<h1 id="flux.2-klein-4b-aio-or-sub-second-image-generation">🚀 FLUX.2 [klein] 4B AIO | Sub-Second Image Generation</h1>Ultra-Fast • 4-6 Steps • Text-to-Image + Image Editing • All-in-One • Apache 2.0<hr /><h2 id="what-is-flux.2-klein-4b-aio">✨ What is FLUX.2 [klein] 4B AIO?</h2>FLUX.2 [klein] 4B AIO is an All-in-One repackage of Black Forest Labs' newest compact image generation model. This version includes VAE, Text Encoder (Qwen3) and UNet in a single file – just load and go!"Klein" means "small" in German – but this model is anything but limited. It delivers exceptional performance in Text-to-Image, Image Editing and Multi-Reference Generation, typically reserved for much larger models.<hr /><h3 id="update">🔄 UPDATE</h3><h3 id="flux2-klein-4b-aio-nvfp4">⚡ Flux2-klein-4B-AIO-NVFP4</h3>Fast generation with blackwell in just a few seconds — even at 4 steps, and scales nicely with more steps 🚀🕒 Performance<ul><li>Prompt executed in 2.08 seconds</li><li>██████████████ 4 / 4 steps</li><li>~2.65 it/s</li></ul>✅ Extremely fast ✅ Stable ✅ Great quality for a distilled setup<hr /><h3 id="example-generation">🖼️ Example Generation</h3><edge-media url="bd48640d-b1f0-455b-aa0f-3550dd472001" type="image" filename="Flux2-Klein-4B-L_00019.png"></edge-media>Prompt:Anime, powerful anime illustration with vibrant dark fantasy colors, one adult woman inspired by Jouryuu, tall imposing presence, long hair flowing dramatically, intense anime eyes, wearing an ornate battle-inspired dress referencing official visuals, heavy fabric and strong silhouette, standing confidently in a ruined temple environment, low-angle camera enhancing dominance, dramatic backlighting with red and violet tones, strong shadows, intense cel-shading, bold anime lineart, intimidating yet elegant presence, correct anatomy, no text, no watermark.💡 This setup is optimized for speed, making it ideal for quick iterations, testing ideas, or just having fun generating without long wait times.Have fun generating — and as always, thanks for all the feedback and support 🙌✨<hr /><h2 id="available-versions">📦 Available Versions</h2><h3 id="fp8-aio-(~7.7-gb)-recommended-for-most-users">🟡 FP8-AIO (~7.7 GB) – Recommended for most users</h3><ul><li>Precision: FP8</li><li>UNet: FP8</li><li>Text Encoder: FP8</li><li>VAE: BF16</li><li>Best for: Most users, quick tests, everyday use, lowest VRAM</li></ul><h3 id="fp16-aio-(~15-gb)-for-older-gpus">🔵 FP16-AIO (~15 GB) – For older GPUs</h3><ul><li>Precision: FP16</li><li>UNet: FP16</li><li>Text Encoder: FP16</li><li>VAE: BF16</li><li>Best for: Older GPUs (GTX 10xx, RTX 20xx), broadest compatibility</li></ul><h3 id="bf16-aio-(~15-gb)-maximum-quality">🟢 BF16-AIO (~15 GB) – Maximum quality</h3><ul><li>Precision: BF16</li><li>UNet: BF16</li><li>Text Encoder: BF16</li><li>VAE: BF16</li><li>Best for: RTX 30xx/40xx/50xx, professional/commercial work</li></ul><h3 id="nvfp4-aio-(~5.7-gb)-maximum-speed">🔴 NVFP4-AIO (~5.7 GB) – Maximum speed ⚡</h3><ul><li>Precision: NVFP4</li><li>UNet: NVFP4</li><li>Text Encoder: FP4-mix</li><li>VAE: BF16</li><li>Best for: RTX 50xx only, ultra-fast generations, Blackwell GPUs, low VRAM + extreme performance</li></ul><hr /><h2 id="key-features">🎯 Key Features</h2><ul><li>⚡ 4-6 Step Generation – Sub-second inference on modern hardware</li><li>📦 All-in-One – No separate VAE/Text Encoder download needed</li><li>🎨 Unified Architecture – T2I, I2I Editing &amp; Multi-Reference in one model</li><li>📐 1024×1024 native – Optimized for this resolution</li><li>💾 Low VRAM – Runs on consumer GPUs with ease</li><li>📜 Apache 2.0 – Fully open for commercial use!