Your AI-generated Art

Posting this here because it relates to "AI Art" more than the other topics on MyBB and didn't think there needed to be a dedicated thread for this, but here goes.

I’ve just open-sourced a custom ComfyUI node pack called CamSim v2 that simulates how a real digital camera behaves instead of just stacking generic “filters”.

It’s probably niche, especially for those not using ComfyUI, but I figured I’d share it in case anyone here is into AI image realism, cinematic looks, or metadata fakery 😅

GitHub:
👉 https://github.com/FLi79Za/comfyui_camsim2

What It Does​

CamSim v2 simulates:
  • 📷 Real camera exposure logic (aperture, shutter, ISO, EV comp)
  • 🌙 ISO-based noise & dynamic range behaviour
  • 🎯 Motion blur based on shutter + focal length + stabilisation
  • 🌈 Chromatic aberration
  • 🌑 Lens vignette
  • 🔥 Pixel-level exposure pushing (not just brightness sliders)
  • 🎞 3D LUT (.cube) support
  • 🌫 Depth of field (radial or depth-map driven)
  • 🏷 Real EXIF metadata writing (Make, Model, Lens, GPS, Date, etc.)
  • 📁 Real camera-style filenames (IMG_0001, DSC_####, etc.)
The idea is to treat your generated image like it was captured by an actual camera body + lens combo, rather than just “post-processed”.

You can pick things like:
  • Canon R5
  • Nikon Z6 II
  • Sony APS-C
  • Fuji X-T4
  • Smartphone profiles
  • Custom iPhone profiles
And pair them with matching lenses (or mismatched if you’re chaotic).

Why I Built It

When generating images in ComfyUI, everything often looks “digitally perfect”.

Real cameras don’t behave like that.

They have:
  • ISO grain behaviour
  • Reciprocal shutter blur
  • Stabilisation limits
  • Lens artefacts
  • Exposure physics
CamSim v2 tries to model that pipeline.

It’s not a Photoshop replacement — it’s more of a “virtual camera pass”.

How To Install​

  1. Go to your ComfyUI folder
  2. Open:



    ComfyUI/custom_nodes/

  3. Clone or download this repo into that folder:




    comfyui_camsim2
  4. Restart ComfyUI.

You’ll find the nodes under:

CamSim v2

Main node:

  • CamSim v2 Apply (IMAGE → IMAGE)
Just feed your generated IMAGE into it, choose a body + lens, tweak exposure and realism strength, and output.

Bonus: It Writes Real EXIF

If you export via the node, it can write:
  • Camera make & model
  • Lens model
  • Aperture
  • Shutter
  • ISO
  • GPS (optional)
  • Randomised or ranged dates
Yes, the image will actually look like it was shot on an R5 or iPhone 16 Pro Max.


Who This Is For​

  • People doing cinematic AI work
  • Anyone chasing realism instead of hyper-clean renders
  • People who enjoy “camera logic”
  • Nerds like me who care about EXIF tags for no rational reason

If anyone here is using ComfyUI and wants to try it out, feedback is welcome.

And if nobody cares, that’s fine too, it was fun to build 😄
 
Last edited:
Built a ComfyUI node and workflow that actually uses the actual text prompt to create the image as a "word cloud" style background for the image. It allows you to define the blending mode, font, the text placement and so on. Quite happy with how it turned out. You can optionally also use a different piece of text to derive the word cloud from, but it's interesting (to me at least) to see the actual words to create the image, as part of the image.

Cyborg.jpg
 
You must log in or register to reply here.
Back
Top
Sign up to the MyBroadband newsletter
X