▼ Recommended Settings ▼
Negative Prompt
Sampling Method
DPM++ SDE Karras
Sampling Steps
12 (8 ≈ 16)
Restore Faces
Off
Hires Fix (! Required !)
On
Upscaler: Latent (bicubic antialised)
Hires Steps: 5 (4 ≈ 10)
Denoising Strenght: 0.55 (0.4 ≈ 0.7)
Upscale by: 2
CFG Scale
3 (2 ≈ 5)
Clip Skip
1 (1 ≈ 2)
VAE
None (Baked)
▼ Merge Based On ▼
▼ More Info ▼
Any questions or suggestions you can contact me through Discord Cisney Gassai#5108 or Twitter @Realisian.
For donations or requests you can do it on my Ko-fi page <3.
Description
Models updated
Recipe changed
LoRAs merged
FAQ
Comments (29)
One of the best models. V6 is fantastic.
Produces good quality results, but it only produces slight variations of the same Asian female model, which unfortunately makes it quite limiting.
Try adding more defining features so you can get a different character.
Sometimes my prompts will choose a random face and keep trying to use that same face, but once I add any specific hair color, race, or outfit it changes.
Just like nature! Hashtag got’em.
Any plan for SDXL?
Not really. I don't see a big enough difference in quality for SDXL models. You can generate really high resolutions with SD1.5 models like this one if you tweak some settings a bit to prevent weird stuff. Look at this one, for example: https://civitai.com/images/6280034. I forgot to enable hypertile for VAE, so there are some artifacts in the colors, but as you can see, the resolution is really high.
@Cisney_Gassai SDXL follows the prompt way better than SD1.5
@Cisney_Gassai especially sdxl generates hands way better than sd1.5
@drusniel @jumpmann maybe you guys are right, that's something I haven't tested, are there any models you recommend me to try?
@Cisney_Gassai Pony Diffusion V6 XL is the best anime model imo but i've not found the best realistic model yet. Anyway i belive you can feel the power of sdxl with Pony XL
@Cisney_Gassai I like RealVisXL models. Those are the most balanced in my opinion.
@Cisney_Gassai JuggernautXL and AlbedoBaseXL for realistic images
One of the best generators
Can you add black people to the model please
Use this lora it gives very good results:
https://civitai.com/models/112594/skin-tone-slider-lora
A very pleasant mix. Given the set of models on which it is based, it would be strange if it turned out to be bad... ;) I liked the light unobtrusive Asian touch on the faces and the muted colors.
The only thing is, a slight feminine focus turns some male characters into women or too feminine. Otherwise, everything is fine. A good merge!
Hi, does anyone know how to get a human character to lay on a bed on their back or on their side. Thanks
lying, on back, from above
Should do.
@Cisney_Gassai Thanks, very helpful
do you provide fp32/unpruned version? i want to try using it to train LoRA
Sadly no... :c, but I'll keep it in mind if I ever make a new version <3.
6
This works great, good faces ngl
This model is not available to me. How are people still able to use it?
Sadly you have to use it locally, civitai now only allows top 100 bidded models on their on-site generator.
@Cisney_Gassai Thanks for the reply. Is the model available on any other platform like Civitai? I don't have the technical chops to run it or on a remote compute.
@adroit17084870 I'm sure it is but not uploaded by me so I don't really know those other platforms.
i have been using this for a month now
the only setting that works for me are (ComfyUi)
sampler: ddpm
scheduler: beta