Stable Virsion: 2.0
Quality Tag:
Anime example: (anime coloring:2.4), close up, face focus, breasts, focus, (masterpiece:1.1), (best quality:1.1), (newest:1.1), (absurdres:1.1), (highres:1.1)
I used the Mikoyan(style) tag during development and submitted samples, so if you add the Mikoyan(style) tag, it may be easier to stabilize the drawing.
I'd like a prompt for model performance testing...
Base model is Hassaku XL (Illustrious)
WIP Now!!
I'll upgrade it gradually to suit my tastes
v1.0 I tried to give the body a three-dimensional feel.
v1.1 Improved lighting and rendering stability
v1.2 Changes to coloring and improved drawing stability with almost no change to the drawing content
v1.3 It wasn't bad, but it didn't seem like it was of a quality worth releasing, so I skipped it.
v1.4 I made some adjustments to make the expression look stronger. The drawing may have become a little easier to break down.
v2.0 Change to manga style
v2.1 Changed to give it texture(Weapon rendering has become unstable
v2.2 I aimed for a more manga-like expression.
v2.0a It was developed based on v1.4 to prevent the pose from collapsing, and v2.2 was used, but since the features unique to the 1.x series have been reduced, the next version may be modified based on 1.4 again.
I've made so many that I don't know what's good anymore
Description
Sampling method:LCMSchedule type:SGMUniformSteps:4CFG Scale:1.0Hires. fix:x1.5Hires. fix 使用前提です。Hires. fix steps:4Hires. Upscaler:Latent (nearest-exact)Latent 系以外は遅くなります。Denoising strength:0.7LoRA Weight:1.0
GrayColor CustomModel is meant to be flexible, so I think it can create all kinds of concept art depending on the tags — not just anime, but other styles too.
FAQ
Comments (16)
I noticed in your Mordred images for 1.4 you used the "shingeki no kyojin style" tag - is that a built-in to the base, something you trained in, or a LoRA/embedding that's not listed in the resources?
I’d like you to try it with various models, but it seems that most popular anime are rendered in imitation using the [Anime Name style] tag.
@GrayColor So I just tried it:
- shingeki no kyojin, pokemon, Spy x Family, Gankutsuou, Ookami to Koushinryou, Aggressive Retsuko style
- 6 test prompts
- 5 models: WAI-NSFW, YiffyMix 61, Anime Screenshot Merge (Noob), TewiNAIV40, and of course, GrayColor 1.4
And it seems like a pretty weak method of controlling style - the effects were extremely intermittent, fairly weak, and highly dependent on the specific prompt (short prompts lacking additional details did better). They were also more likely to cause concept bleed (recycled faces and features) than have any significant style impact.
You're absolutely right that the effect is subtle and inconsistent — especially with certain prompts or models. That said, there is a definite effect, even if it’s relatively weak. With the right combination of tags and more detailed prompts, you can tease out style influences more clearly, though it’s admittedly not a strong or reliable method on its own.
By the way, in the case of Shingeki no Kyojin, it seems to lean more toward the anime’s visual style rather than the manga.
Nice to meet you, I am using GrayColor - CustomModel. However, the generated images often contain logos and artist names. How can I fix this?
If possible, could you share one of the images where logos or artist names appear?
I'd also love to know what kind of prompt you used when generating it — that way, I can better understand the issue and help you find a solution more accurately.
I'm sorry. I think it's probably because it was generated simply with Euler A like the previous Illustrious.
You're always a great help! I'll give it a try!
great model but doesn't be affected by negative prompt?
This is the best model I've ever used. It's been like a year and I keep going back to this model. Amazing work!
We would appreciate it if you could update to the v2.2 series.
I like v2.2 the best.
Thank you! I'm happy to hear you like the v2.2 series.
I'd love to hear your thoughts—are there any specific improvements or features you'd like in future updates?
@GrayColor
I really like the art style, but I’m a bit concerned that the structure of the hands, arms and legs tends to collapse easily. So, I’d be happy if that aspect could be improved.
If I include “realistic, semi real” in the prompt, v2.2 generates an art style that I really like.
I'm also looking forward to this new version.
Thank you! I'm happy to hear you like the v2.2? series.
I'd love to hear your thoughts—are there any specific improvements or features you'd like in future updates?
Details
Files
graycolorCustommodel_v14HyperdmdRequest.safetensors
Mirrors
graycolorCustommodel_v14HyperdmdRequest.safetensors
graycolorCustommodel_v14HyperdmdRequest.safetensors
graycolorCustommodel_v14HyperdmdRequest.safetensors
graycolorCustommodel_v14HyperdmdRequest.safetensors
gr4y_color_1_4.safetensors
graycolorCustommodel_v14HyperdmdRequest.safetensors
graycolorCustommodel_v14HyperdmdRequest.safetensors