~ Recommended Settings & Training Parameters Below! ~
This is a Qwen-Image style lora for general color film photography. No trigger word is required but photography keywords (ex: "a photograph of... ...shallow depth of field, natural lighting...") can help the style come through! You can make your images look even more like film by adding film grain after the image is generated. I didn't add film grain on any of the example images as to show the model in it's raw form. I trained this model on all real (no synthetic) data, and masked all faces and hands to attempt to maximize compatibility with character loras. This is my first Qwen lora. Please let me know what you think if you use it. :)
Recommended Settings
Strength ... 0.70 - 1.00
Steps ... 20 - 50 (or above)
Shift ... 2.00
CFG ... 2.0 - 4.0
Sampler/Scheduler ... any but res_2s/bong_tangent (RES4LYF) is good.
This model and Qwen in general does best with long detailed captions and high resolutions with a high step count. Check the showcase for examples + ComfyUI workflow.
Custom nodes I used in the workflow are:
Rgthree (https://github.com/rgthree/rgthree-comfy)
Optional: Post-processing nodes (https://github.com/EllangoK/ComfyUI-post-processing-nodes)
Training Parameters
Because I hate gatekeeping, in the spirit of open source goodness, here's what I used ~
Tool ... Diffusion-pipe
LR ... 0.0003
Dim or "Rank" ... 8
Steps ... 1500
Epochs ... 150
Scheduler ... AdamW8Bit
46 total images, 1 repeats, Alpha Mask enabled
Resolutions trained ... 1024x1024, 1024x768, 1024x679