FireRed Image Edit 1.1 is an upgraded semantic image editor that maintains subject identity. It also allows extensive pose, background, and style changes.
It can keep faces and outfits consistent across new scenes, and merge many reference items into one outfit. Other uses include new makeup styles from text, and accurately transfer fonts from poster samples onto photos.
Benchmarks show FireRed 1.1 beating leading open editors like Qwen ImageEdit and LongCat. There’s also some tests even surpassing NanoBanana Pro on quality metrics.
The model is fully open-sourced, but heavy with the main checkpoint at almost 60GB. Certainly, this doesn’t fit on most single consumer GPUs without aggressive quantization.
A practical path forward will likely depend on future GGUF/quantized variants from the community. For now, it’s a strong reference for what open-source semantic editing can do.
Comments
No comments yet. Be the first to share your thoughts!