black-forest-labs

/

FLUX.2-small-decoder

like 103

Follow

Black Forest Labs 36.3k

It's the same decoder for all open variants of our models. So you can download the Small decoder for FLUX.2 [klein] 9B, 4B, KV and FLUX.2 [dev]

The problem is, there is 3 files to download 🤔😅

Folder path:

black-forest-labs/FLUX.2-small-decoder

Files in the folder:

diffusion_pytorch_model.safetensors
full_encoder_small_decoder.safetensors
small_decoder.safetensors

By the way, thank you Stephen 😎

I downloaded full_encoder_small_decoder.safetensors and it works fine in ComfyUI.

Maybe small_decoder.safetensors would work too, not sure, but giving them a try probably won't hurt.

None of them work for me

I didn’t understand where exactly they should be put in Comfy!?

update comfy ui, put the full_encoder_small_decoder.safetensors in the vae folder
small_decoder.safetensors is just the small decoder
diffusion_pytorch_model.safetensors is for diffusers

update comfy ui, put the full_encoder_small_decoder.safetensors in the vae folder
small_decoder.safetensors is just the small encoder
diffusion_pytorch_model.safetensors is for diffusers

Thank you, it really worked!!!!

update comfy ui, put the full_encoder_small_decoder.safetensors in the vae folder
small_decoder.safetensors is just the small decoder
diffusion_pytorch_model.safetensors is for diffusers

Update ComfyUI == Break all my subgraphs :(

update comfy ui, put the full_encoder_small_decoder.safetensors in the vae folder
small_decoder.safetensors is just the small decoder
diffusion_pytorch_model.safetensors is for diffusers

Hi,
Thanks for the release!

I’ve updated ComfyUI, placed full_encoder_small_decoder.safetensors in the vae folder, and selected full_encoder_small_decoder.safetensors as the VAE in the Flux.2 VAELoader node.
However, I’m getting this error:
RuntimeError: Error(s) in loading state_dict for AutoencoderKL:
size mismatch for decoder.conv_in.weight: copying a param with shape torch.Size([384, 32, 3, 3]) from checkpoint, the shape in current model is torch.Size([512, 32, 3, 3]).

Any advice on what could be wrong?

--> EDIT : it works with comfy UI v19 (14/4/2026)

update comfy ui, put the full_encoder_small_decoder.safetensors in the vae folder
small_decoder.safetensors is just the small decoder
diffusion_pytorch_model.safetensors is for diffusers

Hi,
Thanks for the release!

I’ve updated ComfyUI, placed full_encoder_small_decoder.safetensors in the vae folder, and selected full_encoder_small_decoder.safetensors as the VAE in the Flux.2 VAELoader node.
However, I’m getting this error:
RuntimeError: Error(s) in loading state_dict for AutoencoderKL:
size mismatch for decoder.conv_in.weight: copying a param with shape torch.Size([384, 32, 3, 3]) from checkpoint, the shape in current model is torch.Size([512, 32, 3, 3]).

Any advice on what could be wrong?

same issue, updated

hello, i have same mismatch error for comfyui v18.5.

edit: it is working with comfyui v19.