The non-stop lewd pictures generator.
How it works
random character + random clothes + random expression + random pose + random location + random style = image
This is a big prompt combinator. Every single word, including wildcards, is tested to work with PonyDiffusion Turbo.
How to use
Go through installation (guide below);
Check "Extra Options" and "Auto Queue" checkboxes in ComfyUI menu;
Press "Queue prompt".
Installation
Install Impact Pack custom nodes;
Install Custom Scripts custom nodes;
Download PonyDiffusion Turbo model and put it into your model folder (probably "\ComfyUI\models\checkpoints");
Move the "ponydiffusion" folder from the archive into your wildcard folder (probably "\ComfyUI\custom_nodes\ComfyUI-Impact-Pack\wildcards");
Run it!
How to modify
The workflow is mainly the collection of prompt parts. It is made to be easily customized. The syntax is fully explained here.
For example, adding this line in the middle of "Character" node (Bondage Edition) will make Tinker Bell randomly appear in generations:
(tiny fairy Tinker Bell, tinker_bell_\(disney\), fairy wings)|Note that Bondage Edition uses a small amount of characters with keywords to guide their appearance, while Lesbian Edition uses a huge wildcard "pd_female_character.txt". You can easily modify both workflows, choosing your preference between quality and diversity.
You can also use LoRAs, dynamically! Right LoRA will be selected only when the right part of the prompt is selected. For that, you'll need to:
Replace "ImpactWildcardProcessor" nodes with "ImpactWildcardEncode";
Connect model and CLIP to them;
Put your LoRAs inside according to syntax;
Join outputs together with "ConditioningCombine" nodes (you can use "ConditioningConcat" for Place & Style. It might or might not improve results);
Connect the resulting condition to "KSampler Turbo" node.
Notes
"Save Image" node is disabled by default (Ctrl + M).
It is tricky to recover a prompt from an image. I genuinely recommend using "SD Prompt Saver" node from SD Prompt Reader custom nodes to easily save and load the prompt for further processing. It's quite trivial, but there is a guide in comments anyway.
Description
Lesbian Edition is a surprise to everyone, including myself.
05.06.2024 silent update: fixed tiny typo: "__ponydiFFusion/pd_mystery_style__"
FAQ
Comments (37)
I would seriously love a version with loras
Not planned, too much variety to cover. Only if you do it yourself, for your preference.
@Postpos unfortunately I don't really know how to do complex comfy workflows like this. And for whatever reason yours give me the best quality, whenever I try my own the images all come out distorted. The vae is connected so idk why.
@Taco360 Do you get images full of noise? Than it's "CLIP Set Last Layer" not being set to 2.
Is it scewed siluettes with burned colors? Than it's PonyDiffusionTurbo with non-Turbo scheduler (or non-DPM++ SDE sampler).
Everything else is most likely prompt fault.
@Postpos I do have clip set -2, but I use regular pony or pony fae not turbo, the prompts I use turn out fine on the on site generator, and they turn out fine with your workflow. Obviously it's something with my workflow. I'm using low vram too.
@Postpos do you think you could make a simple pony workflow with clipset and loras? So far I've not seen a comfyui setup that has it. They always ignore clipset.
@Taco360 It's pretty trivial, screenshot is enough: https://civitai.com/posts/3323664
@Postpos thanks. I'll try it out seems exactly what I needed. I'll have to figure out how to add Lora's. Shouldn't be too hard.
