Hey everyone!
Excited to share something I've been working on: Wan Gone Wild 1.3B T2V, specifically fine-tuned for NSFW content.
Overview - What is this thing?
Wan Gone Wild 1.3B T2V is designed to be a powerful starting point for generating adult-themed short video clips from text prompts. It's been trained on a massive dataset: the top 1,000 posts from around 1,250 different NSFW subreddits. The goal was to create a model that has a broad understanding of NSFW scenarios, aesthetics, and actions.
You can find the full, more technical model card over on Hugging Face: https://huggingface.co/NSFW-API/NSFW_Wan_1.3b
Why I'm releasing this now:
Think of this model as a "pre-trained" NSFW model. While you can generate videos with it directly, its real magic shines when you use it as a base for training your own LoRAs.
Why? Because it already has a foundational understanding of:
Core NSFW Anatomy: Yep, it knows what a penis, vagina, breasts, etc., are supposed to look like.
Common Sexual Acts: Concepts like blowjobs, masturbation, various positions, and general interactions are already part of its knowledge.
General NSFW Aesthetics: It gets common lighting, settings, and visual styles often found in adult content.
This means you don't need to teach your LoRA these fundamentals from scratch! You can focus your LoRA training dataset exclusively on that specific niche concept, character, art style, unique action, or specific motion you're aiming for. NSFW Wan will help "fill in the rest."
In my own tests, I've seen really promising LoRA results for various concepts within just 10-20 epochs when built on top of this base model. It significantly cuts down the effort needed to get to high-quality, specific results.
Important Note & How to Use:
This current version was fine-tuned for 10 epochs on images. Because of this, you might notice some deterioration or lack of complex motion in the generated videos if used standalone. This is expected!
Apply a LoRA: This is highly recommended to get the desired motion, style, and overall video quality.
Consult prompting-guide.json: This file (included in the repo) is super useful! It breaks down common keywords and phrases from the training data, which can help you craft more effective prompts, especially for specific niches.
This is a Beta - Let's Explore Together!
I'm releasing this as a sort of "beta" version. I wanted to get it out to the community to see what you all can create with it and to gather feedback. I'm still committed to improving it, with the next major step being to fine-tune it on actual video content to enhance motion and temporal consistency. Your explorations and feedback will be invaluable!
Community & Support:
Come hang out on our Discord! Share your creations, ask questions, discuss prompting, and help us make this even better: https://discord.gg/mjnStFuCYh
Ethical Considerations:
This model is for adult users (18+/21+) and is intended for creating fictional, synthetic media. Please use it responsibly, legally, and ethically. Don't use it to create non-consensual content or to harm/impersonate real individuals.
License:
Steal this model! (Basically, do what you want with it, but be cool and responsible.)
Can't wait to see what you all cook up with it! Let me know your thoughts!