Krea 2 LoRA training is 50% off, and we're running a Krea 2 Training Contest with 1.8 million Buzz in prizes across four LoRA categories.

⚡ Krea 2 Training Contest + 50% Off Training

A new Illustrious model, capable of creating high-quality images in a second (depending on the hardware, obviously). I wanted to make a DMD model that wouldn't lose detail or image quality, focused on good prompt understanding, recognizing complex and dynamic poses and images, good facial expressions, and specializing in action because I particularly love that style. I really hope everyone can try it and share their images.<ul><li>The VAE is baked</li><li>Upscaler: Im using 4x-ClearRealityV1 upscale by 1.5 Steps 3 denois 0.2 Hires CFG 1,</li><li>Width and height: 896*1152</li><li>CFG Scale: 1.2</li><li>Steps: 8</li><li>The sampling method i use is Euler A-Karras or LCM-Exponential</li><li>Neg prompt: worst quality, low quality</li></ul>

I wanted to retrace my steps a little and create a model less focused on a pictorial image and more on clean, action-packed images.This model was built around a simple idea: anime should feel dynamic, cinematic, and larger than life.While many anime checkpoints focus primarily on character accuracy or clean rendering, this model pushes much harder toward dramatic visual storytelling. It excels at extreme perspectives, impossible camera angles, intense action sequences, and highly detailed compositions that make every image feel like a frame taken from a high-budget animated film.What Makes This Model Different?Dynamic Camera Work<ul><li>Extreme low-angle and high-angle shots</li><li>Wide cinematic perspectives</li><li>Aggressive foreshortening</li><li>Action-focused framing that creates a strong sense of motion</li></ul>Impossible Angles<ul><li>Designed to handle viewpoints that would normally break composition</li><li>Better spatial understanding in complex poses</li><li>Stronger depth perception in crowded scenes</li><li>More convincing perspective distortion</li></ul>Action and Energy<ul><li>Fast-paced combat scenes</li><li>Dramatic movement effects</li><li>Dynamic poses with natural flow</li><li>Powerful visual impact even in single-character illustrations</li></ul>High Detail Density<ul><li>Rich environmental details</li><li>Intricate clothing and accessories</li><li>Enhanced texture rendering</li><li>Layered scenes that reward close inspection</li></ul>Cinematic Lighting<ul><li>Dramatic contrast</li><li>Atmospheric effects</li><li>Strong highlights and shadows</li><li>Visual storytelling through light and color</li></ul>The GoalThe objective was never to create another generic anime model. This model was trained and tuned to generate images that feel alive, energetic, and ambitious. Whether you're creating epic battles, dramatic character reveals, fantasy worlds, futuristic cityscapes, or impossible action scenes, the model is designed to embrace complexity instead of avoiding it.Every part of the training process focused on one question:"What if every image looked like the most exciting frame of an anime movie?"The result is a model that thrives on motion, perspective, scale, and visual intensity, delivering illustrations that feel bold, immersive, and unforgettable.<ul><li>The VAE is baked</li><li>Upscaler: Im using R-ESRGAN 4x+ Anime6B upscale by 1.5 Steps 3 denois 0.2 Hires CFG 1,</li><li>Width and height: 896*1152</li><li>CFG Scale: 1.2</li><li>Steps: 8</li><li>The sampling method i use is LCM-Exponential, you can try Euler-A</li><li>Neg prompt: worst quality, low quality</li></ul>

IlluSat-DMD-AnimeV6

For this new version, I wanted to create a brushstroke aesthetic, one that simulates a bit of painting. The images aren't as bright as in other models; they have more subdued tones. What I haven't lost, however, is the aspect I love: impossible angles and frenetic action. As I've said many times, I don't consider this model better than the previous ones; it's very different and brings its own unique qualities. I hope you enjoy it as much as I do.<ul><li>The VAE is baked</li><li>Upscaler: Im using R-ERSGAN 4x+ Anime6B upscale by 1.5 Steps 5 denois 0.2 Hires CFG 1.1</li><li>Width and height: 896*1152</li><li>CFG Scale: 1.1</li><li>Steps: 8</li><li>The sampling method i use is LCM-Exponential or LCM-SGM Uniform or Euler A-Karras </li><li>Neg prompt: worst quality, low quality</li></ul>

IlluSat-DMD-AnimeV5

I wanted to create a version that combined the best features of the two previous models, but with an even more exaggerated mix of action, exaggerated poses, facial expressions, and impossible angles. I hope you like it as much as I do and that you can try it out and share your examples. Thank you.<ul><li>The VAE is baked</li><li>Upscaler: Im using 4x-ClearRealityV1 upscale by 1.5 Steps 3 denois 0.2 Hires CFG 1,</li><li>Width and height: 896*1152</li><li>CFG Scale: 1.2</li><li>Steps: 8</li><li>The sampling method i use is Euler A-Karras or LCM-Exponential or LCM-SGM Uniform</li><li>Neg prompt: worst quality, low quality</li></ul>

IlluSat-DMD-AnimeV4

As I've mentioned in other models, this version isn't necessarily better than the previous one; it's just different. As always, the name of the game is action, action, action—dynamic images with a different aesthetic. I hope you like it and share your example images, as that's the only way to improve.<ul><li>The VAE is baked</li><li>Upscaler: Im using 4x-ClearRealityV1 upscale by 1.5 Steps 3 denois 0.2 Hires CFG 1,</li><li>Width and height: 896*1152</li><li>CFG Scale: 1.2</li><li>Steps: 8</li><li>The sampling method i use is Euler A-Karras or LCM-Exponential</li><li>Neg prompt: worst quality, low quality</li></ul>

IlluSat-DMD-AnimeV3

For this version, I changed practically everything. I wanted to focus on one of my obsessions, which is dynamic images, showing a lot of movement and impossible angles. I also tried to improve the facial expressions to make them more emphasized, and finally, to improve the textures. I left behind something I liked, which was the thick lines, to give it an aesthetic closer to anime.<ul><li>The VAE is baked</li><li>Upscaler: Im using 4x-ClearRealityV1 upscale by 1.5 Steps 3 denois 0.2 Hires CFG 1,</li><li>Width and height: 896*1152</li><li>CFG Scale: 1.2</li><li>Steps: 8</li><li>The sampling method i use is LCM-Exponential mostly but it also works with Euler A-Karras</li><li>Neg prompt: worst quality, low quality</li></ul>I really hope everyone can try it and share their images.