💯Everything-Hen v3 [N-SFW & SFW] Face fix recom.

Recomended version: v3.1, last one. Recommended: Euler a with 30 steps (22 for closeups). It looks pretty balance at everything, even hands. This is a model merged to be used for N-SFW generation and it can be improved in the future. To get the best quality, you need to use x1024 resolution or use the enhance option with a normal resolution and lower steps (for example: Euler a is +20 steps). I recommend you to enhance instead using Hires fix that it is a big waste of credits. Enhancing without prio costs around 600 credits (from x1,5 to x2 with less or more steps). Wanna make different kind of generations? Play with the first two numbers of my negatives until 1.4 weight. Add '3D' or 'realistic' in negatives if you want more anime style. Important: If you use LoRAs, keep negatives at 1.4 instead 2.0.

Tips: Keep the prompts simpler, prompts with many weights can burn the model. Quality tags are not that necessary. My model is slow, but good at higher steps, fix your with images with face fix and enhance it or use higher resolutions instead.

⚠️ Here's my tutorial on how to get the best quality possible: https://youtu.be/lF5csBbVIX4

I recommend face fix for full body or you will get worse results. Enhancing may also help. Favorite custom resolution for PixAI: 1024x560, 1024x640, 960x610, 960x530. Others: 920x530, 960x640. If you use full body try to re-enhance the model, HiRes fix or use more resolution.

Best overall samplers for Everything-Hen v3: Euler a, DDIM, DPM++ SDE Karras

Samplers: I didn't test enough but you can use a free Euler a/DDMI generation with 20 steps and then enhance. If you don't do that, use +30 steps, but enhancing will always be better. I didn't test other DPM++ samplers, but it can also be good.

This is a model that works the best with highres fix in Stable diffusion. If you will enhance, you can use less steps. If you don't, use more for better quality.

Hen v3 is the experimental model mixed up with Everything-Mix v3_2D model as a base and then adding better N-SFW. Maybe this model have lost some N-SFW (I don't think so), but it won a style similar to my Everything-Mix models. Be very careful with quality tags because it can increase realism and give you weird results, this model is sensitive.

Current versions: · Hen v3_release: The "beta" version of Hen v3. · Hen v3_EX: The version I am using for merging. This version has good body but weird eyes that may be fixed in the following versions. · Hen v3: The improved version that has better hands. · Hen v3.1: Composition changes. Better eye/face generation. More flexibility to 3D.

Prompt tips: try to detail what you want in prompt and detail the color of the eyes. Prompts that are usually fine: 1girl, detailed eyes, black iris...

I recommend you to enhance instead using Hires fix that it is a waste of credits. To get a similar result than highres fix in Stable diffusion, you can use the resolution 640x448 and then enhance. This will make your generation worse, but cost will be cheaper.

Note: If you want to tone down the 3D of this model try "anime" in prompt or my 2Dfy LoRA. For the negatives I have used my generic negatives. You can test others if you want, maybe it shows better quality. If you have better success with other negatives, let me know.

Curiosity: Hen v3 was going to be Everything-Mix v3, but I still don't know how I want to make Everything-Mix v3. Maybe I need to mix it with my model DarkAngel-Mix. This is a model I made to be used in Stable Diffusion so I may share it in the future.

Issues: Men N-SFW seems not optimal.

Off-topic: If you like AI music I have an AI band. Please subscribe! https://www.youtube.com/@ForTheSkies