<h1 id="about-this-model:"><span style="color:rgb(34, 139, 230)">About this Model:</span></h1><p>I collected <strong>180+</strong> high-quality animated rule34 (uncluding 2d, 3d, many of poses, characters and styles), and goodly trained on it. Natively supports lenghts <strong>from 1 to 5 seconds</strong>, <strong>16fps</strong>. That means you can freely generate videos from 17 to 81 frames. Use 4-8 steps for drafts, and 20-40 for good quality results. I recommend to use few seeds to find the best, it really matters to the main concept of animation. Best resolution is 512x512 - 1024x1024, but you can use other. Supports from square to horizontal and vertical. LoRA strength good on 1.0 - you can don't touch this, but may conflict with other LoRAs, not tested!</p><p></p><h2 id="prompting:"><span style="color:rgb(130, 201, 30)">Prompting:</span></h2><p>Your prompts must be a simple:</p><pre><code>cowgirl, lewd and seductive, heavy breathing, looped sex, static background and camera</code></pre><p>Sometimes model ignores some words, but goodly supports up to 20-30 tags/words of description. All prompts in training written by me, bruh, because it's simple, but also simple to write your and generate :3</p><p>That's not a rule, but you can use "good quality" in prompt and "bad quality" in negative prompt, it will slightly work. Use normal CFG around 3.5-5.</p><p></p><h2 id="trained-only-on-i2v!"><span style="color:rgb(253, 126, 20)">Trained ONLY on I2V!</span></h2><p>Natively not trained on "text2video", but fully supports "image2video" format. Maybe later I will do new version, if you will support this thing, because for this I need to prepare a new dataset, and train again ¯\_(ツ)_/¯</p><p></p><h2 id="bad-for-irl-porn!"><span style="color:rgb(253, 126, 20)">Bad for IRL porn!</span></h2><p>This model mostly trained on animated 2d and 3d scenes, but not specific on real-life porn. Only ~2% of data has been IRL videos, but this LoRA don't make the IRL results worse (I guess).</p><p></p><h2 id="other:"><span style="color:rgb(21, 170, 191)">Other:</span></h2><p>I will read all your feedback and watch public posts, hope that's not creepy.. I really love what I made, because it's fast (5B) and best for this size with quality dataset.</p><p>For training 14B (honestly 28B) model I need a much more time and money, so I not motivated now, but it possible in future. You can support me with "follow" button, like the model, etc. what you can! Maybe I will place here some links later..</p>

Files with SHA256: ce4634b6fa06c15197c0c116d4c20ca8c406e97f96bf52550f8905b7525cb1ce