-
-
Notifications
You must be signed in to change notification settings - Fork 140
Flux
Calamdor edited this page Sep 5, 2024
·
4 revisions
This page is a work in progress as information is learned about Flux.
Flux is a DiT Transformer flowmatching model that has high learning potential but it is a very large and also slow model to work with.
Like SD3, a hugginfgace key or a local copy of the diffusers model is needed.
- Finetune is not currently possible.
- Embeddings do not likely work, due to the nature of T5.
- Single file safetensors is currently not supported as a model import
- Full Dora does not currently work completely with the Comfy Lora Loader
- Lora is currently the only recommended training.
- NF4 is possible with the model weights. INT8 requires bits and bytes fixes and is not currently supported.
- The OneTrianer Lora can be used in Comfy in the standard Lora Loader.
- Flux can learn at lower resolutions, for example 512 and 768, which is a benefit for people with less powerful graphics cards.
- It is possible to train a Flux Lora on 12GB. 8GB cards are likely too close to the limit, and would require windows to be using a different GPU for VRAM usage as Windows can quite a bit of VRAM.