invoke-ai
[bug]: Cannot use other FLUX checkpoints with VAE or CLIP embedded.
Is there an existing issue for this problem?
- [X] I have searched the existing issues
Operating system
Windows
GPU vendor
Nvidia (CUDA)
GPU model
RTX 3060
GPU VRAM
12 GB
Version number
5.1.1
Browser
Chrome latest version
Python dependencies
No response
What happened
Flux in cmd reverts back to menu selection when inpainting or generating with custom FLUX checkpoints. I've tried using the following model; https://civitai.com/models/652009?modelVersionId=963489 but it fails and doesn't generate. I think the issue may be the fact that there is no "None" option in VAE or CLIP selections, and they are marked as required, not optional.
What you expected to happen
Generate the images as instructed like in FLUX Dev
How to reproduce the problem
No response
Additional context
No response
Discord username
No response
I'm trying to use an NF4-quantized FLUX.1 dev from here: https://huggingface.co/lllyasviel/flux1-dev-bnb-nf4/blob/main/flux1-dev-bnb-nf4-v2.safetensors It seems to work, but still, I can't disable VAE, T5 and CLIP provided by Invoke in favor of the ones already embedded into the model. I'm not sure if they are really used or not.
Same here, it take looooads of Vram even I select fp8 model, it is unusable with Flux....
@LiJT try using FLUX (Quantized) installed from Starter Models tab in Invoke. It's NF4.
What is the state here? I guess i am in same situation: i have a Flux model with embedded CLIP T5 and VAE but "Invoke" is disable and tells me that those files are missing.
Are you plan to support Flux models with embedded stuff? probably a checkbox would be enough - no validation needed i think.
