Face Foundry SD1.5 - Consistent Characters from Face Find

5.0

0 reviews
35
6.6K
1.7K
0
Description

Discover a unique face for your character! This is is part of the Face Finder workflow to generate diverse, random characters, based on randomly generated names. By doing this, we sidestep the use of celebrity images and names, ensuring a pool of original names and faces for your characters, free from ethical dilemmas.


How to use it:

0a. Download all the stuff (there's a lot, sorry). Check the notes around the lower nodes for explanations.

0b. Generate a random face using one of my Face Finder workflows if you haven't already.

(SDXL) https://openart.ai/workflows/wilburn/face-find-sdxl/OsArpYrUDTAuqYsR0nKc

(SD1.5)  https://openart.ai/workflows/wilburn/face-finder-sd15/T6gQM9mafaesRfAIaPqH

1. Place the face into the Face Reference Image node to the right

2. (optional) Import an image to pull the style from (make sure to enable the style IPAdapter to the right by right-clicking and select "Set Group Nodes to Always")

3. Don't forget the prompt.

4. Try it out!


Tips:

1. The SD1.5 models tend to work best imitating faces that have been generated by them to begin with. And you can reinforce the effect by using the random name you used to make the face (as long as you are using the same model).

2. SD1.5 models tend to be biased female, and seem to edge toward the young side. So even when feeding an adult face, you will need to play with the prompts some to get what you want. I suggest looking at some of the age embeddings on civitai.com.

3. Because of the way IPAdapters seem to work, I find it difficult to prompt for the type of image and style I want at the same time as getting the face right. So I like to generate a generic image to get the pose and style, and then use that as the style image and re-sample it. That makes for a much better aesthetic than just a face swap in my opinion.

4. If you leave the style image/IPAdapter off, you will probably need to play with the strengths of the face models or you will pick up too much of the face image and your prompt will get ignored.


Next Steps (How I use this in my overall workflow):

1. I generate like 100 faces using one of my Face Find workflows

2. Pick about 5 faces I like, then generate several more faces with the same name

3. Pick a favorite face for the character, upscale and detail the faces

4. Use the face image in a face replace workflow, like THIS ONE! This IPAdapter method is my preferred for now, though I am closely watching for development of InstantID.


Additional comments:

-- Please use this responsibly. This is not for non-consensual deep fakes. Ever. That's not just a disclaimer. I build this whole set of workflows to try to protect real people's faces. These are crazy powerful toys we get to play with. So let's be respectful.

-- I cherry-picked which images I uploaded. But I didn't modify the workflow or text prompt at all between the different characters. So while I did pick the best of the batch to showcase, I am confident that with some intentional prompting you can get even better results.

Node Diagram
Discussion

(No comments yet)

Author

5
8.9K
152
37.6K

Reviews

No reviews yet

Versions (1)

  • - latest (a year ago)

Primitive Nodes (11)

IPAdapterApply (2)

IPAdapterApplyFaceID (1)

InsightFaceLoader (1)

Note (6)

PrepImageForInsightFace (1)

Custom Nodes (37)

  • - CR Integer Multiple (2)

  • - CR VAE Input Switch (1)

  • - CR Apply LoRA Stack (1)

  • - CR Seed (1)

  • - CR SD1.5 Aspect Ratio (1)

  • - CR LoRA Stack (1)

ComfyUI

  • - KSampler (2)

  • - VAEDecode (3)

  • - PreviewImage (5)

  • - ImageScale (1)

  • - ImageUpscaleWithModel (1)

  • - VAEEncode (1)

  • - VAELoader (1)

  • - CLIPSetLastLayer (1)

  • - CLIPTextEncode (2)

  • - UpscaleModelLoader (1)

  • - LoraLoaderModelOnly (1)

  • - LoadImage (2)

  • - CheckpointLoaderSimple (1)

  • - CLIPVisionLoader (1)

  • - SaveImage (1)

  • - IPAdapterModelLoader (3)

  • - PrepImageForClipVision (2)

  • - DZ_Face_Detailer (1)

Checkpoints (1)

dreamshaper_8.safetensors

LoRAs (2)

15\detail_slider_v4.safetensors

ip-adapter-faceid-plusv2_sd15_lora.safetensors