Dream DeepThroat — Experimental LTX2.3 LoRA
This LoRA is provided as-is, strictly for experimental purposes.
It is a motion LoRA for image-to-video generation. It guides the base model toward a specific type of motion/action, shaping how the subject moves rather than altering identity or appearance. The output is always driven by your input image, your prompt, and the capabilities already present in the underlying base model.
No identity-targeted datasets were used in training, and the LoRA does not reconstruct, swap, or expose any specific person's likeness. It only influences motion and action dynamics.
I take no responsibility for your generations. Anything you create with this LoRA is entirely your own responsibility. By downloading or using it, you agree to comply with all applicable laws and platform rules.
Hey everyone! 👋 This is my second attempt at training my own LoRA, and today I'm sharing my experimental release of DeepThroat LoRA.
This LoRA is rank 64, trained on 40 videos over 2000 steps. It works exclusively in I2V mode and may produce weird results in T2V, so keep that in mind. For input, I feed in characters generated with Gemini or Qwen Image Edit.
It's trained on a side pose, so it can give unpredictable results when you add extra stuff to the prompt. I'm planning to train future versions on different angles and situations — so stay tuned for updates and new versions!
For inference I run the LoRA at weight 1.0. I also stack it with Penile Praxis at weight 0.85 for more realistic anatomy.
I also built my own training program, Just LTX Trainer, so you can train a LoRA without touching a terminal — you can read about it here.
Thanks for checking it out, and any support or feedback means a lot! Either way, I'll keep experimenting and sharing my results for free. ❤️
Basic prompt:
```
DR3AM_D3EPTHR0AT The girl teleports into the deep throat position and sucks the guy's cock deep while slurping.
```
This prompt was user in all my inference generations and as caption in training
Description
Current limitations of this Experimental LoRA:
Only I2V (T2V can produce bad results)
Only side view (POV and so on doesn't support)
Trained on simple dataset
FAQ
Comments (8)
Did you also train on males?
If not can you do that in next update? :)
I have just verified that this version does not function with male characters. It also fails to operate with anime content unless additional lore is utilized.
I will take this into account prior to training the next version.
Thank you!
I do not understand the thumbs downs. Why shouldn't gay men and straight women who like gay scenarios be catered to as well?
@playnproto266 The people giving it thumbs down believe gay men are too precious to be sexualized. Their love for men is pure and genuine.
@playnproto266 Nothing against gays but if this was my lora and someone asked me to train on males my response would be "Respectfully, no". Why? Because most people creating loras are doing so out of their own interest, not to "cater" to other people, let alone niche content that interests ~2% of the populous. I would not want to download, view or caption content that I have no interest in that would make me grimace to watch. It comes off entitled, like why can't people who like furry content or morbidly obese people be catered to as well? Well, they can -- they should just go train their own loras. Problem solved.
@NANONYL121 I mean, it's a reasonable question. You can just say as you did, "respectfully, no" and that can be that. No need to passively-aggressively thumbs it down especially since this ISN'T you lora and the creator is perfectly happy to acquiesce to a reasonable request.
@Zewsic Great!
I was spent to the bone but this bad boy brought me back.