Перейти к содержимому

Qwen Image 2.1: New Viggle Turbo Update vs Pruna vs Alibaba vs 40 Steps

ComfyUI Workflow Blog

0:00 / 0:00

Qwen Image 2.1: New Viggle Turbo Update vs Pruna vs Alibaba vs 40 Steps

3 756 просмотров · 8 часов назад
ComfyUI Workflow Blog
22,6 тыс. подписчиков
3 756 просмотров · 8 часов назад
I tested the new speed LoRAs for Qwen Image 2.1 against the normal 40 steps in ComfyUI, with the same prompts on my RTX 5090. One is the fastest, one is the closest to 40 steps, and one changed the face in a simple edit. What I compared: 40 steps (no speed LoRA) Viggle Turbo, new version (6 steps) Pruna (8 steps) Alibaba Fun Acc (4 steps) I skipped Pruna 5-step because I wanted to see how good the 8-step version is. What I tested: Portrait, text on a cafe sign and menu, hands, a street scene, a watch, easy and hard character sheets, storyboards, 2K resolution, a prompt written by Qwen 3.5, and image editing (colour change, object removal, text replacement, a rainy night scene, and an outfit change). Speed on my RTX 5090 (portrait, 1024 x 1024): 40 steps: about 13.9 seconds Viggle Turbo: about 3 seconds Pruna 8-step: about 8 seconds Alibaba Fun Acc: 1.82 seconds Speed for a simple edit (T-shirt colour change): 40 steps: 16 seconds Viggle Turbo: 3.8 seconds Pruna 8-step: 4.7 seconds Alibaba Fun Acc: 2.9 seconds Before you start: All of them work at 1K. Viggle supports 1K and 2K, and up to 1536 for editing. Pruna does not support 2K, so it may not work. None of them support 4K. For 4K, use an upscaler. Viggle needs its own node (the unmerged node). Save viggle_turbo.py in your ComfyUI custom_nodes folder. Alibaba Fun Acc is not supported natively in ComfyUI yet, so I made my own node to run it. Outfit tip from the video: put the whole outfit in one image and use one Load Image node. Links: My workflow:   / qwen-image-2-1-170642530   My Alibaba Fun Acc node: https://github.com/aistudynow/ComfyUI... Download viggle_turbo.py (save it in ComfyUI/custom_nodes): https://huggingface.co/Viggle/Qwen-Im... Viggle Turbo LoRA and workflows: https://huggingface.co/Viggle/Qwen-Im... Pruna-Qwen-Image-2.1: https://huggingface.co/PrunaAI/Pruna-... Alibaba Fun Acc: https://huggingface.co/alibaba-pai/Qw... License: All of these LoRAs use the Qwen Research License. Check the license before you use them for paid work. These are my results on my PC, so try them on your own images too. Which one will you use? Tell me in the comments. #QwenImage #ComfyUI #AIImageGeneration