Hey Mr_Fei! Thanks for checking out the model.

You are completely right, the showcase renders definitely lean into that stylized, dark sci-fi and anime vibe because that is a style I personally love and tested a lot!

However, under the hood, the model actually has a very strong realistic foundation. If you want to push it away from the sci-fi look and get a pure, documentary-style realistic photo, you just need to guide the prompt a bit.

Try using realistic photography terms in your prompt like "8k documentary photography, natural lighting, highly detailed skin pores, raw texture". At the same time, make sure to add "sci-fi, cyberpunk, neon, glowing, bioluminescent, space" to your negative prompt to strip away the futuristic elements.

Hey AceGCR! Thanks for the feedback, and man, I totally feel your pain regarding the hard drive space. My own storage is absolutely screaming right now too, haha!

To answer your question honestly: the reason I went with a full checkpoint instead of just a LoRA comes down to how deeply the lighting, textures, and that signature "AfterHours" deep-black void needed to be baked in.

When you run multiple heavy LoRAs together, they often fight each other or "burn" the image if you push the weights to get that level of contrast. By blending and refining these elements directly into the base model's core transformer network, the lighting physics (like the rim-lighting and bioluminescent bleed) works flawlessly and naturally out of the box at a low CFG (1.0 - 1.2), without needing any extra prompt steering or complex LoRA stacking.

That being said, since drive space is tight for so many of us, I actually just made a highly optimized INT4 version available for download as well! It cuts the file size almost in half, making it way easier on your storage and VRAM while still keeping all that rich contrast and detail intact.

If you want to stick to your own custom base models, you can also use the "Extract LoRA" tool in Kohya or OneTrainer to pull a neat ~200MB LoRA out of it.

Thanks again for checking it out, and let me know how the INT4 version works for you if you decide to give it a spin! 🙌

Hey @ferrrett33! Thanks for the question.

To give you a completely honest and direct answer: No, there are absolutely no specific NSFW or adult LoRAs added into the blend under the hood. The core adjustments were strictly focused on lighting physics, deep contrasts, and premium textures. [1]

it targets the base model's core transformer and lighting networks, it has not altered or forced any specific body shapes or anatomical modifications from the base model. The underlying anatomy and flexibility remain completely intact, exactly as they are in the original Krea 2 structure.

So whether you are rendering characters or environments, the body shapes will follow your prompt and whatever style/anatomy tokens you choose to use, without the model forcing any unwanted body distortions on its own.

Hope that clears it up, and let me know if you have any other questions! 🙌

@ferrrett33 Spot on, @ferrrett33! You guessed it perfectly. That is exactly why you are getting that washed-together look.

AetherFlow v1.0 is NOT a Turbo or Lightning model! It is a full-weight, high-contrast architecture. Because of how deeply the "AfterHours" black levels and the complex lighting physics are baked into the network, it needs room to breathe in the sampler. Running it at 8 steps just doesn't give the mathematical layers enough time to resolve the textures and clean up the noise, which results in that surreal, bleeding effect you saw.

For the absolute best results and to get that crisp, clean photographic realism, try switching up your workflow to these settings:

Steps: 20 to 25 (minimum)

CFG / Guidance: Keep it quite low, around 1.0 - 1.2 (this is super important to keep the lines sharp and prevent the high contrast from burning)

Sampler: DPM++ 2M SDE Karras (or your favorite high-quality sampler) [1]

Give it a shot with 25 steps and that lower CFG, and you will see the model lock in those clean, unblemished surfaces and deep shadows perfectly! Let me know if that cleans up the renders for you! 🙌 [1]

Thank you so much!