</li><li>🔧 LoRA-compatible – Base version ideal for fine-tuning</li></ul><hr /><h2 id="recommended-settings">⚙️ Recommended Settings</h2><ul><li>Steps: 4-6 (step-distilled, more steps ≠ better)</li><li>CFG: 1.0 ⚠️ CRITICAL!</li><li>Sampler: euler</li><li>Scheduler: simple (or "normal")</li><li>Resolution: 1024×1024 (native)</li></ul><h3 id="critical:-cfg-must-be-1.0!">⚠️ CRITICAL: CFG Must Be 1.0!</h3>This is a distilled model optimized for CFG 1.0. Higher CFG values will produce worse results!<pre><code>✅ CFG 1.0 = Correct
❌ CFG 3.5+ = Wrong, will look bad
</code></pre><h3 id="additional-notes">Additional Notes</h3><ul><li>4-6 Steps are optimal! The model was step-distilled for fast inference</li><li>No negative prompts needed – works but not required</li><li>Natural language prompts – Just describe what you want to see</li></ul><hr /><h2 id="installation-(comfyui)">📥 Installation (ComfyUI)</h2><h3 id="quick-start">Quick Start</h3><ol><li>Download your preferred version (FP8/FP16/BF16)</li><li>Place in <code>ComfyUI/models/checkpoints/</code></li><li>Load with the "Load Checkpoint" node</li><li>Generate!</li></ol><h3 id="folder-structure">Folder Structure</h3><pre><code>ComfyUI/
└── models/
 └── checkpoints/
 └── flux-2-klein-4b-bf16-aio.safetensors (or fp16/fp8)
</code></pre><hr /><h2 id="example-prompts">🎨 Example Prompts</h2><h3 id="photorealistic">Photorealistic</h3><pre><code>A professional photograph of a barista making latte art in a cozy 
coffee shop, morning light streaming through windows, shallow depth 
of field, shot on Sony A7III
</code></pre><h3 id="digital-art">Digital Art</h3><pre><code>A majestic dragon perched on a crystal mountain peak, aurora borealis 
in the background, fantasy digital painting, highly detailed scales, 
dramatic lighting
</code></pre><h3 id="product-photography">Product Photography</h3><pre><code>Minimalist product photo of a luxury perfume bottle on white marble, 
studio lighting, reflection, commercial photography
</code></pre><hr /><h2 id="capabilities">💻 Capabilities</h2><h3 id="what-flux.2-klein-4b-can-do:">✅ What FLUX.2 [klein] 4B can do:</h3><ul><li>Text-to-Image (T2I) – High-quality image generation from text</li><li>Image-to-Image (I2I) – Single-reference editing</li><li>Multi-Reference – Multiple input images for controlled transformations</li><li>Text Rendering – Improved text rendering in images</li><li>Photorealistic – Professional photo quality</li><li>Artistic Styles – Diverse artistic styles</li></ul><h3 id="limitations:">⚠️ Limitations:</h3><ul><li>Optimized for 1024×1024 (other resolutions possible but not optimal)</li><li>4B model – less detail than larger models for complex scenes</li><li>Distilled version – less output diversity than base models</li></ul><hr /><h2 id="technical-details">🔧 Technical Details</h2><ul><li>Parameters: 4 Billion</li><li>Architecture: Rectified Flow Transformer</li><li>Text Encoder: Qwen3-based</li><li>Inference Steps: 4-6 (step-distilled)</li><li>Native Resolution: 1024×1024</li><li>Precision: BF16 / FP16 / FP8</li><li>License: Apache 2.0</li></ul><hr /><h2 id="comparison:-4b-vs-9b">🆚 Comparison: 4B vs 9B</h2><h3 id="flux.2-klein-4b">FLUX.2 [klein] 4B</h3><ul><li>Parameters: 4B</li><li>VRAM: ~8-13 GB</li><li>GPU: RTX 3090/4070+</li><li>Quality: Very Good</li><li>License: Apache 2.0 ✅</li><li>Commercial Use: Yes!