⚠️ WARNING: Python 3.12+ is NOT supported. Use Python 3.11.8–3.11.9 only.
This workflow is designed for multi-angle dataset production using Qwen Image Edit 2511 (Lightning).
It focuses on generating consistent subject representations across multiple camera angles, making it suitable for dataset creation, reference generation, and training preparation.
Every image has the according name and is ready for Lora-Training!
The workflow does not rely on Lightning step LoRAs and uses the Lightning base model directly.
Optional Multi-Angle LoRA support enables controlled perspective and view variation while maintaining visual consistency.
Intended for users working with ComfyUI who need reliable, repeatable multi-view image outputs for dataset and training-related workflows.
Description
Initial release of the Multi-Angle Dataset Workflow for Qwen Image Edit 2511 (Lightning).
Designed for consistent multi-view dataset generation using ComfyUI.
FAQ
Comments (8)
gives error. says Node 'ID #652:640' has no class_type. The workflow may be corrupted or a custom node is missing.: Node ID '#652:640'.
Sorry, do you have the latest comfy ui version? I downloaded it and testet it again, its working fine on my computer.
By far one of the best workflows out there, but you really need to update the instructions. on the exact same box you say "don't use any 1/8 steps LoRA", and a few lines further down you say "you MUST use a 8 steps LoRA, mandatory". took me a while to figure out, but now it's running smoothly.
could you let us in into the secret you figured out?
What worked for me was eliminating the turbo LoRA, and use the nunchaku version of the checkpoint with 8 steps. You can download it here: https://huggingface.co/nunchaku-ai/nunchaku-qwen-image-edit-2509/blob/main/lightning-251115/svdq-fp4_r128-qwen-image-edit-2509-lightning-8steps-251115.safetensors
You need to play around a little with the LoRA strenght to see what works for you, but it will probably be between 0.8 and 1.0
Ignore the name of the checkpoint file on huggingface, this is the Qwen Image Edit 2511 version, not the 2509
@vevelstad568 why do we even care about speed when the quality of this stage in LoRA training is paramount? Can't we just toss these lightning/speedups/nunchakus in the garbage bin and pump steps to 40 since that's ideal?