v1 - BASE WORKFLOW
v2 - BASE + DYPE NODE - (YOU MUST INSTALL DYPE NODE IN COMFYUI MANAGER CUSTOM NODES!)
dype node natively generates images at higher resolutions than the base model can generate. the base model can natively generate up to 1024 resolution. with dype, you can generate up to 3K resolution without an upscale model. the node also cleans up artifacting and hallucinations from the model through a model patch inside the node. VERY USEFUL! be warned, generations times will increase by up to 5x depending on specs. node is preset with correct settings and values are set at 2K landscape by default.
use the following values for proper resolution:
3840 - 3k
2160 - 2k
1440 - 1k
for landscape/portrait, just interchange the corresponding lower value.
example: 2160w X 1440h
dype struggles with portrait values, run multiple seeds to try and correct stretching. (youll see)
condensed subgraph workflow for the NEW Qwen Image 2512!
Lightning LoRA is compatible and workflow is set to 8 steps.
IMPORTANT!
shift can be adjusted if images come out artifacted. currently set to 4, test values anywhere from 4.0 to 13.0 to correct artifacting. 13.0 works the best for me personally
you will activate the ACTIVE text encoder (Instruct UD), the mmproj is an inactive text encoder that is passively activated through the workflow, no need to place it in. you just activate the regular UD instruct encoder in the workflow and it will call on the other.
GGUF Model
https://huggingface.co/unsloth/Qwen-Image-2512-GGUF/tree/main
NEW 4-Step 2512 Lightning LoRA
https://huggingface.co/lightx2v/Qwen-Image-2512-Lightning/tree/main
8-step Lightning LoRA (old)
https://huggingface.co/lightx2v/Qwen-Image-Lightning/tree/main
ACTIVE Text Encoder (BOTH REQUIRED) (INSTANT DOWNLOAD)
INACTIVE mmproj Text Encoder (BOTH REQUIRED) (INSTANT DOWNLOAD)
https://huggingface.co/unsloth/Qwen2.5-VL-7B-Instruct-GGUF/resolve/main/mmproj-BF16.gguf
VAE (INSTANT DOWNLOAD)