iLLuSTRiouS from HaDeS

🌀 Illustrious from HaDeS — v2 Series v1.35.X

🔧 Major upgrade in the v2 lineup — Closer to the 2.5D ideal

This version is a significant improvement over the original v2 release. The v2 series aims to deliver a highly aesthetic 2.5D-realistic visual style — staying just near photorealism without ever fully crossing into it. With v1.35.X, we’re getting even closer to that sweet spot.

✅ What's new in v1.35.X:

• Stronger prompt adherence

• Improved shadow rendering and lighting gradients

• Enhanced depth and surface control

• Better consistency across outputs

• Partial mitigation of unwanted photorealism (still happens occasionally — see notes below)

⚠️ Notes on model behavior:

The model may still occasionally produce fully photorealistic images, which is not the intended outcome. This is a known limitation — I'm working to fully lock the output into the 2.5D space in future updates.

Unlike the v1 series, v2 is not plug-and-play.
It requires thoughtful prompting, carefully chosen activation keywords, and a solid negative prompt. These vary from image to image, so I can’t provide a universal guide — it’s a model for those who enjoy exploring and fine-tuning.

However you can check the following link for more information about the v2 series:

Unlocking the Full Potential of Illustrious from HaDeS v2 Series.

⚠️ If you're not an advanced user or prefer quick results, I strongly recommend downloading the v1 Series Final Stable, which produces excellent outputs with almost any prompt and requires little to no tweaking.

This model is not for everyone — but for those who like to experiment, tweak, and refine, the reward is huge once you understand where this model excels and where it struggles.

🔁 Recommended Settings:

• Sampling method: Euler A or SDE++ 2M Karras

• Sampling steps: 30

• Hires fix:

  • Upscaler: 4x_fatal_Anime_500000_G

  • Steps: 4

  • Denoising strength: 0.05

• Resolution: 1152 × 1536

• CFG Scale: 5-7

• VAE: Baked

These settings provide better shadow control, highlight shaping, and stable 2.5D depth — especially with the suggested upscaler.

⚠️⚠️⚠️ WARNING ⚠️⚠️⚠️

🔥 V1 SERIES FINAL STABLE 🔥

🚫 WORKS BETTER WITHOUT NEGATIVE PROMPT 🚫

💡 Focus only on what you WANT to see ✨🎯

🔥 Illustrious from HaDeS — v1 Series Final Stable

🌟 Final official release of the V1 series

🔧 Key improvements over previous versions (v1.0.12 / v1.0.16):

• 🎨 Enhanced color fidelity: richer, more consistent and vivid tones

• 📌 Strongest prompt adherence: better responsiveness without sacrificing quality

• 🧬 New VAE: now using EQ-SDXL.VAE by KBlueLeaf, delivering better contrast and visual definition

• 🧱 Rock-solid output stability: this is the most refined and polished version in the entire V1 lineup

• ⚠️ WARNING ⚠️

  ❌ Negative Prompt is NOT necessary

  🚫 Not even recommended

  🎯 Just describe what you want to see ✨

🛑 This is the final update of the v1 series.
Future development will continue in new branches (V2 and experimental lines).

🔽 If you’ve worked with v1.0.12 or v1.0.16, this version is a clear upgrade across the board while staying true to the Illustrious from HaDeS vision.

🌀 Illustrious from HaDeS — v2 Series 🔥

✨ A true Evolution

After the strong reception of v1.0.16, I’ve spent tons of hours training, fine-tuning, and testing what is now a new generation:
Illustrious from HaDeS — v2 Series
A model I actually began training even before releasing v1.0.12.

The v2 Series doesn’t just improve rendering and realism — it redefines the balance between emotion, texture, anatomical coherence, and visual storytelling.

It’s a highly flexible model capable of generating both stylized 2D with depth (2.5D) and semi-realistic scenes, and with the right prompt, it can even approach true photorealism.

Prompt guide here.

🧠 Based on: Illustrious from HaDeS v1.0 and trained at the same time
📌 Status: Stable Release
🔧 Type: SFW + NSFW (explicit capable if prompted)
📩 For commercial collaborations → [email protected]

🔥 What’s Better in v2 Series Compared to v1.0.16

• 2.5D Rendering — Adds volume and depth to 2D style without losing stylization

• Semi-realism & Photorealism — Can approach realistic rendering with the right prompt

• Improved Facial Anatomy — More accurate proportions, better expression control

• Emotional Subtlety — Faces now convey believable emotions, not exaggerated reactions

• Skin Detail — Visible pores, gradients, and soft lighting with subsurface scattering

• Clothing Realism — Fabrics show tension, natural folds, and proper light interaction

• Natural Hair Rendering — Layered strands, soft shadows, and believable highlights

