Recommended to enhance the image, HiRes fix or use higher resolutions like 1024x.
Hen versions will be more experimental than my Everything-Mix normal versions. 2D to 2.25D model. This model is less flexible with dimensions but it won a better 2D face and N-SFW body. This is the model that merges all the good experimental models with Mix v2 or v2.1_EX; like v3_blurness and Everything-Hen v0.2. This model was going to be Everything-Mix v3 but I thought of improving even more than version. You can notice a similar texture than my Everything-Mix models and maybe you can like this one better than Everything-Hen V2.
I recommend face fix for full body or you will get worse results. Enhancing may also help.
⚠️ Here's my tutorial on how to get the best quality possible: https://youtu.be/lF5csBbVIX4
Best overall samplers for Everything-Hen v1: DPM++ SDE, DDIM, Euler a
Hen v1.0 won some background blurness but at the same time it has some sharpness. Eyes/hands should be better now. The N-SFW gen has been improved and hands are the same than in Everything-Mix V2. Recommended use of LoRA for hands.
· v1_release is the initial version that may be more experimental. I'll try to fix the issues I see with this version. Face has been tuned to be more 2D and body to be more 3D.
· v1.0 is the same version than the release but it only changes certain merging blocks. It still has blurness but not that much as the release version. Hand generation is the same as V2.
I don't have tested negatives that much so try others if you are not getting good results.
· v1.1_EX has theorical improvements in hand and N-SFW generation. I tested it very little but sometimes it looks worse in samplers like Euler a than v3.0. It will require more steps or resolution.
Negative prompts: Maybe there are better ones but mines are decent.
Samplers: Not tested enough, but this is what I usually recommend for any model:
Euler a Steps recommended: +30 steps. DPM++ samplers steps recommended: 20+ steps.
Looks like best Alpha v1 sampler is DDMI.
Negative prompts are still not tested and would require more testing and time. Test others or change numbers from mine if you want, maybe it will show better results.
For full body is highly recommended to enhance the image of fix it outside PixAI.
Issues: v1_release has too much blurness. It has been fixed in for the following versions but now it less flexible and sometimes it has bad anatomy (missing legs). Maybe negatives can fix those issues, but I'll try to fix it in the next version.
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.
Best samplers I tested yet in general for every model:
Euler a and DDMI = Good quality and better composition (in my opinion)
DPM++ SDE and DPM++ 2M = Cheaper, faster and good quality
DPM++ 2M SDE Karras = Very good quality in some prompts but requires more steps, so it is the most expensive for PixAI.
DPM++ 2S a Karras (& 2S) = It works in rare ocasions. Worth to try in closeups.
