*some previews may be unavailable here due to auto-censorship. To see the full range of model capabilities (including NSFW), view the complete gallery <a target="_blank" rel="ugc" href="https://t.me/axiom_team_ch">on Telegram</a>Axiom is a complex merge designed to solve the common trade-offs in Illustrious/Noob-AI models. While base models are diverse but unstable, and specialized models offer stability at the cost of a "hard-coded" style, Axiom bridges this gap. It eliminates the anime-only bias of Illustrious and the loss of character authenticity found in realistic models.<h2 id="key-features:">Key Features:</h2><ul><li>Perfect Balance: A versatile foundation that maintains stability without sacrificing diversity. It covers anime, illustrations, 3D, and semi-realism within a single, pretty balanced model.</li><li>Advanced-User Ready: Optimized for broad prompt interpretation - delivering authentic ethnic features and complex textures that standard anime models often simplify.</li><li>Superior Control: Extremely high compatibility with ControlNet and IP-Adapter, allowing for precise scene, pose, and style management.</li><li>Universal LoRA Support: High compatibility with both Illustrious and Noob-AI LoRAs.</li></ul><hr /><h2 id="technical-guide">Technical Guide</h2>Recommended Settings:<ul><li>Generation: Euler a Polyexponential | 30 Steps | CFG 5.5~ | Default Illistrious/Noob-AI Resolutions</li><li>Upscaling: Hires/SD Upscale x1.5–2.0 | Denoising ≤ 0.46 | Steps ≤ 23</li><li>Upscalers: ESRGAN, UniScale Balanced, or NKMD SIax (sharp upscalers recommended)</li><li>Software: WebUI - ReForge ("Old" branch recommended for bug-free ControlNet support)</li></ul>Prompting Structure:<ul><li>Positive: <code>[subject/scene], masterpiece, very awa, best quality, absurdres, newest</code></li><li>Negative: <code>lowres, (worst quality, bad quality:1.2), text, watermark, old, [negative]</code></li></ul>ControlNet &amp; IP-Adapter:<ul><li>IP-A: Weight 0.66–0.9 | Start 0.11–0.33 | End 0.66–1.0</li><li>General Tips: Avoid 1.0 weight on all parameters. Lineart Realistic is strongly recommended for high-fidelity textures.</li></ul><hr /><h3 id="links-and-notes:">Links &amp; Notes:</h3><ul><li>Links: <a target="_blank" rel="ugc" href="https://huggingface.co/axiom-team">HuggingFace</a><a target="_blank" rel="ugc" href="https://huggingface.co/axiom-team"> </a>| <a target="_blank" rel="ugc" href="https://t.me/axiom_team_ch">Telegram </a>| <a rel="ugc" href="https://tensor.art/u/968104586772539491">TensorArt </a></li></ul><ul><li>Notes:<ul><li>Some previews include post-processing (ADetailer/External edits) to demonstrate maximum potential for advanced workflows.</li><li>Metadata is intentionally stripped from previews to prevent simple copy-pasting. As this model is built for advanced users, the goal is to encourage unique creations rather than replication.</li><li>For support, discussions, or any questions, feel free to join the Telegram channel.</li></ul></li></ul><hr /><h3 id="license-and-usage-terms">License &amp; Usage Terms</h3>By downloading or using Axiom, you agree to the following:<ul><li>Personal/Commercial Renders: Allowed for individuals (selling images is okay).</li><li>Corporate/Entity Use: Strictly prohibited without explicit permission. Includes any platform or service where the model is hosted, regardless of their own Terms of Service.</li><li>No Hosting: Uploading or hosting on any generation services/cloud platforms is strictly prohibited.</li><li>No Derivatives: Using Axiom as a base for merging, training, or selling new models/selling the model itself is prohibited.</li></ul>

Axiom

Optimized Hybrid (FP8 UNet / FP16 CLIP) Key Benefits (compared to FP16):<ul><li>VRAM Optimization: Lower memory footprint (depends on hardware and software configuration).</li><li>Storage Optimization: ~37% smaller file size (4.07GB vs 6.46GB).</li><li>Performance: Native hardware acceleration for RTX 40xx/50xx.</li><li>Visual Fidelity: High accuracy with subtle variations in fine details.</li></ul>

Files with SHA256: afef2b636e476904793f5b4b97726dc2e1d8964c68ce6d5055ad4b4d603fdc4b