@Postpos unfortunately loading it into comfy doesn't work for some reason. there might not be any data in the pic
@Taco360 save into a text file and rename it to "JustPony.json":
{ "last_node_id": 269, "last_link_id": 584, "nodes": [ { "id": 233, "type": "CLIPSetLastLayer", "pos": [ 680, 360 ], "size": { "0": 210, "1": 60 }, "flags": {}, "order": 5, "mode": 0, "inputs": [ { "name": "clip", "type": "CLIP", "link": 460 } ], "outputs": [ { "name": "CLIP", "type": "CLIP", "links": [ 461, 462 ], "shape": 3, "slot_index": 0 } ], "properties": { "Node name for S&R": "CLIPSetLastLayer" }, "widgets_values": [ -2 ] }, { "id": 50, "type": "VAEDecode", "pos": [ 1420, 300 ], "size": { "0": 140, "1": 50 }, "flags": {}, "order": 12, "mode": 0, "inputs": [ { "name": "samples", "type": "LATENT", "link": 574 }, { "name": "vae", "type": "VAE", "link": 577 } ], "outputs": [ { "name": "IMAGE", "type": "IMAGE", "links": [ 570 ], "shape": 3, "slot_index": 0 } ], "properties": { "Node name for S&R": "VAEDecode" } }, { "id": 226, "type": "EmptyLatentImage", "pos": [ 920, 860 ], "size": { "0": 220, "1": 80 }, "flags": {}, "order": 7, "mode": 0, "inputs": [ { "name": "width", "type": "INT", "link": 457, "widget": { "name": "width" } }, { "name": "height", "type": "INT", "link": 458, "widget": { "name": "height" } } ], "outputs": [ { "name": "LATENT", "type": "LATENT", "links": [ 459 ], "shape": 3, "slot_index": 0 } ], "properties": { "Node name for S&R": "EmptyLatentImage" }, "widgets_values": [ 1024, 1024, 1 ] }, { "id": 3, "type": "CheckpointLoaderSimple", "pos": [ 280, 300 ], "size": { "0": 370, "1": 110 }, "flags": {}, "order": 0, "mode": 0, "outputs": [ { "name": "MODEL", "type": "MODEL", "links": [ 470 ], "shape": 3, "slot_index": 0 }, { "name": "CLIP", "type": "CLIP", "links": [ 460 ], "shape": 3, "slot_index": 1 }, { "name": "VAE", "type": "VAE", "links": [ 575 ], "shape": 3, "slot_index": 2 } ], "title": "Base model", "properties": { "Node name for S&R": "CheckpointLoaderSimple" }, "widgets_values": [ "XL\\Pony\\ponyDiffusionV6XL.safetensors" ] }, { "id": 267, "type": "Reroute", "pos": [ 680, 230 ], "size": [ 75, 26 ], "flags": {}, "order": 6, "mode": 0, "inputs": [ { "name": "", "type": "*", "link": 575 } ], "outputs": [ { "name": "VAE", "type": "VAE", "links": [ 576 ], "slot_index": 0 } ], "properties": { "showOutputText": true, "horizontal": false } }, { "id": 268, "type": "Reroute", "pos": [ 1310, 230 ], "size": [ 75, 26 ], "flags": {}, "order": 10, "mode": 0, "inputs": [ { "name": "", "type": "*", "link": 576 } ], "outputs": [ { "name": "VAE", "type": "VAE", "links": [ 577 ], "slot_index": 0 } ], "properties": { "showOutputText": true, "horizontal": false } }, { "id": 21, "type": "CLIPTextEncodeSDXL", "pos": [ 920, 610 ], "size": { "0": 220, "1": 202 }, "flags": {}, "order": 9, "mode": 0, "inputs": [ { "name": "clip", "type": "CLIP", "link": 462 }, { "name": "text_g", "type": "STRING", "link": 477, "widget": { "name": "text_g" } }, { "name": "text_l", "type": "STRING", "link": 22, "widget": { "name": "text_l" } }, { "name": "width", "type": "INT", "link": 578, "widget": { "name": "width" } }, { "name": "target_width", "type": "INT", "link": 579, "widget": { "name": "target_width" } }, { "name": "height", "type": "INT", "link": 580, "widget": { "name": "height" } }, { "name": "target_height", "type": "INT", "link": 581, "widget": { "name": "target_height" } } ], "outputs": [ { "name": "CONDITIONING", "type": "CONDITIONING", "links": [ 133 ], "shape": 3, "slot_index": 0 } ], "title": "Negative", "properties": { "Node name for S&R": "CLIPTextEncodeSDXL" }, "widgets_values": [ 1024, 1024, 0, 0, 1024, 1024, "score_4, score_5, low quality, blurry", "score_4, score_5, low quality, blurry" ] }, { "id": 220, "type": "PrimitiveNode", "pos": [ 570, 940 ], "size": { "0": 210, "1": 80 }, "flags": {}, "order": 1, "mode": 0, "outputs": [ { "name": "INT", "type": "INT", "links": [ 442, 445, 458, 580, 581 ], "slot_index": 0, "widget": { "name": "height" } } ], "title": "Height", "properties": { "Run widget replace on values": false }, "widgets_values": [ 1024, "fixed" ], "color": "#232", "bgcolor": "#353" }, { "id": 7, "type": "CLIPTextEncodeSDXL", "pos": [ 920, 360 ], "size": { "0": 220, "1": 202 }, "flags": {}, "order": 8, "mode": 0, "inputs": [ { "name": "clip", "type": "CLIP", "link": 