v4.3 [final]
I recommend to choose this one or v2. If you don't care about speed and want great anime results use this model + LCM sampler + 20 steps. It's simply better than any model I know in this aspect.
Recommended settings:
6+ steps (8 + 4 steps of upscaler is optimal for my use cases, feel free to experiment).
A few options of samplers:
CFG 1.8 | Euler A
CFG 2.0 | LCM - better for anime and other abstract things.
CFG 1.8 | DPM++ 2M SDE with BETA scheduler (normal doesn't work)
Use style tags such as "anime style, flat coloring" or artist/title name. It will help model to know what you want from it.
The fastest generation: use CFG 1.0 and set lora CFG_slider on negative -0.4 value. Lora listed in suggested resources. (CFG 1.0 ignores negative prompt. Don't use it if you need negative)
The best coherency: use CFG 2.8 and CFG_slider lora on +0.75 weight (for Euler A).
v3.2B
Recommended settings:
CFG 1.8
LCM / [kl optimal / sgm uniform]
5-10 steps
v3.2A
Recommended settings:
CFG 1.8 (works not bad on cfg 1 as well)
Euler A / normal
5-10 steps
v1.99 (v-prediction)
Use title or artist name to get a specific style.
Eauler A - Beta scheduler (or normal)
8-16 steps
CFG: 1.0 recommended for 2x speed. Use more if you want to use negative prompt.
v2
Sampler: Euler A / Euler A cfg++
CFG: (depends on sampler)
Alternative sampler: LCM
Description
Use v2 if you want better art styles. This one is good in other things.
FAQ
Comments (5)
Wow super fast model and the quality is still very good
OH MY GOD! I don't know who you are, but you didn't sell your soul to make this masterpiece?) I get simply amazing pictures with this model, and besides, its speed is like switching from a bicycle to a Lamborghini.. The styles are very easy to change and adjust, I am simply amazed to the depths of my soul by this model.. The highest respect! 👍👍👍
After some time I found some downsides in this model
1 - The main problem is that very often the model wants to draw extra people. For example: 2 are standing, a guy and a girl, the guy puts his hand on the girl's shoulder and hugs her, that's how it should be, but in reality, the model REALLY WANTS to draw a second guy in the middle, or a girl, or several people, faces, in the background. Which should not be there. It even happens that you make pictures where the girl is just standing, and another one appears in the background, and even the solo tag can not always help. Although the second case is still rare, usually problems are when there should be 2 people, especially 2 girls.
2 - The model can be very difficult to set up to get a good background, and not something very abstract. Although if everything is done correctly, the backgrounds are amazing.
3 - It seems that the model does not care at all about your negatives, at least in most cases, and I twisted CFG if anything.
I have now written some criticism, but without negativity, I just shared my experience. This model is still amazing, and I will continue to use it
It's just a mix without extra training, so it can't be as stable as other trained models. Stability breaks after almost any mixing and it's recommended to do an extra training to improve relations among weights. I have no resources to do so.
Negative is a useful feature, but with some extra loras you can generate good images on CFG 1.0 which is 2x times faster. CFG 1 ignores negative prompt completely.
If you want a very good prompt coherence use this model: https://civitai.com/models/1308285?modelVersionId=1623762 . I have tested v.2.0 - it's really good. But it have downsides - scenes look very generic without a long prompt and a lot of details. This model is not for lazy prompters.
I would prefer my mix for original, fast generation with small prompts.
@Shio_N It's a pity that you don't have the resources to develop this model, because it's amazing and deserves more...
I'll try to download the model that you advised me. Let's see what results it will show me)
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.