What this workflow does

This basically makes your IRL photo into a collage/school yearbook photoshoot. It's got some wacky and not so wacky combinations - anything could happen 😉 - It will try to make your image younger or older (depending on the image subject) and then time travel you into the nostalgic past of a 90's student leavers yearbook.  It is ment to look deliberately 90's bad photo style.

How to use this workflow

Best results are with SDXL Models but can also be used with SD models.

I'm personally still not fond of Flux and SD3 - but feel free to try them.

This makes use of the DLIB face enhancer that is in ComfyUI-Bringing-Old-Photos-Back-to-Life.

You may encounter issues while trying to install dlib on Windows.  A Helpful guide is found here -> https://github.com/cdb-boop/ComfyUI-Bringing-Old-Photos-Back-to-Life?tab=readme-ov-file#installation-troubleshooting

Once you install this node, you may have to manually download and install the models -> https://github.com/cdb-boop/ComfyUI-Bringing-Old-Photos-Back-to-Life?tab=readme-ov-file#2-models

🟦 Configuration 🟦

Checkpoint Loader - Select your Checkpoint Model and Optional VAE

Global Sampler - Default one is good here, but if your Checkpoint Model suggests a different one, then you can try that.

Control Canny Model - Here you select the ControlNet Canny Model to match your Checkpoint Model Type (SD/SDXL)

Control Depth Model - Here you select the ControlNet Depth Model to match your Checkpoint Model Type (SD/SDXL)

dlib Face Enhancer Model - Point this at FaceSR_512\Latest_net_G.pth

dlib Face Detector Model - Point this at shape_predictor_68_face_landmarks.dat

Global Seed - Self-explanatory, if you like an image but need to tweak stuff, then you can fix the seed here if you want to

CLIP Set Last Layer - "-1" should be good for all here, but some checkpoint models may suggest a different clip, so change here if needed.

InsightFace - Select your provider flavour.  CPU works for everyone but pick another if your PC supports it for faster processing.

Upscale Model (4x) - Pick a x4 upscale model.  Must be a x4.

I suggest Nomos8kHAT-L_otf

Detailed Upscale Model (4x) - Pick a x4 Detailed upscale model.  Must be a x4.

I suggest 4x_NMKD-SuperscaleV2_2k

Upscale Model (1x Skin Detailer) - Pick a 1x Skin Detailer model.  Must be a x1.

I suggest 1x-ITF-SkinDiffDetail-Lite-v1

🟪Your Image Configuration🟪

Load Your Image - Load an image of yourself (or a subject).

Ideally this needs to be a ratio of 1:1, but I have a built-in routine to work it out.

It doesn’t need to be high resolution.  Try to stick to a max of 1024x1024

Gender Selector - Select your gender.  Select non-binary if you prefer not to select.

Remember, AI uses "best guess by looks" when it comes to gender and tends to swing between male, female and non-binary, so please be forgive full as it does not know any better.

Extra Injector - 99% of the time you can leave this as True.  But sometimes you may see face distortions or artifact.  If this happens, lock the Global Seed to "Fixed" and switch this to "false" and try again.

Use Filter - This basically sharpens the injector image.  Can create novel old photo sharpness.  Up to you, have a play 😊

Enhance Your Photo First? - This will pass your IRL photo though an enhancement filter.  It sometimes helps, especially if your image is slightly off quality/blurry.

Enable Focusing - Basically, this will turn on the Upscaling routine.  Upscaling takes a hot minute to process so if you want, you can turn this to "No" until you get an image you like, then lock the Global Seed to "Fixed" and switch this to "Yes" to produce the final Image.

🟫 Processing 🟫

No need to change anything here.

Tips about this workflow

No real tips other than pick an IRL image that clearly shows your (or your subjects) face without obstacles.

Try to keep the IRL image ration 1:1 and size 1024x1024 max.

Version Changes

v1.0

• Release Workflow