461 }, { "name": "text_g", "type": "STRING", "link": 583, "widget": { "name": "text_g" }, "slot_index": 1 }, { "name": "text_l", "type": "STRING", "link": 584, "widget": { "name": "text_l" } }, { "name": "width", "type": "INT", "link": 441, "widget": { "name": "width" } }, { "name": "target_width", "type": "INT", "link": 443, "widget": { "name": "target_width" } }, { "name": "height", "type": "INT", "link": 442, "widget": { "name": "height" } }, { "name": "target_height", "type": "INT", "link": 445, "widget": { "name": "target_height" } } ], "outputs": [ { "name": "CONDITIONING", "type": "CONDITIONING", "links": [ 134 ], "shape": 3, "slot_index": 0 } ], "title": "Prompt", "properties": { "Node name for S&R": "CLIPTextEncodeSDXL" }, "widgets_values": [ 1024, 1024, 0, 0, 1024, 1024, "score_9, score_8_up, score_7_up, 1girl, portrait, a woman in a garden, blonde, blue eyes", "score_9, score_8_up, score_7_up, 1girl, portrait, a woman in a garden, blonde, blue eyes" ] }, { "id": 13, "type": "PrimitiveNode", "pos": [ 570, 650 ], "size": { "0": 210, "1": 110 }, "flags": {}, "order": 2, "mode": 0, "outputs": [ { "name": "STRING", "type": "STRING", "links": [ 22, 477 ], "slot_index": 0, "widget": { "name": "text_l" } } ], "title": "Negative", "properties": { "Run widget replace on values": false }, "widgets_values": [ "score_4, score_5, low quality, blurry" ], "color": "#223", "bgcolor": "#335" }, { "id": 219, "type": "PrimitiveNode", "pos": [ 570, 810 ], "size": { "0": 210, "1": 80 }, "flags": {}, "order": 3, "mode": 0, "outputs": [ { "name": "INT", "type": "INT", "links": [ 441, 443, 457, 578, 579 ], "slot_index": 0, "widget": { "name": "width" } } ], "title": "Width", "properties": { "Run widget replace on values": false }, "widgets_values": [ 1024, "fixed" ], "color": "#232", "bgcolor": "#353" }, { "id": 53, "type": "SaveImage", "pos": [ 1590, 300 ], "size": { "0": 480, "1": 530 }, "flags": {}, "order": 13, "mode": 0, "inputs": [ { "name": "images", "type": "IMAGE", "link": 570 } ], "title": "Save result", "properties": {}, "widgets_values": [ "PonyDiffusion/PonyDiffusion" ] }, { "id": 72, "type": "KSampler", "pos": [ 1180, 300 ], "size": { "0": 210, "1": 262 }, "flags": {}, "order": 11, "mode": 0, "inputs": [ { "name": "model", "type": "MODEL", "link": 470 }, { "name": "positive", "type": "CONDITIONING", "link": 134 }, { "name": "negative", "type": "CONDITIONING", "link": 133 }, { "name": "latent_image", "type": "LATENT", "link": 459 } ], "outputs": [ { "name": "LATENT", "type": "LATENT", "links": [ 574 ], "shape": 3, "slot_index": 0 } ], "title": "KSampler Slow", "properties": { "Node name for S&R": "KSampler" }, "widgets_values": [ 376483869389350, "randomize", 25, 8.5, "euler", "karras", 1 ] }, { "id": 269, "type": "PrimitiveNode", "pos": [ 280, 470 ], "size": { "0": 500, "1": 130 }, "flags": {}, "order": 4, "mode": 0, "outputs": [ { "name": "STRING", "type": "STRING", "links": [ 583, 584 ], "widget": { "name": "text_g" }, "slot_index": 0 } ], "title": "Positive", "properties": { "Run widget replace on values": false }, "widgets_values": [ "score_9, score_8_up, score_7_up, 1girl, portrait, a woman in a garden, blonde, blue eyes" ], "color": "#232", "bgcolor": "#353" } ], "links": [ [ 22, 13, 0, 21, 2, "STRING" ], [ 133, 21, 0, 72, 2, "CONDITIONING" ], [ 134, 7, 0, 72, 1, "CONDITIONING" ], [ 441, 219, 0, 7, 3, "INT" ], [ 442, 220, 0, 7, 5, "INT" ], [ 443, 219, 0, 7, 4, "INT" ], [ 445, 220, 0, 7, 6, "INT" ], [ 457, 219, 0, 226, 0, "INT" ], [ 458, 220, 0, 226, 1, "INT" ], [ 459, 226, 0, 72, 3, "LATENT" ], [ 460, 3, 1, 233, 0, "CLIP" ], [ 461, 233, 0, 7, 0, "CLIP" ], [ 462, 233, 0, 21, 0, "CLIP" ], [ 470, 3, 0, 72, 0, "MODEL" ], [ 477, 13, 0, 21, 1, "STRING" ], [ 570, 50, 0, 53, 0, "IMAGE" ], [ 574, 72, 0, 50, 0, "LATENT" ], [ 575, 3, 2, 267, 0, "*" ], [ 576, 267, 0, 268, 0, "*" ], [ 577, 268, 0, 50, 1, "VAE" ], [ 578, 219, 0, 21, 3, "INT" ], [ 579, 219, 0, 21, 4, "INT" ], [ 580, 220, 0, 21, 5, "INT" ], [ 581, 220, 0, 21, 6, "INT" ], [ 583, 269, 0, 7, 1, "STRING" ], [ 584, 269, 0, 7, 2, "STRING" ] ], "groups": [], "config": {}, "extra": { "ds": { "scale": 1, "offset": [ -180, -130 ] } }, "version": 0.4}
@Postpos thanks it works! I've also figured out how to add Loras to the other setups. All I needed to was just click on the checkpoint loader and add the load Lora.
