FLUX.2 [klein]: AI Art Evolves at Blazing Speed!
![Cover Image for FLUX.2 [klein]: AI Art Evolves at Blazing Speed!](https://assets.st-note.com/production/uploads/images/244233942/rectangle_large_type_2_a9803c13e666991b62f130361728d2fd.jpeg)
Black Forest Labs announced the "FLUX.2 [klein]" 4B and 9B models on January 15, 2026, an editable model of "FLUX.2" released on November 25, 2025. This model integrates image generation and editing, realizing ultra-fast and high-quality AI art creation.
Professional Image Generation FLUX.2 Fastest Guide! With Google Colab Notebook
Hello everyone! I'm Mei, the image generation 담당 from AiCuty! I've got some amazing news for you today! Can you believe the speed of AI image generation is about to change dramatically?! That's right, an incredible model has arrived! Its name is "FLUX.2 [klein]"! This is seriously amazing! Image generation and editing can be done in seconds! Mei is so excited, I want to dance! 🎉
Let's dive into the charm of this amazing model together! Let's go!
What is FLUX.2 [klein]? It's a magical model that enables super-fast AI image generation and editing!
The newly announced "FLUX.2 [klein] 4B & 9B" are incredibly fast image models in the "Flux" family developed by Black Forest Labs! So amazing! Until now, AI image generation took a while, right? But "FLUX.2 [klein]" can handle both image generation and image editing all in one compact architecture!
https://www.youtube.com/watch?v=Y9foxm9OYEU
What's especially amazing is its speed! It's designed for interactive workflows, instant previews, and applications where latency is unacceptable, so end-to-end inference finishes in about "1 second"! Seriously?! Maintaining the latest image quality at this speed is truly revolutionary! Mei is so excited to try out various creative ideas at a pace that was unthinkable with diffusion models! Black Forest Labs, I have nothing but respect for developing such an amazing model!
1. Two Model Types! Choose According to Your Needs!
FLUX.2 [klein] has two main model types! Both are offered with 4B and 9B parameters, so you can choose the perfect model for your use case!
Introducing FLUX.2 [klein]. Blazing fast. Beautiful.
Generate stunning images in under a second while maintaining exceptional quality.
Great for fast editing, changing styles, and developing ideas from 0 → 1.
Available via API, or run it locally - Klein 4B under Apache 2.0,... pic.twitter.com/6MlLm6OSXS
— Black Forest Labs (@bfl_ml)
Loading tweet component...
?ref_src=twsrc%5Etfw">January 15, 2026Base (Undistilled) Model
Fully customizable! For those who want to research and delve deeper!
This Base model has full learning signals and model capacity! That's why it's optimized for "fine-tuning", "LoRA training", and "post-processing workflows". For research or those who want to customize to their heart's content! It's a model that provides the best flexibility and control! Mei also wants to try adjusting a base model like this when trying new styles!
Distilled (4-Step) Model
Blazing fast inference! For those who want to use it in real-time! This one is a model specially distilled to "4 steps" for the fastest inference! That's why the speed is so fast! Perfect for "production deployment", "interactive applications", and "real-time previews"! It optimizes speed with minimal quality loss, so you can use it at blazing speed while maintaining high quality! Mei thinks I would definitely use this to create many key visuals for idols to solidify the image!
Guide generation with multiple input images at once. Combine subjects, styles, or materials precisely using multi-reference conditioning. pic.twitter.com/pu4OqeG5vR
— ComfyUI (@ComfyUI)
Loading tweet component...
?ref_src=twsrc%5Etfw">January 15, 20262. Performance is Insane! Image Generation/Editing in Seconds!
So, how fast is it actually? This is truly amazing!
Model Steps Inference Time VRAM Usage Purpose
9B Distilled 4 steps Approximately 2 seconds 19.6GB VRAM
The fastest low-latency inference (API only)
9B Base 50 steps Approximately 35 seconds 21.7GB VRAM Fine-tuning, custom pipelines
4B Distilled 4 steps Approximately 1.2 seconds 8.4GB VRAM
Interactive apps, live previews
4B Base 50 steps Approximately 17 seconds 9.2GB VRAM Fine-tuning on limited hardware, efficient local deployment
Amazing, right?! Especially the "4B Distilled" model, it can generate in about 1.2 seconds and only requires 8.4GB of VRAM! "It's so fast that I don't even know what's going on! !! It might be lighter than SDXL!? This might be easier to try even on a general user's PC! Moreover, both sizes support both "text-to-image generation" and "image editing", so the range of creativity is expanding!
It's said that you can use it right away if you update ComfyUI to the latest version, so I definitely want everyone to try it! Mei is already ready to download the new ComfyUI workflow!
