Flux has recently garnered a lot of positive feedback from model enthusiasts, with its image quality surpassing that of SD3 and MJ. Many claim that Flux is the SD3 everyone has been hoping for, so I'm also quite curious about Flux's capabilities and will share the installation process with everyone.
Official Reference Image:
Flux Official Homepage: https://blackforestlabs.ai/
Artificial intelligence startup Black Forest Labs has been established and has released its first set of text-to-image AI models, FLUX.1. This German-based company, founded by researchers who developed Stable Diffusion, aims to create advanced generative AI for images and videos.
The models come in three versions, including Pro, Dev, and Schnell.
Black Forest's charts claim that their Pro and Dev models are the best image generators to date, while their weaker Schnell ranks between Midjourney v5 and Ideogram.
Descriptions of the three versions:
Pro version offers the best performance but is closed-source and requires API payment.
Dev version is open-source but not for commercial use, and requires at least 24GB of VRAM to run.
Schnell version is open-source and can be used commercially.
If you want to experience it online, there are also websites available:
Schnell: https://replicate.com/black-forest-labs/flux-schnell
Dev: https://replicate.com/black-forest-labs/flux-dev
Pro: https://replicate.com/black-forest-labs/flux-pro
Local Deployment
The latest version of ComfyUI is now compatible with the FLux models, so you just need to upgrade the kernel to the latest version to experience it.
So we first update ComfyUI.
If you are using the official ComfyUI integration package, click this update.
If you are using Qiuye's launcher, update the kernel as shown in the example below.
People with low VRAM may find it difficult to run locally; my local computer setup is decent, so I'm experiencing the Schnell version here.
Download the large model: https://huggingface.co/black-forest-labs/FLUX.1-schnell/tree/main
This link provides a 23.8GB large model and a 335MB Vae model, and we need to download both.
After downloading, place the large model in the: ComfyUI/models/unet/ folder.
The VAE should be placed in your ComfyUI/models/vae/ folder.
For those with average computer configurations, you can download the Flux-fp8 lite version, which is only 11.9GB: https://huggingface.co/Kijai/flux-fp8/blob/main/flux1-dev-fp8.safetensors
The Vae is still the universal ae.sft.
In addition to the large model and VAE, we also need to download the Clip model: https://huggingface.co/comfyanonymous/flux_text_encoders/tree/main
This is actually the SD3 Clip model; for good configurations, we choose fp16, and for average configurations, we choose fp8; clip_l is a must-download.
After downloading, we place it in the: ComfyUI\models\clip folder.
After installing the large model, VAE, and Clip as mentioned above, we can run our workflow. The official website provides a simple ComfyUI workflow: https://comfyanonymous.github.io/ComfyUI_examples/flux/
We save the image provided by the official website.
Then drag the image into ComfyUI; if you don't want to go through this hassle, you can also join our communication group to directly obtain the workflow.
Enter the workflow, and ensure that the model loads correctly, then click to join the queue and wait for generation.
Image Generation Effects
Prompt 1: a bottle with a rainbow galaxy inside it on top of a wooden table in the middle of a modern kitchen
Prompt 2: photograph of a black board in an old classroom. On the black board in chalk are the words “Lets make some really pretty stuff together” with a red chalk heart after the words. Sunlight is streaming in from the window
Prompt 3: filmic photo of a group of three women on a street downtown, they are holding their hands up to the camera
Prompt 4: macro photography of a miniature little village on top of a flower
FP8 version, Prompt 5: close up fujifilm photo of a man's eye and face
Prompt 6: beautiful anime artwork, a cute anime girl standing alone at night darkness wearing an oversized black raincoat with a small bag. she is holding a black umbrella, the umbrella has multicolored LED hidden inside that shine down on her, small rose petals flutter through the air around her, anime screencap style, red eyes, medium hair
Summary
Overall, the test results are quite good, with FLUX outperforming SD3.
Hand generation is more stable, with fewer cases of malformed hands and feet.
The distilled version produces images quickly, within 30 seconds. The Dev version takes longer but offers better image quality.
The model has strong compatibility: it works well for styles ranging from二次元, portraits, realism, to landscapes.
----------------------------------------------------------------------------------------
Zhanzhangsucai AI Tutorials is the AI drawing tutorial platform under Zhanzhangzhijia
A vast amount of AI free tutorials, continuously updating dry content
To learn more AI drawing tutorials in depth, please visit the Zhanzhangsucai AI Tutorials website:
