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
FAQ
Comments (9)
what happens when you use the model
LCM lora?
DPO?
HYPERSDXL?
SDXLLIGHT STEPS?
DMD2?
https://huggingface.co/tianweiy/DMD2
https://civitai.com/models/514626?modelVersionId=571908
https://civitai.com/models/216190/lcmandturbomix-lora-only-12mb-8-step-sampling-effect-is-superior-to-using-lcm-or-turbo-alone
https://huggingface.co/ByteDance/SDXL-Lightning
https://civitai.com/models/242825/dpo-direct-preference-optimization-lora-for-xl-and-15-openrail
https://civitai.com/models/195519/lcm-lora-weights-stable-diffusion-acceleration-module
are they gonna work?
Its just a mix of 2 different LCM loras + a strange model mix. Usually all lora accelerators have problems with illustrious because they are trained on SDXL, rarely on Pony. Illustrious is very different from SDXL, so it's difficult to make it work properly (extra/missing body parts and other content artifacts). I have tested different LCM lora and just mixed best looking combination together. It's possible to use other lora of such type, but it's more likely will lead to more strange artifacts.
dmd2 4steps works poorly even on good clear Illustrious models (mutated/missing/extra body parts is a very often case). It works only in 4 steps, so you can't make quality better as 5 or more steps will just ruin image.
This model is just a low quality "quick" solution. If I find something better I will post a new version. Sadly, I have no resources for training some sort of Accelerator for Illustrious. Mixing is the only thing I can do.
It looks like I have some big progress... I will upload v.2 today, I think.
@Shio_N @Shio_N or this? https://huggingface.co/Zuntan/NoobHyperDmd
@daskmaster this is a good one. It looks like it's compatible with any Illustrious checkpoint. Not sure why it's not popular, but thanks for the link. CFG 1 will give you 2x speed on non-cfg samplers, and it's 4-6 steps only which means it's like 2-3 steps. The only minus with CFG 1.0 - it will ignore your negative prompt.
@Shio_N yep, I hope you find this useful. By the way, can I use samplers other than lcm? If so, what CFG should I put on them?
@daskmaster if you are asking about my model: Euler A cfg++ and any cfg in range 0.4-2.5. I'm using 0.7 or 2.3. Low cfg can do more original arts, high cfg is more precise. If you are asking about Hyper DMD - Euler A works good on cfg 1, but it looks like Hyper DMD is ok only with short prompts. If you have more than 10 tags, some of them will be ignored. It's fast but useless for specific cases
Details
Files
Available On (3 platforms)
Same model published on other platforms. May have additional downloads or version variants.