Shameless plug for my deviant art page https://www.deviantart.com/le-razzz
Follow me there for great Ai video content.
V4 is now out. New concepts
no bra pokies
anal fisting
vaginal fisting
big hanging boobs
bondage
spread pussy
futanari
anal prolapse
I recommend .7 for the Lora and .3 for unconditional. Euler look more realistic now so you can get faster realistic generation.
V3 add a bunch of sexual position concepts
Here's the full list of all the concepts so far:nude woman
nude man
pov vaginal sex
pov doggy style vaginal sex
reverse cowgirl vaginal sex
sideview doggy style vaginal sex
doggy style anal sex
pov anal sex
reverse cowgirl anal sex
pov blowjob
blowjob
sloppy blowjob
handjob
titjob
vaginal dildo
lesbian pussy licking
huge natural breasts
face cum
hairy pussy
pussy insertion (name your object)
puffy nipples
big fake boobs
upskirt
ass spread
creampie
gaping pussy
gaping anus
self fisting
Note: There might be a bit of pussy and anus confusion sometime. In that case put a bounding box where the other hole should be.
V2 add men and a bunch of new concepts.
Don't go higher than .90 or it will start creating bad quality image.
Keep the steps at 18 instead of 12 as you may get generations error from time to time.
Sweet spot is between 0.50-0.90.
res_2s = best definition
res_2m = most realistic and faster than 2s
euler = fastest but less realistic
https://huggingface.co/RazzzHF/Realism_Engine_Ideogram_4/resolve/main/rei4_v2_workflow.json
V1 - First RE Ideogram 4 release. Trained on female DS only so far. Male is coming in the next versions.
Sweet spot is between 0.50-0.60. Higher will degrade the quality.
MAKE SURE YOU USE JSON PROMPTS.
res_2s res_2m is recommended.
Description
Added man and other.
FAQ
Comments (44)
First off, massive props. You are doing God's work! Second, is there a list of concepts you could share, with some prompt specificity (beyond what's in the sample images)? Would massively help to know
Hats off mate
really good, just wish it wouldn't change other elements of the image
I use it at 0.15 - 0.3 strength depending on the image, then it simply provides lighting and style adjustments instead of changing the actual elements
res_mutistep+beta57
res_multistep+linear_quadratic
this two step sampling is good quality at my comfyui.
Can you share this workflow?
Thanks for sharing your best settings.
@HcupMania , sorry to ask but is it faster than res_2s/res_2m? thx in advance!
@smu157 Include in this page https://civitai.red/models/2352925/ziny?modelVersionId=3025876 or workflow only link pixeldrain.com/u/h2qsip4X this. :)
@HcupMania Thanks!
really great quality with V2. One remark: it is not possible to change the eyes direction anymore. anytime i try to add "(s)he is looking at her hands" or what ever, character is always looking at camera. Without lora, it works for that.
What strength are you using the lora at?
Really great stuff. Happy to toss a couple bucks your way to keep training these.
I am instantly sold on big fake boobs👍
Wow! Ideogram 4 has definitely become the best text-to-image template since Lora 2.0! I'm getting incredible and realistic results without any restrictions for any type of NSFW scene. I recommend everyone use the (official Kijai workflow for Ideogram 4) where you write the natural-form prompt and it automatically converts it to the .json version. It also uses bounding boxes where you write and position whatever you want in the scene inside the boxes; it's like a scene outline. By adding just three boxes indicating, for example, three elements of the scene (a naked woman kneeling, a naked man sitting with his erect penis exposed, a pool in the background), the censorship filter is completely disabled. With this workflow, I've NEVER had any problems with censorship filters; I can create any type of NSFW scene.
Thank you and congrats! We look forward to more amazing updates like this!
I appreciate your comment! Can you link the Kijai workflow? Not finding it for some reason.
res_2m comes out as like pop art collage style? res_2s is incredible though. Gen times are longer than id like so ill stick with euler for now
O 'sampler res_2' brings an insane level of realism to the skin, lighting, and even anatomy, but it really takes much longer to generate, but I think it's totally worth it.
Did you put it at 18 steps like I recommended ?
@razzz Wow. Res_2s really does pop the realism and lighting and stuff at 18 steps. I normally use 28 steps with a sampler like Euler, but 18 seems to work fine with res_2s.
