Not exactly film grain, but it can apply an interesting grainy effect to your images. Give it a try and see what you think.
V04
This is my final attempt to produce a LoRA that approximates a film grain effect. Realistic analog film grain produces features that are difficult to reproduce digitally. I spent a lot of time producing new training images and then running the output through ChatGPT to determine the best prompt to help produce a realistic film grain result. Results vary across models. Some models produce better results. Take a look for yourself.
Each of the prompts includes the following lines (more or less):
35mm Kodak Portra 400 pushed one stop, fine natural grain, soft highlight roll-off, neutral skin tones. Slight shadow grain clumping. Indoor tungsten lighting, visible film grain.
A monochrome image.
A bright grainy photo, overexposed. With an artistically pronounced film grain effect in varying intensity across the image. Natural lighting.Note the parts: neutral skin tones and indoor tungsten lighting
Exclude these as appropriate, depending on your subject and setting. ChatGPT suggested all of this, and it seemed to produce a better result. Maybe you can come up with something better? I still think that, at times, the result is a little flat and smooth.
Let's see your results.
Description
FAQ
Comments (7)
If you don't mind giving me the information, but how many steps? how many images? and what LR did you use? I'm trying to get the hang of Qwen loras.
I use AI-Toolkit for all my Qwen training, largely using the default settings in the UI. Learning rate is 0.0001. 43 images. 3000 steps. My image count is usually 40-50.
@CreativeEdge many thanks! :)
I'm a bit confused, none of the example pics have any grain
Perhaps, then, it isn't as effective as I hoped. 😶. There are plenty of other examples on the site. I imagine one will suit your needs better.
@CreativeEdge you nailed all the 'next gen' models for v4.. great results!
Hi, @Narz, I'm delighted you think so. Thanks so much for your feedback.