• Scene Integration — Characters blend naturally into backgrounds with proper lighting

• Crowd Variety — Secondary characters now show variation, emotion, and positioning

• Clean Output — Great anatomy, minimal glitches, even at native resolution

• Prompt Flexibility — Works across styles: stylized, semi-realistic, cinematic

• Hires Stability — Excellent results when using Hires.Fix — no need for re-upscale

• Better Artifact Control — Less distortion in hands, eyes, and dynamic angles

• Cinematic Composition — Great for storytelling, full-body shots, and scenes

• More Consistent Results — Less trial and error; model understands structure better

⚙️ Recommended Generation Settings

📏 Native Resolution: 1024x1408
Fast and highly stable. Ideal for testing or base generations.

🖼️ Sample Images Configuration:
All showcased examples were generated at 1152x1536 with Hires.Fix x2, producing outputs around ~2304x3072.
This setup delivers exceptionally high quality with minimal need for post-processing.

🌀 Sampler: Euler a
Simple and extremely stable — consistently gave the best results in testing.
Other samplers (like DPM++ 2M SDE Karras) work as well but may introduce slight variance.

🔢 Steps:

• 20–30 → For simple prompts or fast previews

• 30 → For standard scene complexity

• 30+ → Recommended for ultra-high quality or complex prompts.

🎚️ CFG (Classifier-Free Guidance):

• Range: 5 to 7

  • 5 → Best for realism and soft transitions

  • 7 → Adds contrast, sharper edges, or stylized dramatism

💡 Negative Prompt Usage:
For optimal results, a well-balanced negative prompt is recommended.
It helps maintain clean anatomy, avoid common artifacts, and reduce visual noise — especially in high-res generations or busy scenes.

Prompt guide here.

🧬 Why “v2 Series”?

The name v2 Series reflects the evolution of the model into a multi-branch generation. This is not a single update — it's the beginning of a modular lineup.

Alongside this main branch, designed for versatility and powerful 2.5D rendering, there will be additional versions:

• A flat stylized variant, focused on pure 2D output with zero embellishment

• A photorealistic variant, specialized for natural detail and high-end realism

Each version will share the same v2 dataset, but be optimized for different styles, workflows, and artistic intentions.

If you like the model, please consider:

❤️ Leaving a like – it helps visibility and growth
🖼️ Posting your generations – the more the better!
🐞 Bug tracking & edge cases – your feedback helps identify generation quirks and improve the next version
🎯 A new bounty will likely be launched soon to showcase the best this version can do — stay tuned!

💬 Leave your feedback, share your creations and help shape the future of this project.
Every generation matters.

— Devildonia

🔥 Illustrious from HaDeS v1.0.16 – The Bridge to V2

🧠 Based on: Illustrious from HaDeS v1.0.12
📌 Status: Stable Release
🔧 Type: SFW + NSFW (explicit capable if prompted)
📩 For commercial collaborations → [email protected]

✅ What’s New in v1.0.16?

This is not just a patch — it's a solid upgrade before the launch of the upcoming v2.0 series.
Version 1.0.16 addresses several issues from v1.0.12 and pushes the model further in quality, adherence, and flexibility.

🧬 Key Improvements:

• ⚡ Better image quality, especially at high resolutions

• 🎯 Stronger prompt adherence — simple prompts work on the first try; complex ones succeed quickly if well-structured

• 👁️ Eye rendering greatly improved at HD+ resolutions

• ✋ Hands more accurate and stable

• 🌄 Improved background generation

• 🎨 More flexible rendering styles — easier to shift aesthetics via prompt

• 💡 Improved lighting and expressiveness (not as dramatic as v2, but a clear step up)

• 😮‍💨 Character expressions are livelier and more emotive

• 🔞 NSFW rendering is more robust than in v1.0.12

🧪 Prompt Simplicity – A Clear Upgrade

All showcase images were generated using only:

masterpiece, best quality, (+ your prompt)

🧼 No negative prompt was needed in almost any example — only used to avoid specific unwanted repetitions in rare cases.
🔍 Of course, you can still add negative prompts to further clean, refine, or enhance the output — they can definitely help create more spectacular results.
🧠 Likewise, while "masterpiece, best quality," was enough to demonstrate the base strength of the model, you’re free to enrich your prompt with detailed keywords and style modifiers to reach visual results on par with more reasoning-heavy models.

