search

invoke-ai / InvokeAI

Invoke is a leading creative engine for Stable Diffusion models, empowering professionals, artists, and enthusiasts to generate and create visual media using the latest AI-driven technologies. The solution offers an industry leading WebUI, and serves as the foundation for multiple commercial products.
https://invoke-ai.github.io/InvokeAI/
Apache License 2.0
23.72k stars 2.43k forks source link

[bug]: Cannot use other FLUX checkpoints with VAE or CLIP embedded. #7137

Open SafwanPK opened 1 month ago

SafwanPK commented 1 month ago

Is there an existing issue for this problem?

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

dsent commented 4 weeks ago

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.

LiJT commented 3 weeks ago

Same here, it take looooads of Vram even I select fp8 model, it is unusable with Flux....

dsent commented 3 weeks ago

@LiJT try using FLUX (Quantized) installed from Starter Models tab in Invoke. It's NF4.