@alcaitiff Thanks! So, as I’m not using Comfy I don’t need v3 and can just wait for eventual future updates :)

@stamp There is a miscommunication it seems. This version apparently is a significant improvement especially when it comes to using it with other loras.

@VKBSOD I don’t know about that, but for sure after testing it seems more consistent than v2, so even for Draw Things it’s worth it :)

Am I doing something wrong I can't get anything to work, if I say a woman is giving a man a blow job (with better detail prompt obviously) she is just holding her mouth open and the man is sometimes there sometimes not but never with his penis out

@Yahulleyyy - also try with this: https://civitai.com/models/2218365?modelVersionId=2508871

@biodizain its all good my testing was just with terrible prompts ,once i changed this it worked great

No, all LoRa characters are like this. For example, if you're drawing a celebrity, using LoRa higher than 0.3 will forcibly modify the character's facial features. At level 1, it will almost entirely replace the face with the training material character. (Of course, for purely erotic content, the drawback of not caring who it is is negligible). Therefore, 0.2~0.4 is recommended.

simple solution - don't be a degenerate by making celeb deepfakes,

@QualityControl Not deepfakes. Using a celeb in the prompt is just the easiest way to get a suitable and pretty face.

Oh it works... Just my prompts were beyond awful... Went into much more detail and works like a charm🍻

@Yahulleyyy can you give us prompt for blowjob. I am in V5, still cannot get the woman to oral. It never goes inside the mouth.

@alcaitiff yes, I tried doing this, still not getting the same result, as scheduler you used "FlowMatchEulerDiscreteScheduler", but it's not in my list. Where can I find this?

@alcaitiff much appreciated!

@alcaitiff I'm getting much better results now, thank you!

bro help us

Any luck with generating a scene, then inpainting with the Character lora over?

@LOPxd I've had desired results by chaining the loras together then balance them around 1.2 The mystic and character lora should be the strongest. And you should use 1 or two other loras and put them into the negative so that it doesn't exceed 1.2. For example if your character lora is 0.7 and you mystic is 0.6 you add them up and figure out what room you have to adjust. Keep adjusting the negatives and positives until you get desired results. Also using the right sample and scheduler helps a ton.

@Superelmostar675 thank you a lot!

All LoRA can not pass the weight o 1.00. You can use the native node from Comfyui merge models, add 2 loras to merge model, make it by par and you can add how much LoRAs you want. Simple as that