@razzz i did yes and it helped, but gen times are too long for my liking haha. its good to know it works though and will do that for my higher quality renders
@thumperunit441 res_2m 18 steps and main lora between .6 to .9 and unconditional Lora at .4 is a good middle ground between time and quality I find.
Res_2s is far more detailed but triples the generation time around 1 minute 50+ generation time. I can get around 50-60 seconds with Euler but the quality is a bit lower.
I also feel like V2 is a bit overcooked, getting a lot more body horrors than V1. But it may be user error as Im still learning this lora and model. I'm also not using bounding boxes to place the limbs positions, thats the real power of this model so its probably why I'm getting more body horrors; im lazy.
Did you increase the step count to 18 like recommended ? This greatly reduce those.
@razzz yes im using the "default" quality setting at 20. I also found through experimenting that if you have the top lora and the uncond lora at an even weight like 0.6/0.6 it starts to break the model and limbs starting going in odd directions. I'm currently using a weight of top lora 0.8 and uncond 0.5 and its working fine now. It seems like the top lora should always be higher weight than the uncond lora or things break. Still testing though so nothing solid.
So V2 is very different from V1.
@fox23vang226 Yeah and same with V3. As I'm still experimenting with the settings we'll see some differences with each iterations.
How do you use this with the text prompt helper? I get this message when I try it. There is not an input for brightness etc. So I don't know how this works.
[INFO] got prompt
[ERROR] Failed to validate prompt for output 179:
[ERROR] * Ideogram4PromptBuilderKJ 259:
[ERROR] - Failed to convert an input value to a INT value: bg_brightness, , invalid literal for int() with base 10: ''
[ERROR] Output will be ignored
[WARNING] invalid prompt: {'type': 'prompt_outputs_failed_validation', 'message': 'Prompt outputs failed validation', 'details': '', 'extra_info': {}}
[INFO] got prompt
[ERROR] Failed to validate prompt for output 179:
[ERROR] * Ideogram4PromptBuilderKJ 259:
[ERROR] - Failed to convert an input value to a INT value: bg_brightness, , invalid literal for int() with base 10: ''
[ERROR] Output will be ignored
[WARNING] invalid prompt: {'type': 'prompt_outputs_failed_validation', 'message': 'Prompt outputs failed validation', 'details': '', 'extra_info': {}}
I'm currently training an ideogram nude Lora too, it should be done and uploaded here tomorrow if everything goes to plan but the samples so far are not looking too hot. Will likely end up with a trash Lora. I used GROK to caption the data and grok is absolutely retarded. Had to spend all day yesterday manually unfucking the mess grok made in the captions.
I went with Gemini 3.1 pro through API. It's uncensored and pretty accurate.
@razzz I screwed up the Lora, only had one type of face in the training data and she got trained into the lora so every generation of a woman is her. This dataset was only 40 pictures and only neck down shots to avoid faces, and ONLY ONE image had a/her face in it, and this was at just 1500 steps and Ideogram already converged that quickly on her face, crazy!. This model responds so fast to data.
Can you please train a lora with the male data only. The v2 lora when doing solo male gens gives men vaginas on their testicles.
@razzz that one is soooo bad
@theally when is Ideogram 4 going to get its own base model filter for searching?
Unfortunately @theally no longer works at civitai since May 1st: https://civitai.red/articles/28893/farewell-civitans
@NowhereManGo Good. He was one of the clowns that would break the rules but it was ok because he was admin. One of the losers that kept banning playtime_ai, the single greatest loss this site has had.
@NowhereManGo oh, I don't know how I missed that. Thanks. Thats sad news.
@J1B You are welcome.
@tajona5803464 I don't know much about theally, so no comment😅
this is great, i do notice that the eyes are sort of messed up; the symmetry is off when lora is engaged
How did you train this, i really want to train my lora on this
superb, ty.
but i notice that ideogram4 is kinda undertrained when it comes to basic poses and ideas, which sometomes ends up with body horrors. will you be maybe teaching it more poses and nsfw concepts as well?
You can pose using JSON and bounding boxes. check out these image and download the PNGs for the workflow with bounding boxes: Image post by Rocket_AI | Civitai
Holy shit the sample images look amazing, congrats bro, this is generational.