β¨3rd Edition OUT NOW!!!β¨
NOW OVER 9,700 COMBINED downloads!!!β€οΈ
per sacrificium purificant
This model is the next evolution of a pure-photorealistic fantasy model. Capable of creating any character & any setting you can feasibly conceive!
It's the most detailed & thoroughly made model I've released as of yet & I spent tireless nights fine-tuning this to be the most epic model I could manage. I spent roughly a month generating examples and running it through its paces. It's also capable of using most LORA's quite effectively so you should be able to create whatever you like.
ALL THE MAIN EXAMPLES ARE 1:1 FOLLOWING MY INSTRUCTIONS YOU SHOULD BE ABLE TO REPLICATE THEM EASILY!
This model can do pretty much any style fantasy art at a very high level. The main concepts:
βοΈ fantasy art (retro & modern)
βοΈ characters / rpg characters
βοΈ photorealistic / realistic
βοΈ analog style
βοΈ surrealism
βοΈ oil painting
βοΈ concept art
βοΈ cinematic
βοΈ scifi/science fiction
βοΈ landscapes & scenery
(all of the above list can & should be used as triggers and weights)
It knows most common D&D races and it will produce some interesting great results when prompted correctly. (It has some minor issues with complicated races such as dragonborn & tieflings, etc. but you should be able to use a LORA to render these effectively. <3)
How to use
There is no embedded VAE but it was designed to be used with this:
β vae-ft-mse-840000-ema-pruned.ckpt
I also recommend using the following negatives:
EasyNegative or EasyNegativeV2 (The all around catch all. Always good in a bind)
badhandv4_anime (absolutely your go to neg for anime hands)
bad-hands-5 (good for the more realistic images but will not work amazingly for anime)
bad-artist \ bad-artist-anime (Iβve had good and bad luck)
bad_prompt \ bad_prompt_version2 (works very good sometimes)
(I typically use (bad-hands-5) & (bandhandv4_anime) at the same time to cover all bases.)
(Here are some example settings for you to start with, but please keep in mind YMMV!)
Steps: 30 = Sweet Spot / 10-150
Sampler: DPM++ 2M Karras, DPM++ SDE Karras, Euler A, DDIM,
CFG scale: 4-11.5 = Sweet Spot / 4.5-15
Size: 512x768, 768x512, 640x840, 840x640, 600x896, 896x600, (any size)
Clip skip: 1
Hires Fix: Enabled
Upscale by: 1.5-2
Hires Steps: Equal to Half or Exact the exact amount of Steps (or higher)
Upscaler: Remacri Upscaler, Lollypop Upscaler, 4x_NMKD_DigArt_10k|
( I use DigArt 10k almost exclusively with everything except for 2D illustration & animation )
Description
3RD EDITION / 3.0!!!
Check the main info for more details!
A PHOTOREALISTIC FANTASY MODEL!
Use this vae β vae-ft-mse-840000-ema-pruned.ckpt
New & Improved! More consistency w/ updates for more utility in fantasy & realism.
FAQ
Comments (18)
looks great, but please release a 2gb version at some time. My SDD is so full.
Coming soon
It's up!
@DarkAgentΒ Holy crap! that was quick. Thanks so much, seriously. When people do this it really does help free up valuable space. Cant wait to try it!
For anyone with same request - just prune it yourself, 2Gb will work fine. It is literally 2-5 minutes of time.
This is a SUPERB model.
I've noticed if I use Euler A I get a very nice photographic effect, but if I use DPM++ 2M Karras I get a more painterly artists style.
The Painterly style is more interesting but introduces more chances of coherency problems, but euler a seems stable AF.
Most of it is in the prompting. There is not much emphasis on sampling method.
Euler generally tends to be a most stable method because it's one of the original and oldest sampling methods
@DarkAgentΒ Yes, I'm using literally using hundreds of dynamic loras, some with art styles in them.
I must say your model is handling nearly everything I throw at it. It's incredibly dynamic.
One of the best stable diffusion checkpoints I've ever used, and I'm not just saying that.
THANKS SO MUCH for pruning it for me earlier.
I'm getting great photorealism style results with DDIM, and PLMS as well. EXTREMELY COHERENT.
It it standard behavior of these samplers, this effect is present on all models SD1.5 and SDXL. That's why people use different samplers for specific style.
@littlefluffyballΒ I hope your having fun arguing with people 1 year ago. Things have changed a lot since then.
@DarkAgent Hiya! What is the stylistic difference between this and your dreamscapes and dragonfire checkpoint? I love using the latter but only kinda notice a slight difference between the rendered style of the generation examples. Is this one supposed to embody a more hyper realistic photorealism while the latter is like surrealistic 2.5D? (Full disclosure, I don't really have any idea what I am saying, lol. I just know that I can tell that there is a difference, but can't put my finger on what it is.)
Truth be told.. DS&D was actually just a random happenstance when it was originally made.
It was never originally intended as a photoreal model when made... but when all was said and done it sort of came out that way.
Illustro was a shift to my DIRECT attempt to create a photoreal capable model that still had the chops & diversity of a fantasy model.
They may have similarities for sure... as all models do produce SOME SIMILAR result based off seed range and or similar settings.. but they are not the same in the way they were made.
(2.5D generally is defined as semi-realism meaning it has alot of depth and close to photoreal but not quite)
Illustro is a way more complex model by far and purpose built for photoreal. It was a hard balance to include fantasy into a model and retain photoreal. As fantasy generally is anything but
any plans on making an inpainting model?
Any tips on how to create elves with less extreme ear shapes? I'm thinking "upswept" rather than the wide/horizontal style.
use: (pointed ears:0.5) or -0.3-0.2 till you get desired result
if that doesn't work use: (ear size:0.3) in the negative and dial that in
Images quality of SD1.5 on pair of SDXL, this model in different ligue than other common SD1.5 models. Makes correct faces, hand, feet, no messed limbs.
What is the difference to "Dreamscapes & Dragonfire" model?
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.