What the Flux just happened
Suuup - this is SinkIn Newsletter, a 5 minutes read made at sinkin.ai to cover the most interesting stuff in the Image AI world.
Flux, the new image AI model from Black Forest Labs, has gained significant attention in the AI community. Developed by former Stability AI engineers who originally worked on Stable Diffusion, Flux.1 is an open-source text-to-image model that has been praised for its high-quality outputs. It has been compared favorably against industry giants like Midjourney and DALL-E, particularly in terms of realism, prompt adherence, and diversity of generated images. The model is quickly being integrated into various platforms and is seen as a strong competitor in the generative AI space. You can try it out at runflux.ai.
An image generated by Flux
It’s not been long before people figured out how to train LoRA for Flux. AI headshots, as one of the most popular applications of generative AI, can be achieved with Flux LoRA too. Results have shown that it can reach better resemblance and quality than previous Stable Diffusion methods (e.g. Dreambooth). We do have an AI headshot service suitup.ai and has recently updated its technology with Flux LoRA. Try it out to see how well Flux does the job.
Headshots generated with Flux LoRa
Not only you can train Flux LoRA for headshots, you can also train it to produce product images. It captures the details of the product accurately and produces far better results than previous Stable Diffusion methods. Tap the image below to check out the example of a yellow sofa. You can check out this example of an LV bag too.
Generating product image with Flux LoRA
The community has been moving fast to port the Stable Diffusion technologies to flux. XLabs has released Canny, Depth and HED models. Each ControlNet is trained on 1024x1024 resolution and works for 1024x1024 resolution. They are already on version 3 which are more realistic and they can be used directly in ComfyUI. An Openpose model has just been made available too.
Flux Finetunes
The community has also been pushing out Flux finetune checkpoints. Some noticeable ones include:
Flux.1 BNB NF4: Flux optimized for low VRAM GPUs.
Flux Unchained: a 4 step, SFW/NSFW capable tune of the Flux Schnell 4 step distilled model.
Flux_Realistic: aims to create more diverse and realistic faces.
Meme of the Week
That’s it for today, hope you’re as pumped as we are about Flux!
What'd you think of today's edition? |