[Stable Diffusion] Prompt Sharing and Learning Thread

F

felldude

Member
Aug 26, 2017
479
1,450
me3 said:
If you got the settings to share, i've got some very fun and problematic sets to try it on.
Great fun to see what i can turn some poor ppl into this time, with any luck it might be themselves :p
Click to expand...
This is a good
You must be registered to see the links


Some things that stood out to me very low learing rate 0.0005
Cosine with restarts and warm up
low memory being on (I have this for XLSD but always had it off for SD 1.5)
Adam8bit, instead of Adam full FP16
20 epochs

All those setting may be meaningless without the weighted captions.

It could stand to be cleaned up but here are some the settings with examples of a caption


{"ss_cache_latents":"True","ss_caption_dropout_every_n_epochs":"0","ss_caption_dropout_rate":"0.0","ss_caption_tag_dropout_rate":"0.0","ss_clip_skip":"2","ss_dataset_dirs":""n_repeats\": 4, \"img_count\": 74}}","ss_datasets":"[{\"is_dreambooth\": true, \"batch_size_per_device\": 2, \"num_train_images\": 296, \"num_reg_images\": 0, \"resolution\": [512, 512], \"enable_bucket\": true, \"min_bucket_reso\": 256, \"max_bucket_reso\": 1024, \"tag_frequency\": {\

\"completely nude\": 23, \"sweatdrop\": 6, \"swept bangs\": 5, \"bangs\": 10, \"fingernails\": 1, \"sweat\": 20, \"large penis\": 6, \"naughty face\": 6, \"foreskin\": 1, \"bed\": 9, \"dark-skinned male\": 1, \"interracial\": 1, \"navel\": 23, \"hand on hip\": 2, \"open mouth\": 25, \"standing\": 10, \"full body\": 10, \"cleft of venus\":


img_count\": 74, \"num_repeats\": 4, \"color_aug\": false, \"flip_aug\": false, \"random_crop\": false, \"shuffle_caption\": true, \"keep_tokens.ss_epoch":"10","ss_face_crop_aug_range":"None","ss_full_fp16":"False","ss_gradient_accumulation_steps":"1","ss_gradient_checkpointing":"False","ss_learning_rate":"0.0005","ss_lowram":"True","ss_lr_scheduler":"cosine_with_restarts","ss_lr_warmup_steps":"148","ss_max_grad_norm":"1.0","ss_max_token_length":"225","ss_max_train_steps":"2960","ss_mixed_precision":"fp16","ss_network_alpha":"8","ss_network_dim":"16","ss_network_module":"networks.lora",","ss_noise_offset":"None","ss_num_batches_per_epoch":"148","ss_num_epochs":"20","ss_num_reg_images":"0","ss_num_train_images":"296","ss_optimizer":"bitsandbytes.optim.adamw.AdamW8bit","ss_output_name":"","ss_prior_loss_weight":"1.0","ss_sd_model_hash":"7fcf3871","ss_sd_model_name":"model.safetensors","
 
Last edited:
  • Like
Reactions: Mr-Fox
me3

me3

Member
Dec 31, 2016
316
708
felldude said:
This is a good
You must be registered to see the links


Some things that stood out to me very low learing rate 0.0005
Cosine with restarts and warm up
low memory being on (I have this for XLSD but always had it off for SD 1.5)
Adam8bit, instead of Adam full FP16
20 epochs

All those setting may be meaningless without the weighted captions.

It could stand to be cleaned up but here are some the settings with examples of a caption


{"ss_cache_latents":"True","ss_caption_dropout_every_n_epochs":"0","ss_caption_dropout_rate":"0.0","ss_caption_tag_dropout_rate":"0.0","ss_clip_skip":"2","ss_dataset_dirs":""n_repeats\": 4, \"img_count\": 74}}","ss_datasets":"[{\"is_dreambooth\": true, \"batch_size_per_device\": 2, \"num_train_images\": 296, \"num_reg_images\": 0, \"resolution\": [512, 512], \"enable_bucket\": true, \"min_bucket_reso\": 256, \"max_bucket_reso\": 1024, \"tag_frequency\": {\

\"completely nude\": 23, \"sweatdrop\": 6, \"swept bangs\": 5, \"bangs\": 10, \"fingernails\": 1, \"sweat\": 20, \"large penis\": 6, \"naughty face\": 6, \"foreskin\": 1, \"bed\": 9, \"dark-skinned male\": 1, \"interracial\": 1, \"navel\": 23, \"hand on hip\": 2, \"open mouth\": 25, \"standing\": 10, \"full body\": 10, \"cleft of venus\":


img_count\": 74, \"num_repeats\": 4, \"color_aug\": false, \"flip_aug\": false, \"random_crop\": false, \"shuffle_caption\": true, \"keep_tokens.ss_epoch":"10","ss_face_crop_aug_range":"None","ss_full_fp16":"False","ss_gradient_accumulation_steps":"1","ss_gradient_checkpointing":"False","ss_learning_rate":"0.0005","ss_lowram":"True","ss_lr_scheduler":"cosine_with_restarts","ss_lr_warmup_steps":"148","ss_max_grad_norm":"1.0","ss_max_token_length":"225","ss_max_train_steps":"2960","ss_mixed_precision":"fp16","ss_network_alpha":"8","ss_network_dim":"16","ss_network_module":"networks.lora",","ss_noise_offset":"None","ss_num_batches_per_epoch":"148","ss_num_epochs":"20","ss_num_reg_images":"0","ss_num_train_images":"296","ss_optimizer":"bitsandbytes.optim.adamw.AdamW8bit","ss_output_name":"","ss_prior_loss_weight":"1.0","ss_sd_model_hash":"7fcf3871","ss_sd_model_name":"model.safetensors","
Click to expand...
Some "issues".
You say that the learning rate is "very low", which is a bit surprising consider it's 4-5x what a lot of loras i have uses, that includes Mr-Fox's kendra lora.
Captions, you mentioned weighted captions, but there's no parameter set for it, also there's no actual caption example included (as stated), there's just a tag frequency list, not that it really matters.
It's dreambooth, which apparently make quite the difference (never had the option to really compare so can't say much more)
The clip skip suggests it's for anime so might be less of an issue there, but the rank is far too low to be able to fit much data in to it.
Having done multiple runs there's clear signs there's too little "room"
 
  • Like
Reactions: Mr-Fox
Sepheyer

Sepheyer

Well-Known Member
Dec 21, 2020
1,523
3,589
I tried SDXL 1.0, and it runs prohibitively slow on my 6gb GTX 1660ti. The image below took 80 minutes to render. Meaning in a day I can render ~18 SDXL images. And this is using Euler sampler. I can't imagine how long Heun would take. I might run one for lolz.

Not that anyone owes me anything, but this is bullshit, I want to see the manager, I want my fucking money bek.

So, yeah, no.

ComfyUI_00002_.png
 
Last edited:
  • Haha
  • Wow
Reactions: DD3DD and Mr-Fox
S

Synalon

Member
Jan 31, 2022
191
617
Sepheyer said:
I tried SDXL 1.0, and it runs prohibitively slow on my 6gb GTX 1660ti. The image below took 80 minutes to render. Meaning in a day I can render ~18 SDXL images. And this is using Euler sampler. I can't imagine how long Heun would take. I might run one for lolz.

Not that anyone owes me anything, but this is bullshit, I want to see the manager, I want my fucking money bek.

So, yeah, no.

View attachment 2822015
Click to expand...
Did it take that long using comfyui?
 
  • Like
Reactions: Sepheyer
T

tacoBlade

New Member
Jan 11, 2023
2
4
I tried
You must be registered to see the links
once and generated something i may never be able to generate again whatsoever lol. image (2).png I have started playing with local sd (billion different models on civitai are sooooooo good, loras, adetailer, inpainting) but i can never recreate these. especially this image (25).png .That blue veil/cloud/magical thing I can never generate again is just so surreal.
 
  • Like
Reactions: FreakyHokage, DD3DD, Mr-Fox and 1 other person
R

rogue_69

Newbie
Nov 9, 2021
78
235
Daz to Stable Diffusion to Flow Frames. Created an image in Stable. Used it to do a Face Transfer in Daz (used Face Transfer Shapes, which is a must have for Face Transfer). This got me a basis for the face shape. Rendered a video with the Face Transfered character (with a neutral looking skin, since I wanted Stable to handle the textures). Used the same prompts for the Mov2Mov video in Stable that I used for the Face Transfer reference image. The combination of the Daz Character face shape, with the prompts seems to get me a really consistent character. I also rendered hair and clothing as a separate canvas, so I could easily just plop those on her at any time for the video.
ezgif.com-video-to-gif.gif
 
  • Like
  • Wow
  • Red Heart
Reactions: DD3DD, felldude, devilkkw and 2 others
DD3DD

DD3DD

Active Member
Apr 23, 2019
775
1,442
Mr-Fox said:
Using the latest beautiful creation by the eminent Seph for some "photo bashing" and "deepfakes" with inpaint and roop. Thanks again to Synalon for introducing me to this awesome extension.


View attachment 2827798

Source:
View attachment 2827779

You don't have permission to view the spoiler content. Log in or register now.
Click to expand...
Whoa! This is crazy impressive :love: & crazy scary :eek: if you think about it, pretty much the end of non-anon/digital face protection social media, cause we all don't have an Uncle Ben event in our lives to realize "With Great Power, comes Great Responsibility" and we getting there very shortly.

Most interesting that the Adult Video stars have the most "not correct" feel (face fit I mean) while the normal actresses have the very "viable" realistic feel. Is it just me or do you agree Mr-Fox? I wonder if it's because our brains have seen AV stars already nude more and causes the difference or is it something else? :unsure:

Just curious how long did it take to make so many variations?
 