This demonstrates:

✅ Clean baseline generation
✅ Excellent default understanding
✅ High prompt adherence, even on first tries

⚙️ Recommended Generation Settings

• 📏 Native resolution: 1024x1408 — faster and highly stable

• 🖼️ All examples: 1152x1536 + Hires.Fix x2.5 → 2820x3840

• 🌀 Sampler: Euler a (Simple) — most stable in testing

• 🔢 Steps:

• 🎚️ CFG (Classifier-Free Guidance): 5 to 7 — recommended range for best balance between detail and coherence

  • 20–25 → for simple prompts

  • 30 → for normal complexity

  • 65 → preferred for ultra-high quality / complex prompts (especially in high res)

The model also performs well with other samplers, but Euler a proved consistently the most reliable.

💬 Community Support = Better V2

If you like the model, please consider:

❤️ Leaving a like – it helps visibility and growth
🖼️ Posting your generations – the more the better!
🐞 Bug tracking & edge cases – your feedback helps identify generation quirks and improve the next version
🎯 A new bounty will likely be launched soon to showcase the best this version can do — stay tuned!

🗣️ Why this version matters

v1.0.16 is the final refinement of the original base, crafted to be stable, clean, and versatile — perfect for users who want a reliable and expressive checkpoint while waiting for the more experimental and stylistic leap of v2.0.

💬 Leave your feedback, tag your creations with #IllustriousFromHaDeS, and help shape the future of this project.
Every generation matters.

— Devildonia

🔥iLLustriouS from HaDeS v1.0.12

🎨 Expressive, Aesthetic, Uncompromising

The next-generation model for modern anime aesthetics, emotional range, and razor-sharp consistency.

Built upon the foundation of several carefully selected Illustrious-based checkpoints and refined with custom LoRAs for advanced lighting behavior, iLLustriouS from HaDeS v1.0.12 is a new breed of visual synthesis. It combines cinematic elegance, vivid expressiveness, and a bold artistic identity — delivering images that feel alive, striking, and polished.

Whether you're crafting serene portraits, evocative fantasy heroines, or highly stylized NSFW concepts, this model adapts beautifully to your prompt. Its fidelity in reproducing modern anime styles with high precision and harmony makes it ideal for both SFW and adult-themed creators.

Robust multi-character control: Whether for action scenes, emotional storytelling, or more intimate settings, Illustrious from HaDeS handles multi-subject prompts with strong coherence and interaction.

⚠️ Note: This is a first public release. If you enjoy the model, please consider sharing your results on this page. Feedback, bug reports, and visual samples from the community are highly appreciated — they help guide future improvements.

⚡ Key Features:

• 🎯 Exceptional Adherence – Prompts yield consistent, reliable results with fine control over details

• 😳 Expressive Faces – Emotions are rich and believable — from subtle glances to dramatic tension

• 🧬 Modern Aesthetic – Clean, sharp anime style with next-gen character design appeal

• 🌟 Advanced Lighting – Enhanced with lighting-focused LoRA support for depth, glow, and contrast

• 🔥 NSFW-Ready – The model handles explicit adult content with natural anatomy and bold composition — but will only generate such content when explicitly prompted. Otherwise, results remain fully SFW.

🛠️ Recommended Settings:

From internal testing, we recommend the following configuration for optimal quality and consistency:

• Sampling method: Euler a

• Sampling steps: 20–25

• CFG Scale: 5

• Clip skip: 2

• Resolution: 1024 × 1408

• Hires. fix: ✅ Enabled

• Upscaler: 4x_foolhardy_Remacri

✨ Sample Prompt Example:

(masterpiece), (ultra-detailed), (very aesthetic), (absurdres), (high resolution), <your prompt here>

✨ Negative Prompt Example:

lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality,signature, watermark,anti human body structure ,username, blurry, sketch, bad, fewer, extra, missing, worst quality, bad quality, watermark, unfinished, displeasing, chromatic aberration, artistic error, scan, abstract,big head, artist name, bad eyes, bright pupils, white pupils, ugly, bad face,abnormal toe appearance,monotone background,grayscale,review, extra fingers,extra toes,missing toes,abnormal number of toes,nail polish,trousers ,stockings,multiple people,extra feet,more than five toes,sock

🔓 License:

This model is free to download and use with no limitations, as long as it is for personal, non-commercial use only. Redistribution of the model files for commercial purposes is not permitted. If you create derivative work using this model, please respect the license and credit accordingly.

💬 Final Words from Devildonia:

This is a model I’ve poured a lot of hope and intention into. iLLustriouS from HaDeS is not just about clean lines and flashy colors — it's about giving you a tool that truly listens to your prompt. Whether you’re building a gentle visual poem or pushing limits with NSFW realism, this model will take you there — with elegance, power, and control.

📧 Business Contact:

For commercial inquiries, collaborations or licensing, please contact: [email protected]

If you enjoy it, please leave a ❤️ and follow me here on Civitai. Reviews, feedback, and seeing your creations posted here truly mean the world to me — and they help shape what's coming next.

~ Devildonia