Black Forest Labs 发布了 FLUX 3 Image Klein,一款具有 5.6B 参数的开源文本生成图像模型。1该模型支持原生 4K 生成、像素级图像编辑和最多 10 个参考图像的多参考生成功能。1在文本生成图像、多参考生成和图像编辑任务的内部评估中,该模型排名第一,性能超过了更大的开源模型和部分闭源模型。1
相比前代 FLUX.2 [klein] 9B 模型,FLUX 3 Image Klein 在文本生成图像任务中 Elo 值提升 91 分,图像编辑提升 70 分,同时参数量减少 38%。1该模型针对消费级 GPU 进行了优化,BF16 格式需 11.2GB 显存,FP8 格式需 5.6GB,INT8 格式可在 8GB VRAM 的 NVIDIA GeForce RTX GPU 上运行。1在 RTX 5090 上生成 1MP 图像耗时约 4 秒,在 RTX 4070 上耗时不足 9 秒。1
该模型采用 FLUX Kommunity License 许可证,年收入 500 万美元以下的个人和公司可免费商用。1通过 BFL API 使用,定价从 $0.009 每 1K 图像起。1
Black Forest Labs has unveiled FLUX 3 Image Klein, an open-source text-to-image generation model featuring 5.6 billion parameters 1. The model delivers native 4K generation capabilities, pixel-level image editing, and support for multi-reference generation using up to ten reference images 1.
In internal evaluations, FLUX 3 Image Klein ranked first across text-to-image generation, multi-reference generation, and image editing tasks 1. Compared to its predecessor FLUX.2 [klein] 9B, the new model achieves a 91 Elo improvement on text-to-image tasks and a 70 Elo gain on image editing while reducing parameters by 38 percent 1. The model has been optimized for consumer-grade GPUs, with memory requirements of 11.2GB in BF16 format, 5.6GB in FP8 format, and support for 8GB VRAM NVIDIA GeForce RTX devices in INT8 format 1. Generation speed reaches approximately four seconds for 1-megapixel images on an NVIDIA GeForce RTX 5090 and under nine seconds on an RTX 4070 1.
Access through the BFL API begins at $0.009 per 1,000 images 1. Under the FLUX Kommunity License, individuals and companies with annual revenue below five million dollars may use the model commercially without charge 1.
评论
还没有评论,欢迎留下第一条。