HiDream-E1.1 is the new leading open weights image editing model, surpassing FLUX.1 Kontext [dev] in the Artificial Analysis Image Editing Arena!
HiDream has recently released HiDream-E1.1, an updated version of the HiDream-E1-Full model, which retains its 17B parameter architecture, while introducing support for direct commands and flexible aspect ratios.
The model outperforms the open weights FLUX.1 Kontext [dev] in our image editing leaderboard, though it falls short of leading proprietary image editing models such as GPT-4o and FLUX.1 Kontext [max].
Unlike FLUX.1 Kontext [dev]'s non-commercial license, E1.1 comes with a MIT license enabling unrestricted commercial use. Note that it uses Meta Llama 3.1 8B's text encoders, so also incorporates Meta's Llama 3 license terms.
Currently, the model is only publicly served through the
@vivago_ai app, with weights available on
@huggingface. We are monitoring inference providers and will update when API access becomes available.
See below for comparisons between HiDream-E1.1 and other leading models in our Image Editing Arena, or try it yourself! 🧵