Love v1.3, the face is more consistent now on edit mode compared the v1.1

I love this lora, been using to since the start. Now some technical question: why is anal sex so difficult to "model"? I'm not a technical person myself, I'm very curious. To repeat my self, this lora is best of the best I've ever used!

v1.3 is great! Waiting for a checkpoint, it works usually even better for some reason.

Can I use this on edit?

Nice. Can you merge 1.3 in the checkpoint? for some reason if i use the lora over the base model, images are very degraded

Does anyone know how to make this work on draw things? am i missing a trigger word or something?

1.2 had problems with penetrations and anus, the model don't know the difference between "anal cowgirl" or "vaginal cowgirl"...and vaginas are too big and there a big gap between vaginal hole and clitoris, making the anatomy weird...

I will try 1.3 today and return the feedback.

Works really well! Getting closer to what I can get out of Qwen. Please consider adding some innie vagina to the dataset and a trigger for that. I can hardly generate such a pussy shape without a pass of the generated results through some IL realistic model.

plz do Z-image next

Nice one! Could you by any chance add the following perspectives in nsfw scenes to your training data, they are lackluster and produce a lot of failed images: "🎥Over the Shoulder Shot": "shown from the partner's perspective, camera placed just behind and slightly above the out-of-focus head and shoulder, while focusing on the scene ahead",

"🎥️⬆️Over the Head Shot": "shown from the partner's perspective, high angle camera placed just above the out-of-focus head, while focusing on the scene ahead",

"🎥⬇️Over the Body Shot": "shown from the partner's perspective, low angle camera placed just to the {right|left} of the out-of-focus {torso and hand|waist and leg}, while looking up focusing on the scene ahead",

"🧱Look Through Shot":"camera is positioned far from the subject looking through a gap in the obstacles that greatly obstruct the view, the subject is only partially visible",

"↔Side Heavy Composition": "the subject is on the {left|right} in the frame, opening the center for the view on the scene",

"🎞Multiview Collage": "several views arranged together in the same image, showing the subject from different angles simultaneously, including front, face closeup, back, and side, creating dynamic visual storytelling",

Also, in Telephoto the background is not compressed at all and basically, it's just a full body shot: "🔭Telephoto":"extreme compressed perspective with strong subject isolation and minimum depth, flattened background appears visually closer and softly blurred into bokeh".

1.3 is a solid improvement, biggest jump yet imo. It no longer changes faces to give everyone a "porn" look in edit mode. Works very well with the Miraclein NSFW checkpoint as well, highly recommend folks try that combo!

hello, licking testicles don't work on qwen or klein :(

I'm using Euler 4 steps, the identity is around 50%, is there any recommendations to get the face consistent with reference

The face is sometimes matching, sometimes not. Tried changing the lora strength, but lowering it gives weird artifacts.

Using the Qwen version, no matter what I do, I can't get an anus to show up. Vagina? No issues. Dicks? Sometimes when I don't even want them! But ask for an anus to be visible? Nope. Smooooth skin.

Also works with flux 9b-kv quite nicely.

Can you please do a Lora version, i use drawthings cloud and can’t upload the full merged checkpoint to their cloud, i can only use their cloud models, but i can upload Loras and use those. I’m guessing they dont support lokrs bc this Lora doesn’t do anything. Please add non lokrs version

is flux klein 9b even in bf16 meant to produce a crazy amount of body horror? seems like nudity and flux/even z-image turbo seems to not compute, most gens i try to pull off in it are worse than anything you could make in sdxl. even copying other people's node setups verbatim seems to only give me maybe 3 good gens in every 10 which is insane.

Anyone using the Qwen version 1.3 able to generate condoms in edit mode? For me the condoms it generates look like crayons. I am using qwen edit 2509

Could you add deepthroating for future updates or a zturbo if possible? This is crazy good

i know it's asking a lot, but is an Ernie version possible?

Absolutely amazing! It can do so much stuff in high quality and all kind of styles and camera angles! It's super flexible. Feels more like a whole basemodel than just a Lora! I hope you will continue your amazing work =)

what comfyui workflow should i use?

Keep the good work, it's really good! Thanks!

Excellent lora but I can no longer make skin wet.

Huge ask but could you consider creating a lora version of this dataset? drawthings doesn't yet support Lokr, hopefully that will change soon. I'd use comfyui but my ram goes mental. Thanks so much.

Error getting file.

I can't download this even with vpn, does anyone know how to solve this?

mind sharing some prompts?