |
string
|
### Preloaded with image
[clip - mlfoundations - ViT-B-16--openai]
[clip - mlfoundations - ViT-L-14--openai]
[clip - mlfoundations - ViT-B-32--laion2b_e16]
### Download on demand
#[clip - mlfoundations - ViT-B-32--openai],
#[clip - mlfoundations - RN50x4--openai]
#[clip - mlfoundations - RN50x64--openai]
#[clip - mlfoundations - RN50x16--openai]
#[clip - mlfoundations - ViT-L-14--laion400m_e32]
#[clip - mlfoundations - ViT-L-14-336--openai]
#[clip - mlfoundations - ViT-B-16--laion400m_e32]
#[clip - mlfoundations - ViT-B-16-plus-240--laion400m_e32]
#[clip - sajjjadayobi - clipfa]
#[clip - navervision - kelip_ViT-B/32]
#[cloob - crowsonkb - cloob_laion_400m_vit_b_16_32_epochs]
|
List of CLIP perceptor models to load, one per line. More models will consume more memory. Uses https://github.com/dmarx/Multi-Modal-Comparators.
|
| clip_scale |
integer
|
16000
|
CLIP guidance scale
|
| aesthetic_loss_scale |
integer
|
400
|
Aesthetic loss scale
|
| starting_timestep |
number
|
0.9
|
Starting timestep
|
| model |
string
(enum)
|
finetuned
Options: original, finetuned, ongo, erlich |
Latent diffusion model (ongo and erlich may need to download, taking extra time)
|
| custom_schedule |
string
|
[
[50, 1000, 8],
"gfpgan:1.5",
"scale:.9",
"noise:.55",
[5,300,4]
]
|
Custom schedules, one per line. See the Majestic Guide for documentation.
|
| latent_negatives |
string
|
|
Negative prompts for Latent Diffusion
|
| output_steps |
integer
(enum)
|
10
Options: 0, 5, 10, 20 |
Steps between outputs, 0 to disable progressive output. Minor speed impact.
|
| advanced_settings |
string
|
[advanced_settings]
#Add CLIP Guidance and all the flavors or just run normal Latent Diffusion
use_cond_fn = True
#Cut settings
clamp_index = [2.4, 2.1]
cut_overview = [8]*500 + [4]*500
cut_innercut = [0]*500 + [4]*500
cut_ic_pow = 0.2
cut_icgray_p = [0.1]*300 + [0]*1000
cutn_batches = 1
cut_blur_n = [0] * 300 + [0] * 1000
cut_blur_kernel = 3
range_index = [0]*200+[5e4]*400+[0]*1000
active_function = 'softsign'
ths_method = 'softsign'
tv_scales = [600] * 1 + [50] * 1 + [0] * 2
#Apply symmetric loss (force simmetry to your results)
symmetric_loss_scale = 0
#Latent Diffusion Advanced Settings
#Use when latent upscale to correct satuation problem
scale_div = 1
#Magnify grad before clamping by how many times
opt_mag_mul = 20
opt_plms = False
opt_ddim_eta = 1.3
opt_eta_end = 1.1
opt_temperature = 0.98
#Grad advanced settings
grad_center = False
#Lower value result in more coherent and detailed result, higher value makes it focus on more dominent concept
grad_scale=0.25
score_modifier = True
threshold_percentile = 0.85
threshold = 1
var_index = [2]*300+[0]*700
var_range = 0.5
mean_index = [0]*400+[0]*600
mean_range = 0.75
#Init image advanced settings
init_rotate=False
mask_rotate=False
init_magnitude = 0.15
#More settings
RGB_min = -0.95
RGB_max = 0.95
#How to pad the image with cut_overview
padargs = {'mode': 'constant', 'value': -1}
flip_aug=False
#Experimental aesthetic embeddings, work only with OpenAI ViT-B/32 and ViT-L/14
experimental_aesthetic_embeddings = True
#How much you want this to influence your result
experimental_aesthetic_embeddings_weight = 0.3
#9 are good aesthetic embeddings, 0 are bad ones
experimental_aesthetic_embeddings_score = 8
# For fun dont change except if you really know what your are doing
grad_blur = False
compress_steps = 200
compress_factor = 0.1
punish_steps = 200
punish_factor = 0.5
|
Advanced settings (can override values above)
|
Output schema
The shape of the response you’ll get when you run this model with an API.
Schema
{'items': {'format': 'uri', 'type': 'string'},
'title': 'Output',
'type': 'array',
'x-cog-array-type': 'iterator'}