Update April 28th Added version V4 pruned model.
A model with professional RPG industry artists in mind. Those working in video games, board and tabletop games as well as concept art and book covers should get good use from this model. The new version, V4, packs in more training for creatures and RPG subjects including a more illustrative style.
Detailed tutorial on how I get the results in the preview images.
Check here if you're having trouble getting the same results. Initial generation size and VAE usage is key.
You can prompt any style you need with these models, but the default aesthetic is listed for each of the models in this handy list.
Different models available, check the blue tabs above the images up top:
Stable Diffusion 1.5 (512) versions:
V4 inpaint Inpainting version of V4 that's good for outpainting.
V4+VAE Same as V4 but with the added convenience of having a preset VAE baked in so you don't need to select that each time. Saves on vram usage and possible NaN errors. Speeds up workflow if that's the VAE you're going to use anyway.
V4 New aesthetic added to make it a bit more illustrative. More training for better coherency for all things RPG.
V3+VAE Same as V3 but with the added convenience of having a preset VAE baked in so you don't need to select that each time. Saves on vram usage and possible NaN errors. Speeds up workflow if that's the VAE you're going to use anyway.
V3 Stronger painterly style. High contrast and sharpness. Even more RPG knowledge.
V3 inpaint Inpainting version of V3 that's good for outpainting.
V2 Stronger painterly style. High contrast and sharpness. More RPG knowledge.
V2 offset Noise Offset added making more contrast and bringing the model back to photoreal.
V2 Art Trained model. Very artsy. Strongest painterly style. Less details and bigger brush strokes to mimic digital painting style pre-AI.
V2 inpaint Inpainting version of V2 that's good for outpainting.
V1 Smoother renders with least painterly effect.
V1 inpaint Inpainting version of V1 that's good for outpainting.
Stable Diffusion 2.1 (768) versions:
SD 2.1 768 V1 Strong painterly style, very coherent with hands and objects. Higher native resolution and detail. Not good for nudity.
Not as effective, but here's the LoRA if you like to use that instead:
A to Zovya RPG Artist's Tools LoRA
Do you have requests? I've been putting in many more hours lately with this. That's my problem, not yours. But if you'd like to tip me, buy me a beer. Beer encourages me to ignore work and make AI models instead. Tip and make a request. I'll give it a shot if I can. Here at Ko-Fi
Description
Stable Diffusion 1.5 (512 x 512)
A-Zovya RPG Artist Tools V3 vae
Set VAE to "automatic" in settings
FAQ
Comments (56)
Slightly confused. I found this on the newest SD 2.1768 models page but here it says its SD 1.5? maybe incorrect tag?
In the description, the SD2.1 models are broken out. You'll see different tabs across the top.
Great model! I wonder where I can find the list of the styles I can use as prompts. You say in de description that there's a list but I can't see it anywere.... THANKS!
One of the best models I have used so far, just loving it! Keep up the good work.
Thank you
Apologies if this is a dumb question. I would use the inpaint version of this checkpoint to fix details not to create the original picture?
Yes, exactly, and you can use it for outpainting (expanding an image). No dumb questions, we all start somewhere, and this thread will undoubtedly help someone else out in the future.
So I have to download the V3 lets say and then the inpaint as well and switch between them. One to create from scratch, and other to improve (inpaint)?
@nehilaiart yep that's right. Although, most people just use the regular version for inpainting too since it does a good job, saves switching. but for outpainting, the inpainting model is a must for that process.
Hello Zovya, I noticed that using antonio_riolo's hair collection, the pigtails lora makes character young in your model. I tried putting your adult neg+ as adult embeddings but it still generates young characters. I remove the lora and it's back to normal age, please look into it.
When ı use the same prompt and exactly same things what u shared my paintings look like more dead uncolored unshined they are literally like pixel to pixel and looks oldy how can ı fix this
need to use a VAE
@Zovya thanks
I love the V2 Art model. Are there any chances we get one for V3?
YES!
Good model but there are artist's signatures and watermark texts everywhere. Almost 90% of the images have it. Negative prompts can't take them out.
yeah, starting way back from the beginning of 1.5 people were very cavalier about training without removing signatures and they persist up till today. best way to remove them is inpainting.
@Zovya Don't get me wrong. It was not a negative comment. I'm grateful for the work done.
It is a great model but signatures are useless because they aren't even readable. They look like signatures but in fact, they are just a sketchy bunch of random characters. Not even useful to tell the viewer who is the artist. I agree it is pretty easy to clean it up with Photoshop but the problem is when you create a lot of images, it becomes very time-consuming.
Is there a way, in a future update to remove them, or train some kind of negative IT to remove them? (just a suggestion)
@Scofano I didn't take it as negative. Signatures is just a thing. probably will be for a long time. Like bad hands. People who train are too lazy to remove them before training so it'll be a problem in future models too most likely. Think of it as a reminder of all the hard work of artist's images that went into making all this possible.
@Scofano Also lol if you're going to agree with something the checkpoint creator says, you might wanna make sure they actually said it XD. He said best way to remove them is inpainting, not photoshop. inpainting makes it incredibly easy to remove watermarks. granted, you could make it even easier on yourself by preprocessing the watermark in GIMP or Photoshop first, use the fuzzy select tool to vaguely select the watermarks, then try using color to alpha on whatever the logo color is.
But I think the easiest way by far is to just use batch processing with afterdetailer enabled, detect something that will be in your image( hand, face, person, etc) and just invert the mask.
bring your image into img2img, set the img2img denoising strength to 0
have the adetailer prompt be the prompt that generated the image, including:
(high quality, best quality:1.4), if it isn't already there.
