This model is recommended for use with Comfyui
compatibility with Reforge is not guaranteed.
RECOMMEND setting (NTR MIX Xiii)
{your Prompt}
masterpiece, best quality, amazing quality, very aesthetic, absurdres, newest,
you can add tag scenery, (volumetric lighting), To make the background more distinctive.Use dpm ++ 2m sgm uniform or karras (Sometimes euler is okay...?)
Step 25, cfg 5 (can up to 5.8 by looking at the situation)
If you not use LoRA, author tags, etc., up the step to 28 EXTREAM (very scenery) setting (NTR MIX Xii)
masterpiece, best quality, amazing quality, very aesthetic, absurdres, newest, (scenery, volumetric lighting),
{your Prompt}
masterpiece, best quality, amazing quality, very aesthetic, absurdres, newest, (scenery, volumetric lighting)Use dpm ++ 2m karras or sgm uniform
Step 28, cfg 5.8 (can take it down to 5 by looking at the situation)SOFT (char only, simple background) setting (NTR MIX Xii)
Quality tags (please put Bottom of the tag)
{your Prompt}
masterpiece, best quality, amazing quality, very aesthetic, absurdres, newest,Use dpm ++ 2m beta or karras or sgm uniform
Step 20, cfg 5 (up to 5.8 by looking at the situation)EXTREAM (very scenery) setting (NTR MIX Xi)
Use dpm ++ 2m SDE (recommend SGM UNIFORM or Karras or BETA, choose your own!)
Step 28, cfg 5.8 (can take it down to 5.5 by looking at the situation)this need Two Quality tags, bottom and top (refers to the beginning and end of the prompt.)
Quality tags (please put Bottom and Top of the tag)
masterpiece, best quality, amazing quality, very aesthetic, absurdres, newest, (scenery, volumetric lighting),
{your Prompt}
masterpiece, best quality, amazing quality, very aesthetic, absurdres, newest, (scenery, volumetric lighting)SOFT (char only, simple background) setting (NTR MIX Xi)
Use dpm ++ 2m karras
Step 28, cfg 5.8 (can take it down to 5 by looking at the situation)Quality tags (please put Bottom of the tag)
masterpiece, best quality, amazing quality, very aesthetic, absurdres, newest, volumetric lighting
{your Prompt}
masterpiece, best quality, amazing quality, very aesthetic, absurdres, newest, volumetric lightingNegative Tags (all setting)
lowres, (worst quality, bad quality:1.2), bad anatomy, sketch, jpeg artifacts, signature, watermark, old, oldest, censored, bar_censor, (pregnant), chibi, loli, simple background, Recommendation setting (under NTR MIX 777)
Use dpm ++ 2m Karras (in i2i, use BETA and cfg 5)
Step 28, cfg 7Quality tags (please put Bottom of the tag)
masterpiece, best quality, amazing quality, very aesthetic, absurdres, Negative tags
lowres, (worst quality, bad quality:1.2), bad anatomy, sketch, jpeg artifacts, signature, watermark, old, oldest, censored, bar_censor, (pregnant), chibi, loli, simple background, conjoined, futanari, Description
This model is based on NTRMIX 777 (for LoRA) as the default base. It is a model that uses high quality images (top Civitai) from NTRMIX4.0, trained with LoRA, and merged into 777. The advantage of 777 is that it has a high LoRA recognition rate, and the default color scheme has a slightly more 4.0-like feel.
FAQ
Comments (18)
Great model！ thanks for updating!
The weird thing is that any sampler I use except Eular and Eular A results in a lot of noise in the image
tried a few samplers and it's just low quality and a bunch of artifacts, similar to v777(lora) version. Idk what it is, in comparison v4 and 3.5 are near perfect other than the lora compatibility, great hands, composition, no highres needed, etc.
@yourneighbourRiz Agreed.
Xi has better coloring than v4.0. Amazing!
What are the key aspects of Xi? Also, can this model be used on A1111 or does it have to be used on ComfyUI/Forge?
As far as I know, all are available.
Anyone who has used the XI model, please comment on how it feels to use it and I'll think about switching to the V4.
Any chance for On-Site Generation?
somebody loaded it as a graydient workflow, its fun
How is XI with pony loras?
Shouldn't work too well. A drastically different direction of what SDXL is.
I haven't tested the Model Xi enough yet. My first thoughts are that it is more realistic than V4. Better for those who are bored with the anime-oriented style.
The CFG power seems a bit weaker than V4. It produces softer colors even though I used the same values.
I'm not sure about the Lora compatibility yet. LORA compatibility is slightly better. This model can handle higher weights.
It is good at creating a base image ✅ but not suitable for upscaling.❓
It is not consistent at high resolutions. Colors are pale. V4 is more vibrant and colorful while upscaling. I get more consistent results in V4. Maybe I can fix this with different settings. It's too early to reach a definitive conclusion yet. 🧐
I'm on the opposite side regarding base image and upscaling, on Xi.
DPM++ 3M SDE + Turbo seems to do best for me, but HiRes is an absolute must.
@Fatbuns Maybe the settings I'm using are not suitable. I get great results on V4, but weaker results on Xi. I'll try some different things.
I think this needs a different merge/finetuning approach altogether,
v4 was insanely well rounded and clean (minus the large heads and lora stuff) to the point where most other illustrious model felt kinda mid.
Now these feel more like some base IllustriousXL style-prompt dependent models that don't even function without Hires
@yourneighbourRiz I agree. The V4 model is so good it's almost perfect. It still gave good results in most different settings. Only the LORA compatibility was weak, but it's not really necessary. It already recognizes most characters and styles.
[Xii] appears to be closer to v4 than [Xi] is
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.