  • Red Heart
  • Like
Reactions: Mr-Fox and Sepheyer
Mr-Fox

Mr-Fox

Well-Known Member
Jan 24, 2020
1,401
3,793
DD3DD said:
Whoa! This is crazy impressive :love: & crazy scary :eek: if you think about it, pretty much the end of non-anon/digital face protection social media, cause we all don't have an Uncle Ben event in our lives to realize "With Great Power, comes Great Responsibility" and we getting there very shortly.

Most interesting that the Adult Video stars have the most "not correct" feel (face fit I mean) while the normal actresses have the very "viable" realistic feel. Is it just me or do you agree Mr-Fox? I wonder if it's because our brains have seen AV stars already nude more and causes the difference or is it something else? :unsure:

Just curious how long did it take to make so many variations?
Click to expand...
short answer: a veeery long time.. :LOL: Since it did take a long time I learned a lot and also they are not posted in the order I made them. So it could just be that I did a better job on some than others. I'm glad you liked it.
 
  • Love
  • Like
Reactions: Sepheyer and DD3DD
FreakyHokage

FreakyHokage

Member
Sep 26, 2017
261
356
Mr-Fox said:
Using the latest beautiful creation by the eminent Seph for some "photo bashing" and "deepfakes" with inpaint and roop. Thanks again to Synalon for introducing me to this awesome extension.


View attachment 2827798

Source:
View attachment 2827779

You don't have permission to view the spoiler content. Log in or register now.
Click to expand...
Is this from using a checkpoint or lora? Where can I get it lol? I can't find any decent loras that let me do nswf images of celebs.
 
  • Like
Reactions: Mr-Fox
LuminaJR

LuminaJR

Newbie
Aug 1, 2018
96
41
i made these without any Lora´s Just Img2img a lot of positive and negative prompt along with Textual inversions
 
Mr-Fox

Mr-Fox

Well-Known Member
Jan 24, 2020
1,401
3,793
FreakyHokage said:
Is this from using a checkpoint or lora? Where can I get it lol? I can't find any decent loras that let me do nswf images of celebs.
Click to expand...
It's made with an extension called roop, not a lora.
You must be registered to see the links

Follow the installation instruction, it won't work otherwise. After the installation you need to edit a code inside a py file to turn off the NSFW filter. In Stable-Diffusion\stable-diffusion-webui\extensions\sd-webui-roop\scripts edit cimage.py and replace the code with this:
" import tempfile
def convert_to_sd(img):
return [False, tempfile.NamedTemporaryFile(delete=False, suffix=".png")] "
 
Last edited:
  • Like
Reactions: VanMortis, onyx, FreakyHokage and 1 other person
Mr-Fox

Mr-Fox

Well-Known Member
Jan 24, 2020
1,401
3,793
LuminaJR said:
i made these without any Lora´s Just Img2img a lot of positive and negative prompt along with Textual inversions
Click to expand...
It's not bad but needs more work imo. In the first image the faces on both are a bit wonky. Is she supposed to look pregnant? This happens sometimes with loras or Ti's, an unwanted side effect with swollen belly. I read the prompt, omg that's some beautiful ChatGPT prose..:LOL: I don't think you need to explain the abbreviation "milf" in the prompt, it's common enough and also we are not communicating with some singularity. SD is simply in data and out data.. Even though it hurts the soul, I would simplify that prompt as much as possible. It's not practical to have such a large describing prompt. A lot of redundant words and that has no effect. It's impossible to know what in the prompt affects what in the image. Though I know easy negative etc is popular I think it's better to try to use normal pos and neg prompt instead. You can only adjust the weight of Ti's like easy negative, you have more direct control with normal prompting. I think my main advice is, "more" doesn't equal "better". With SD it's the opposite but should be worded "simpler is better". I'm also guilty of writing large prompts but I think this is some kind of record.:LOL: Also if you insist on using any lora or Ti, don't use so many, it's difficult enough to get one to work properly.
I hope this helped and is not interpreted as something overly negative, it's only meant as constructive.
 
  • Like
Reactions: Sepheyer
me3

me3

Member
Dec 31, 2016
316
708
negative prompts are much less needed than ppl think, you're generally better off starting without them and being specific in your positive prompt.
IE using age to avoid the person being too young or old, or the type and quality of photo.
Negatives mostly come into play to combat "conflicts", like when you're specifying colors that has a habit of coloring everything or the switching place with another color. In those cases you'd use negatives to "disallow" the color on the specific thing.
Alot of these "easy this and that" flood prompts with pointless instructions which most likely has negative impacts on your image. One very basic example would be if you're trying to create something that's dirty, in ruins, "dystopian" etc, and your "easy to use prompting" contains negatives like ugly and/or old.

Start with a simple positive prompt, find the basics you want and work on finding a specific seed you like (optional in many cases), then tweak/expand upon your positive prompt and if you encounter persisting "problems" you can start trying to add some very specific negative prompts.

With the, worryingly high, amount of images i've generated so far i'm seeing very clear signs that negative prompts are byfar less needed than "guides" and "all knowing super important creators" make it out to be.
As Mr-Fox mentioned, stick to the good old
You must be registered to see the links
 
  • Like
Reactions: VanMortis and Mr-Fox
You must log in or register to reply here.
Top Bottom