BREAK ((missing [[[watermark]]], removed signature, artist request)) - the milage for this one may vary, most of the time it won't help
and whatever other quality boosters you need
and then for negative, just put, (watermark, text, artist logo, signature, signed by artist), (low quality, worst quality:1.4), in the negative prompt.
adetailer denoising strength for this should probably be between 3 and 4, but play around with it.
in adetailer, enable the controlnet at the bottom, select inpainting, and then change the model to 'inpaint global harmonious'
If you find it's not working, like nothing is changing try changing the adetailer model, or the detection settings. you're just trying to get it to detect ANYTHING, then it should work.
Of course you may notice I didn't mention just regular inpainting, and that's because you'd have to do it one by one that way obviously.
Try putting (text, watermark, signature, signed, writing:2) in the negative. Extreme upweight helps remove them.
Hi, getting great results so far. How many images do you have in the dataset for the v3 model?
Thanks. Wow that's hard to say cause I'm always training more into it. Hundreds at this point I'd guess.
a-Zovya, I must say hands down best fantasy model!
This Model is amazing for my online tabletop game. I have generated 10,000s of NPC images for all types of classes, jobs, people etc with it using wildcards.
It has made presenting NPCs 100x easier when I can put an image to the face for them to reference and remember them each by!
Here is LOTS of examples. If anyone wants to use any of the pics, feel free to use them however you wish. (if linking the pics is allowed)
http://dxwarlock.com/npc/
very cool. Thanks!
Can you share your prompts?
@charuru https://civitai.com/articles/1291
A full article on how I made them
My absolute favorite non photo realism model.
I only wish it could support more horror, and eldritch monsters instead of relying on loras. It would be perfect for Cthulhu rpgs . Maybe future models could be trained on Chaosium art? Specifically CoC 7e art?
no model can do it all, thus loras. but I train more rpg creatures and stuff into it each time. v4 will be even better
@Zovya can't wait for an sdxl version.
Hey this is a great model! Minor question, when I try to replicate your results (specifically for V3+VAE), how come I couldn't generate the exact same image? It still looks phenomenal so it's not a big deal, but I'm just curious if I installed everything correctly. For your medieval ship image, for example, here's the generation data I used:
from side heavy capital class battleship medival pirate ship in a nebula (asteroid field:1.2) action camera view epic render Style-Volumetric (zrpgstyle) (masterpiece:1.2) (illustration:1.2) (best quality:1.2) (detailed) (intricate) (8k) (HDR) (cinematic lighting) (sharp focus)
Negative prompt: symmetrical (crooked) (broken) (bent) (nude) (bloom) (cartoon) (photo photography photograph) (bad hands) (disfigured) (grain) (deformed) (poorly drawn) (mutilated) (lowres) (deformed) (lowpoly) (CG) (3d) (blurry) (out-of-focus) (depth_of_field) (duplicate) (frame) (border) (watermark) (label) (signature) (text) (cropped) (artifacts)
Steps: 30, Sampler: DPM++ 2S a Karras, CFG scale: 7, Seed: 1817092183, Size: 512x768, Model hash: 8c4042921a, Model: A-Zovya_RPG_Artist_Tools_V3_VAE, Denoising strength: 0.4, Hires upscale: 2, Hires upscaler: 4x_foolhardy_Remacri, Version: ## 1.4.0
Another question: is separate VAE required for inpainting model?
Hello, i hope that's okay for me to answer your question. If you have selected GPU generation instead CPU generation, result is understandble. Because with GPU generation, every seed even if it is the same will change slighly as output picture. If creator has generated his results with GPU generation, most likely you won't get same result with your pc if it is not exact to the creator's.
@Reelai ah gotcha, that makes sense! ty
Whenever I ask it to make a character with white or blond hair it always gives them black eyebrows, any ideas for getting around that?
No. Even inpainting will not help you with that. Some models are slightly better with lighter-colored eyebrows, but most of the time you need to paint manually with an image editor.
This is a great model! It improved the druid I was generating 1000x! Only issue is it doesn't seem to understand that a loxodon is a humanoid elephant, not a normal elephant.
Great work!
I humbly would like to see a model or LORA that can handle equipment better. That is when asking for a sword I usually get nothing or a limp short spear thing. Asking for a bow I get a fishing line, maybe. Asking for a crossbow, I get absolutely nothing.
Always looking for any model that can handle requests for 16:9 images and pulling out from the subject. I know most of the training images are portraits. I don't want portraits or close-up glamour shots. I tend to try to make images that show my character in an environment or on an adventure. Also interested in images that I can use for my windows desktop background. :)
Your work is amazing. Thank you for the time and effort you've put into this, it's greatly appreciated.
Your tutorials and tips have been clear and to the point. Conciseness is a talent many lack (myself included) and it's nice to have good instructions that don't attempt explain advanced AI algorithms and quantum mechanics in detail.
This gives amazing characters, but I get very gray and quite grainy results? Size, sampling method and sample amount does not affect.
Any chances for a SDXL model?
Easily one of the best models out there. I hope you will make an SDXL update
Please make an SDXL model! I can't wait!
Thumbs up this comment for an SDXL version!
Don't forget to donate if you can.
Im a big fan of science fiction,with this model i created haunted space stations,lots of ships,images like from star wars of star trek,even aliens and scenes from horror movies.
This model is just masterpiece because it can create creatures and every area in science fiction you can imagine.
There is no better model in this area,period!
Great job my friend,and i also like Children stories its another masterpiece of yours.
BTW Steampunk embeeding is crazy good,i can finally build stuff from Falout 4,in combination with RPGArtistTools model is just fantastic.
Quickly become my go-to for the modern age of RPGing in the digital space.
When using the Inpainting version, does anyone have a recommendation for a VAE to use? I usually use the baked in version but even there it sometimes gives me grainy, desaturated results for some reason.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.