</li></ul><h3 id="flux.2-klein-9b">FLUX.2 [klein] 9B</h3><ul><li>Parameters: 9B</li><li>VRAM: ~29 GB</li><li>GPU: RTX 4090+</li><li>Quality: Excellent</li><li>License: Non-Commercial ❌</li><li>Commercial Use: No</li></ul>→ 4B is perfect for: Consumer hardware, commercial projects, fast iterations<hr /><h2 id="faq">❓ FAQ</h2>Q: Do I need separate VAE/Text Encoder files?No! AIO = All-in-One. Everything is included in a single file.Q: Can I use this for commercial projects?Yes! The 4B version is licensed under Apache 2.0.Q: Why only 4-6 steps?The model was step-distilled. More steps won't improve quality.Q: Why must CFG be 1.0?This is a distilled model optimized for CFG 1.0. Higher values will degrade output quality.Q: FP8 vs BF16 – What's the difference?FP8 is smaller and faster, BF16 has slightly better quality. For most applications FP8 is sufficient.Q: Does this work with LoRAs?Yes! Especially the Base version (non-distilled) is ideal for LoRA training.Q: What's the difference to the 9B version?9B has better quality but is non-commercial only. 4B is Apache 2.0!<hr /><h2 id="troubleshooting">🐛 Troubleshooting</h2><h3 id="images-look-&quot;washed-out&quot;-or-oversaturated">Images look "washed out" or oversaturated</h3><ul><li>Check CFG – must be 1.0 for distilled model!</li><li>Use 4-6 steps</li></ul><h3 id="poor-text-rendering">Poor text rendering</h3><ul><li>Be more specific in your prompt</li><li>Use simple, short text</li><li>Place text requirements at the beginning of the prompt</li></ul><h3 id="colors-look-off">Colors look off</h3><ul><li>Try BF16 version instead of FP8</li><li>Ensure your monitor is properly calibrated</li></ul><hr /><h2 id="credits">🙏 Credits</h2>Original Model: Black Forest Labs Architecture: Rectified Flow Transformer Text Encoder: Qwen3 AIO Repackage: SeeSee21Official Links:<ul><li>🔗 <a target="_blank" rel="ugc" href="https://bfl.ai/blog/flux2-klein-towards-interactive-visual-intelligence">Black Forest Labs Blog</a></li><li>🔗 <a target="_blank" rel="ugc" href="https://github.com/black-forest-labs/flux2">GitHub Repository</a></li><li>🔗 <a target="_blank" rel="ugc" href="https://huggingface.co/collections/black-forest-labs/flux2">HuggingFace Collection</a></li><li>🔗 <a target="_blank" rel="ugc" href="https://huggingface.co/black-forest-labs/FLUX.2-klein-4B">Original 4B Model</a></li></ul><hr /><h2 id="changelog">📋 Changelog</h2>v1.1 – January 2026 ⚡<ul><li>🆕 Added NVFP4 AIO variant (RTX 50xx / Blackwell)</li><li>⚡ Ultra-fast inference (optimized for 4 steps)</li><li>🧠 Extremely low VRAM usage</li><li>🎯 Designed for maximum speed while keeping good image quality</li></ul>v1.0 (January 2026)<ul><li>Initial Release</li><li>BF16, FP16 and FP8 versions</li><li>All-in-One with VAE + Text Encoder + UNet</li></ul><hr />License: Apache 2.0 – Free for personal AND commercial use! 🎉<hr />The fastest open-source image generation model for ComfyUI! ⚡Download and start creating! 🚀

Files with SHA256: df9c25602d359c22159060220dcbea7f84ab2fc3638fd331ab51eee122f25eaf