A realistic merge with Illustrious and my SDXL merge models.
V4 Ascension
Better, more natural realism and lighting and skin detail.
This model outputs better with square images around 1280x1280.
V3 Rapture
This version has better skin realism and better faces.
Removes the washed-out colour saturation look happening on some images.
In my workflow I am using a CFG of 1.2 but on site it may need to be higher.
V2 Revelations
Better photo realism an works great with my tuned ComfyUI Workflow
V1 Genesis
It seems to like a low cfg of around 2-4 instead of the usual higher values.
Scheduler: Euler a/ancestral - Normal , 20+ steps. CFG 1.1 - 2.0
I would highly suggest you try my ComfyUI Workflow built for this model: https://civarchive.com/models/1255024?modelVersionId=1591700
Description
Better, more natural realism and lighting and skin detail
FAQ
Comments (17)
Workflow is broken. I get "Cannot read properties of undefined (reading 'output')"
Maybe just change the Image Save node to a standard ComfyUI one if you don't have the Image Saver Custom node, it could be that anyway.
I'm receiving the same error. It seems to be stemming from the cg-use-everywhere nodes, but I don't see any logs on why it's occurring.
Hi Jib, tnx for the model!
Whats the TensorRT loader ? looks like you used another type of model while generating the example images.
when I use your flow with the .safetensors model, the results are quite different.
TensorRT is a way to decreas generation time by around 50% on Nvidia cards, it is pretty difficult to setup on Windows: https://github.com/comfyanonymous/ComfyUI_TensorRT
But AI's can help to install all the prerequisites.
But it doesn't change the output quality, what samplers/schedulers are your trying?
The noise injection nodes in my workflow help but I can also get good images from it on Civitai.
@J1B聽I鈥檒l give it a try, thanks!!
Love the V4 version. Seems to produce a slightly more mature base model, than before. A lot less teen like than other base models. Cheers.
Oh, and J1B, Thank you for all the tips.
Acention v4.0 馃榿 great Checkpoint 馃ぉ BIG Thanks
V4 "..square images around 1280x1280"
驴Could you recomend another resolution like 832x1216? 驴Using another resolutions is worse?? 驴why?
It just quite often messes up the face shapes at longer aspect ratios but you still can get some good images out of them, its just less consistent.
VERSION3 = young people, about 20 years, blurred skin detains.
VERSION4 = average age people, more mature > 30 years old = better skin details than v3
*VERSION4 = when rendering an extreme closeup head.. the image has distorsion and deformed at 1280x1280. Looks fine at 832x1216.
After serious testing, i'd say V4 Ascension is if not the best, it's right there fighting for the top spot for realistic IL checkpoints. 馃┓
And there are a few good ones in that top3. So yes, Ascension is THAT good.
And now the long rant for anyone bored at work/college and nothing better to do than read all this:
V4 (as did the V3 before it also tbh) has some specially potent magic mixed in it that somehow gives most variety in backgrounds without specifying each detail, but quite simply has a fun time rendering images because you get both what you want in the images, but also cool background and scenarios and happy accidents without big effort. So can leave prompts open for exploring and fun results leading to cool images.
+ Skin detail is awesome and you don't have to type in prompt about skin details for each imperfections etc, you get all that automatically and can just add if needs something specific, the skin you get with simple prompts is top tier realistic. No skin loras etc needed at all.
+ Lighting and atmospheres really good, probably easier than in most other competitors to do really cool night photos especially without having to use any Loras for it (or fight a long prompt fight), nights look actually dark if wants to, can do a true Nordic forest edgy photos as much as my inner gothic heart desires lol
+ Very good and easy to get 2K and 4K images without long workflow.. can do straight upscale and no artifacts or scuffed quality, but insanely good quality and fast.
Actually instant classic giga good checkpoint for IL realism. So good so, that any other checkpoint creators should take note and/or dip some of this in their own IL checkpoints like it was some spicy treat, because it is.
J1b, this is on par with some of your SDXL classics, your SDXL V18 is amazing, as is the true undying classic, a "hidden gem" known as the V16 Aphrodite, the full fp32 version (because the "full version" is like an album collector's edition lmao) is one of the few checkpoints in my everlast list that are always in use, no matter how many months/years from their release and how newer checkpoints get fancy things improved, they have that special something.
Big thanks for your magical work!
* Eclipsia's trademark courtsy bow *
Thanks for your kind words.
It was my main intention when I started out to make the best looking realistic Illustrious model on Civitai as I though that was achievable, and yeah it is up there. I will probably be working on Z-image, Qwen-Image and Flux Klein for a while, but I will keep my eyes open and if I have any ideas for improving my Illustrious model I may release an updated version in the future.
this model is great for adding skin detail after generating the main composition. super low denoise (0.05-0.1) pass on masked skin areas = really nice!
This checkpoint is fantastic
V3 has its quirks but I would prefer it over V4.