🐇Fabled Illusion (N/SFW)🔮 - CivArchive (CivitAI Archive)

👍 Please Rate! - Your feedback, rating or a follow are greatly appreciated. - Thankyou! 😊

🐇 Any yellow Buzz donated to this model will go toward making it freely available for On-site Generation. 🔮

All example images showcase the base model only, rendered in a single pass at high resolutions, with No detailers, LORAs, embeddings, upscaling or extras of any kind.

(The exception being the DMD2 LORA for a couple LCM examples), There's a simplified workflow embedded in each of my v6 gallery images. (just verify the Magic Node settings with the 🗈 notes section at the bottom, as I may have tweaked them slightly). - Standard SDXL VAE is baked.

📷 Photorealism - with a touch of artistic magic. ✨

V6 - Escape Velocity

Enhanced contrast, saturation & black levels.
Further push into a natural language & photorealism bias, while expanding artistic composition.
Photorealism can now be maintained @ higher CFG values than in prior versions.
Expanded sampler / scheduler options
I've started using the (Kohya Deep Shrink) node, (a default node in ComfyUI). It's an absolute game changer in terms of image stability @ higher resolutions. I made a simple workflow for it, (embedded in each of my gallery images). I don't see a point in ever generating in standard res again, (outside of testing).

CFG Range: 3.0 - 6.5 (I typically use 4.0 -4.8).

Samplers / schedulers: DPM++ 2M_SDE_GPU, DPM++ 2M_SDE, DPM++ 3M_SDE_GPU, DPM++ 3M_SDE, Euler_A, heun, dpmpp_2M, DDPM / Simple, SGM uniform, Karras, Exponential, ddim_uniform, beta, normal, AYS (See 🗈 notes for my preferred combinations & use cases).

Prompting: Natural language is King, tags are still relevant & useful for fleshing out details, but should generally be added after a NatLang description of the scene.

(avoiding negative prompts where possible can improve overall quality)

Medium: RAW photo / professional photo / amiture photo etc...  
Summary: photo of a woman enjoying a beverage at the neighborhood bar...
Tags: alluring gaze, light blue eyes, looking at another, etc. 
Background details / Lighting / photocentric descriptors, I.E. film type, lens, blur effects zoom etc.
(See my sample images for more detailed examples),

Steps: 30 - 40, (I use 36 most often, 10 - 16 for LCM / AYS).

Resolution: I've been using multiples 64 as that seems to be most stable, (Some resolutions I use regularly include: 1152 x 1536, 1280 x 1728, 1344 x 1920 1408 x 1920 & sometimes up to 1536 x 2048. Things will stay quite stable up to about 1408 x 1920 provided you're using the Kohya Deep Shrink node. (load one of my sample images in comfyui for a simple workflow that includes it with recommended settings.

See the new 🗈 NOTES section below for a description of the Kohya Deep Shrink node, my recommended settings, best samplers / schedulers, & additional guidance.

Added the following models:

Artaix Realism SDXL v3 by wildzzz
GrAaM_XL_Juicy by graam558
MoP Mix ERP by GBRX
ASTRAL FORGE XL v2 by MINDZ_FUELING_MEDIA
updated dataset / LORA
(See full list of models used / credits below)

V5 - Cinéma Chiaroscuro

(High realism, capable of high contrast chiaroscuro effect & highly cinematic scenes, greater prompt adherence).

A return to a stronger photorealism bias. (use lower CFG: 3.5 - 5-ish to achieve this). (a photorealistic look can still be achieved in the higher ranges, (6.5-ish) by adding film-centric quality tags to your positive prompt: (avoiding negative prompts where possible can improve overall quality)

CFG Range: 2.8 - 9

Samplers / schedulers: DPM++ 2M_SDE, DPM++ 3M_SDE, Euler_A / SGM uniform, Karras

I generally prefer SGM uniform, DPM++ 2M_SDE @ 32-38 steps.

I've been experimenting with higher resolutions to good effect: Below are just some ideas, don't feel constrained to these ratios

Multiple columns for different ratios. (The highest resolutions can be tricky depending on composition). Adding (malformed, distorted, warped, stretched) can be helpful.

1536 x 1536 | 1840 x 1536 | 1584 x 1232 | 1824 x 1248 | 2016 x 1152

1280 x 1280 | 1528 x 1280 | 1440 x 1120 | 1552 x 1040 | 1680 x 960

1024 x 1024 | 1224 x 1024 | 1152 x 896 | 1216 x 832 | 1344 x 768

(I've been using: 1120 x 1440, 1232 x 1584, 1344 x 1728 most often).

Added the following models:

Lustify GGWP (V7)
BATCH XL ( PHOTO REAL)
Cinema Diffuso XL
Real Humans (LORA)
dark (dramatic chiaroscuro lighting) (LORA)
(See full list of models used / credits below)

Adjusted / rebalanced ratios of preexisting models. Updated / retrained custom style LORAs

V4 - Gravitational Lensing

(More artistically capable, greater prompt adherence, a bit more fantasy vs realism).

Mainly a rebalancing / adjusted ratios of existing ingredients.

Added the following models:

itercomp, (for better natural language handling)
Photoart v6 (gives more bias to an HDR look)
SPO-SDXL (lora), (injects noise / clutter into the image greatly improving background details)
(See full list of models used / credits below)

See v3 bellow for settings as they reman unchanged for this version

V3 - Photo Noir

(Capable of much higher resolutions), (> 2 megapixels).

much greater photorealism. (use lower CFG: 1.8 -3-ish to achieve this). More varied and artistic compositions in the higher CFG ranges, (up to 9-ish or higher if you're getting artsy). (a photorealistic look can still be achieved in the higher ranges by adding the following examples to your negative prompt: (unreal, airbrushed, cartoonish, plastic skin)

I've been experimenting with higher resolutions to great effect:

Multiple columns for different ratios. (The highest resolutions can be tricky depending on composition). I still prefer SGM uniform, DPM++ 2M_SDE @ 38 steps. (I occasionally use 48-50) (length & width can be flipped as needed), (See v2 notes for additional sampler / scheduler info).

