fofr / cog-consistent-character

Create images of a given character in different poses
https://replicate.com/fofr/consistent-character
MIT License
560 stars 54 forks source link

No explanation of workflow, re-named model files? #5

Open ewebgh33 opened 3 months ago

ewebgh33 commented 3 months ago

Hi Was missing a few model files when loading rhe JSON. Need help with one. Found both thibaud_xl_openpose.safetensors and bbox/Eyes.pt via a google search. The openpose was easy enough to find, the Eyes.pt looks like it came from CivitAI - hopefully that's right.

But there is also asking to load as follows:

Prompt outputs failed validation
InstantIDModelLoader:
- Value not in list: instantid_file: 'instantid-ip-adapter.bin' not in ['ip-adapter.bin']
ControlNetLoader:
- Value not in list: control_net_name: 'instantid-controlnet.safetensors' not in (list of length 41)

And I can't find a model for InstandID with THAT filename. I can find others, for example https://github.com/Mikubill/sd-webui-controlnet/discussions/2589 https://huggingface.co/Aitrepreneur/InstantID-Controlnet/tree/main/checkpoints And the original Instant ID github https://github.com/InstantID/InstantID

These link to .bin and a .safetensors but not with the names your workflow has saved. Have you renamed the models you used, instead of using the filenames that most repos have in them?? Otherwise, what InstandID model version from which author is this please?

Thanks

fofr commented 3 months ago

I renamed the file to be unique, as this is a limitation of the way the cog automatically downloads weights from Replicate.

This is the bin file you need: https://huggingface.co/fofr/comfyui/tree/main/instantid

ewebgh33 commented 3 months ago

Thanks @fofr Does the comfyUI version work the same as the replicate version (aside from number of generations when running being 1 not 4/5)? It seems to be... not as accurate? Something like that

(Edit: I am not using COG in any part of this, does that have something to do with results?)

ArianaStar commented 3 months ago

@ewebgh33 I use Conda & WSL2 & COG, my results are very good.