I added a CR Random LoRA_STack node (acutally about 6 of them) to my version. I'd love a node that allowed me to select a lora if a trigger word appeared in the prompt, but for now I've just got tons of random with the occasional burnt image. It's working pretty well.
@badbunny There is actually an explanation on how you can select LoRAs along with keywords. Search for words "You can also use LoRAs, dynamically!" in description.
@badbunny could u share the workflow
@Ihateusernames442 I'll try and add a pic with my variant to this workflow group.
how do we fix an issue where with multiple/auto queue the prompt doesn't seem to change or the picture seems to follow the same prompt unintentionally?
@moretoolsplease
It's probably caused by drag&drop of the image into ComfyUI window. When you open workflow from image, "ImpactWildcardProcessor" nodes (big green ones) set their mode to Fixed, to allow generating the same picture.
Set mode to Populate in green nodes, instead of Fixed. Also, make sure that random seeds are also not fixed.
Good shit, any chance of a version that can include the metadata of the generation (characters, pose, etc) in the saved file?
I genuinely recommend using "SD Prompt Saver" node from SD Prompt Reader custom nodes to easily save and load the prompt for further processing. It's quite trivial, but there is a guide in comments anyway.
How would I get the full negative prompt text out of the workflow? It looks like the node that outputs the negative prompt conditioning takes in the negative prompt text twice - can i just use the base negative prompt as it defaults to?
I tried to drag the string output from the negative prompt to the "negative" string text node to the "negative" input in SD Prompt Saver, and it just refuses to connect.
@yangxiaolong Negative text is default "Primitive" node, it bugs sometimes. Try to connect "Reroute" to negative inputs (and clip_l, clip_g inputs), then "Primitive" to "Reroute".
Alternatively, use any other node with STRING output. Like "Text Multiline" from WAS Node Suite.
I modified the workflow to include a 'Save text file' node for the positive prompt as well as face detailed, eye detailer and an upscale. I've turned off saving the workflow with the image simply because it's now a bit of a noodle soup with my updates.
Any chance of an updated version that's able to load loras in the wildcards?
ImpactWildcardEncode lets you load loras from the prompt entered into it. I'm able to integrate it into my other workflows, but I'm having trouble with yours.
ImpactWildcardEncode has a peculiarity – it can't process inputs dynamically. That means you can't convert text widgets into inputs, it will not work. It is the common source of it not working.
Thanks so much for such great workflow I have a question tho, is it possible to have aDetailer and Ultimate SD upscale integrated with this workflow?
Yes, it's possible. Feel free to do it.
If you are asking if I can integrate it into published workflow, I say I shouldn't. It's not essential, and it will slow down and overcomplicate the workflow.
@Postpos Oh I have no idea where to start, I'd really like these features in the workflow I've tried but fail at, if you could do that it would be amazing it's the only things that missing for me tho overall it's great
I've never used wildcards on Comfy, where can I download the wildcards, do I have to make them myself?
All wildcards are already in the archive (ponydiffusion folder). You can also download them separately from https://civitai.com/models/400472/female-wildcards-for-ponydiffusion. Put the folder into your wildcard folder (probably "\ComfyUI\custom_nodes\ComfyUI-Impact-Pack\wildcards").
@Postpos I didn't find any bondage or bdsm scenes in the wildcard text
@CarolineKimber254 They are in the workflow. "Situation" green node.
Great workflow :-)
I added and SD1.5 upscale and used another Pony-Model... for more realism ;-)
Thanx for sharing
I can´t stop spamming that render button, to see what comes up next. Great stuff, thank.
Check "Extra Options" and "Auto Queue" checkboxes in ComfyUI menu. That allows you to have at least one free hand. 👋