1536 x 1536 | 1840 x 1536 | 1584 x 1232 | 1824 x 1248 | 2016 x 1152

1280 x 1280 | 1528 x 1280 | 1440 x 1120 | 1552 x 1040 | 1680 x 960

1024 x 1024 | 1224 x 1024 | 1152 x 896 | 1216 x 832 | 1344 x 768

(Standard resolutions and other samplers are still fine, see note from earlier versions).

Added the following models:

Lustufy OLT (Fixed Textures)
The Araminta Experiment FV5
RealVisXL V5.0 (BakedVAE)
Photonic Fusion SDXL Finalé
DetailedEyes_XL V3 (lora)
(See full list of models used / credits below)

Adjusted / rebalanced ratios of preexisting models

v2 - Nue Aspect

(Can now do SFW content equally well)

Greatly improved in all aspects, (far more prompt compliant, SFW images are now quite viable).

the artistic side of the model is more robust and photorealism also improved.

Sampler: DPM++ 2M SDE SGM Uniform, (most consistent good results) - I also use DPM++ 3M SDE. Karras and Exponential work well also.
Clip skip: 1 or 2 (should yield same results)
Resolution: (I often use 1024x1328, 1128x1328, 1024x1280, & 1128x1280 in portrait and landscape).
All standard 1MP resolutions work well in portrait and landscape, (depending on context)
CFG: 2.0 - 9.0 (higher for non-photorealistic - I generally use 3.8 - 6.4)
Steps: 30 - 50 (I use 38 or 48 most often) try 40-50 if your struggling to get a post, or good eyes.

I'm now recommending scheduler: SGM Uniform, (seems most constant for well drawn eyes), Karras and Exponential are still fine options.

neg-promt: "plastic skin" can be helpful as you go over 3.8 CFG (roughly). I similarly use cartoonish & or unreal, with different effects in different contexts.

Prompting:

Primarily natural language, tags can be used also, (look to my sample images for examples) .
negative prompts - aside from "underage" or "child" I don't recommend anything be present for all scenarios. This is very context specific. I always recommend starting simple and adding tags as needed. (one exception is "plastic skin" once I go higher than 3.8 CFG), (look to my sample images for examples)

My samples are uploaded without any upscaling or facial detailer.

Added the following models in v2:

BigAsp v2 (you owe it to yourself and it's creator to try it).
Lustify v2 and v4
NatVis: NaturalVision v2.5
OnlyforNSFW118 v2
ColossusProjectXLSFW v12c

(See full list of models used / credits below)

v1.0 Porn Nouveau

(Can do SFW well, but will often refuse)

VAE: sdxl_vae (included)
Clip skip: 1 or 2 (should yield same results)
Sampler: DPM++ 2M SDE Karras (most consistent good results - I also use 3M with exponential)
Resolution: all standard 1MP resolutions work well in portrait and landscape, (depending on context) (I often use 1024x1328 and 1128x1328 in portrait and landscape)
CFG: 2.0 - 9.0 (higher for non-photorealistic - I generally use 3 - 7)
Steps: 30 - 50 (I use 38 or 48 most often)

Prompting:

Primarily natural language, tags can be used also, (look to my sample images for examples) .
negative prompts - aside from "underage" or "child" I don't recommend anything be present for all scenarios. This is very context specific. I always recommend starting simple and adding tags as needed. (look to my sample images for examples)

My samples are uploaded without any upscaling or facial detailer.

v1.0 is a bit over trained for sex. If using a LORA to reinforce a pose / concept, I generally recommend low strengths.

🗈 Notes & Tips:

(to be expanded over time),

Preferred Sampler / Scheduler combos: in order preference: 
DPM++ 2M_SDE_GPU,  / Simple, SGM_uniform, beta, 
DPM++ 2M_SDE,      / SGM_uniform, Simple, beta, 
DPM++ 3M_SDE_GPU,  / Simple, SGM_uniform, beta,
Euler_A,           / Simple, beta, SGM_uniform, exponential,
DPM++ 3M_SDE,      / Simple, SGM_uniform, beta,
heun,              / Simple, beta, (Sometimes garbage, sometimes pure gold)
dpmpp_2M,          / beta, (same as the last one, hit or miss depending on scenario, but when it behaves, it's a clear winner, after some trial & error it becomes more predictable).
DDPM               / ddim_uniform, (High accuracy with clean details, a bit of a washout - overexposed effect, but with higher CFG's & / or charged terms in your prompting, this can become the best option.)
Karras, tends to overcook things and glitch eyes, I almost never use it, (I know people swear by it, it's great for many scenarios / models but generally not with mine. 
Exponential, pairs well with LCM to mitigate washout
AYS, works well, but I prefer LCM with this model.
------
Patch Model Add Downscale (Kohya Deep Shrink), 
is a default node in Comfy that greatly increases image stability by compressing the Unet model during the 1st compositional stage of image generation then re-expanding it for fine details.
Parameters of concern include: 
Block number: (3 is default but 4 can also be useful in some instances).
Downscale Factor: (I use 1.5 as a default but will go as low as 1.25 for resolutions close to standard, & sometimes as high as 2 if pushing the upper limits.
End Percent: (this is the percentage of total steps, the point at witch the unet will decompress. You can generally leave this at 0.35, but I sometimes make slight adjustments if I'm trying to eliminate a pesky error.
-------