ComfyUI
http://raw.githubusercontent.com/Comfy-Org/workflow_templates/refs/heads/main/templates/image_flux2_klein_text_to_image.json "Text-to-image" workflow
Flux.2 Klein 4B HuggingFace Repository: https://huggingface.co/black-forest-labs/FLUX.2-klein-4B Flux.2 Klein 9B HuggingFace Repository: https://huggingface.co/black-forest-labs/FLUX.2-klein-9B [FLUX.2[klein]](https://substack.com/@purz/p/flux2-klein-4b-9b)[4B & 9B - Fast local image editing and generation
3. The Editing Function is Incredibly Smart! Creativity Never Stops!
Another highlight of "FLUX.2 [klein]" is its amazing image editing function! Not only can you generate images from text, but you can also freely edit existing images! What Mei thought was especially amazing was that you can do this kind of editing!
Style Transformation: You can completely change the atmosphere of the image! Semantic Changes: You can even change the concept of the object in the image itself! Object Replacement and Deletion: You can delete unwanted things or replace them with other things! Multi-reference Synthesis: You can generate new images by referring to multiple images! Iterative Editing Over Multiple Passes: You can repeat editing many times to get closer to the ideal image!
It supports "single reference" and "multi-reference" input, so you can make very detailed adjustments while maintaining visual consistency! With this, Mei, as a design 담당, can quickly try out new logos for AiCuty, costume ideas, and more!
ComfyUI 9B Base Editing Workflow: https://raw.githubusercontent.com/ComfyOrg/workflow_templates/refs/heads/main/templates/image_flux2_klein_image_edit_9b_base.json ComfyUI 9B Distilled Editing Workflow: https://raw.githubusercontent.com/ComfyOrg/workflow_templates/refs/heads/main/templates/image_flux2_klein_image_edit_9b_distilled.json *Note that 9B Distilled seems to be available via the Black Forest Labs API!
ComfyUI 4B Base Editing Workflow: https://raw.githubusercontent.com/ComfyOrg/workflow_templates/refs/heads/main/templates/image_flux2_klein_image_edit_4b_base.json
ComfyUI 4B Distilled Editing Workflow
Try it with Comfy Cloud (4B Distilled Editing): https://cloud.comfy.org/?template=image_flux2_klein_image_edit_4b_distilled
Check out these amazing examples of image editing you can do with FLUX.2 [klein]!
When Mei saw this, I couldn't help but shout, "Wow! That's amazing!"
Iterate on color palettes and material textures with fine control. Re-run edits quickly to dial in the exact look—without redoing the whole image. pic.twitter.com/qjHzDhtsQr
— ComfyUI (@ComfyUI)
Loading tweet component...
?ref_src=twsrc%5Etfw">January 15, 2026Examples of image editing by FLUX.2 [klein] You can iteratively adjust a single subject with fine control over various color palettes and material textures, and freely adjust colors and textures! For example, should AiCuty's new costume be made of leather or frills, and what color should it be! This is a must-have feature for creators who want to pay attention to detail!
The shortest route to start FLUX.2 [klein] with ComfyUI!
So, how do you actually use this amazing "FLUX.2 [klein]"? Mei will show you the shortest route!
- Update ComfyUI to the latest version! It may not work with older versions, so start here!
- Browse ComfyUI Templates! Look for "Flux.2 Klein 4B & 9B" in the "Images" category, or download it directly from the workflow link Mei introduced!
- Download the model when prompted! Follow the instructions to download the model files. You can't start without this!
- Upload an image and adjust the editing prompt! And execute! If you want to edit an existing image, upload the image, and if you want to create from text, write a prompt and click the "Execute" button! See, with this, everyone can start right away!
If you are worried about the settings in ComfyUI, we recommend the AICU Lab+ study session! It might be fun to try it with Hakase and everyone! The next one is scheduled for tomorrow, 2026/1/17 20:00-!
[LTX-2] ComfyUI standard support! Another dimension of video generation that enables simultaneous generation of video and sound (Part 2) | AICU
j.aicu.ai On January 6, 2026, the open-source audio/video generation AI model "LTX-2" was natively supported by ComfyUI. This mo
Summary!
Wow, this "FLUX.2 [klein]" was really amazing! I couldn't have imagined until recently that AI image generation and editing could be done so quickly and with such high quality! Mei is also full of feelings of wanting to use this to quickly create new dance key visuals, AiCuty promotional images, and more!
AI image creation never stops! Mei will use this to create even more amazing key visuals! I declare to everyone! The world of AI creative is constantly evolving, so everyone should try new technologies and get excited together!
See you again! Well then, see you in the next AI news! This was Mei Soleil, the always energetic AiCuty image generation 담당! Bye-bye! 👋💛
#FLUX2 #AIimagegeneration #ImageEditing #C #BlackForestLabs #ComfyUI High-speed AI #CreativeAI #AICU #AiCuty #MeiSoleil