Yo! `Wan2.1-I2V-14B-480P` LoRAs are now supported in Diffusers. Go, fire it! In the frame, we have the popular Squish LoRA.

Hackathon today with Lightricks, sponsored by @FAL, Hugging Face and Google With a Modular Diffusers/Mellon sneak peak by apolinario 🌐!

It's amazing that this old SD 1.5 fine-tuned model with images from Studio Ghibli is trending on the Hugging Face Hub thanks to OpenAI. huggingface.co/nitrosocke/Ghi…

Before 2020, most of the AI field was open and collaborative. For me, that was the key factor that accelerated scientific progress and made the impossible possible—just look at the “T” in ChatGPT, which comes from the Transformer architecture openly shared by Google. Then came

You have had your GPT4-o Ghibli moments. Now switch your attention to the latest Diffusers release - 0.33.0 🔥 Bringing you a bunch of new image & video gen models, a wide suite of memory optimizations w/ caching, & `torch.compile()` support when hotswapping LoRAs. Thread ⬇️

please, stop comparing HiDream quantized with Flux or other models, quantization really hurts this model. Soon I'll post a guide on how to run it without quantization on 24GB and 16GB GPUs with almost the same speed as with quantization using diffusers.

HiDream Image LoRA fine-tuning just dropped and it's time to make it yours 🧨 HiDream's sota capabilities (and mit license) bring a lot of potential to explore with fine-tunes 🔥 and of course I had to train a yarn art LoRA 🧶 - more upgrades and features soon! - code, weights

Do you know what's as exciting as Veo 3? Running Wan 2.1 in just 4-8 steps 🏎️💨 I built a demo for Wan 2.1 + CausVid LoRA, which allows high quality video generation in just 4 steps 🤏

We present HeadHunter, a framework for principled analysis of perturbed attention guidance 🤖 Consequently, it enables deeply fine-grained control over the generation quality & visual attributes. Join in 🧵 for insights and "guidance". 1/12

Boy, we shipped, and we shipped hard 🧨 From new SoTA open models to improved support for torch.compile to features, inspiring more accessibility -- this Diffusers release is a blast! What's your favorite? Check out the notes for more!

With Flux Kontext, if you want a more precise shape and positioning you can actually do a bad drawing of what you want. Prompt: change the black drawing of a hat to a real hat.

change the drawing of a hat to a real baseball cap and with the same exact position and color as the drawing

Created a new guide about image compositing with Diffusers and SDXL, for those of you who don't know how to do it and need an open-source and commercially usable solution. huggingface.co/blog/OzzyGT/di…

how is this for prompt adherence? Qwen-Image with 24GB GPU and Diffusers.

so it turns out we don't need to do anything special for the text encoder, this is with both the transformer and text encoder using bitsandbytes with 4-bit quantization, using under 16GB of VRAM and in ~1m40s with a 3090

if you want the code for running Qwen-Image with 24GB and 16GB GPUs using diffusers, you can use the code in this PR while waiting for a merge: huggingface.co/Qwen/Qwen-Imag…

Qwen-Image-Lightning 8-Steps runs in 22s and using less than 16GB with a 3090. You can find the models and the code to test it here: huggingface.co/OzzyGT/qwen-im…