FLUX Kontext Face Swap | Seamless Face Replacement
DualCLIPLoader
Already loads t5xxl_fp8_e4m3fn.safetensors
ComfyUI model file · Text encoder
T5 v1.1 XXL text encoder (fp8 e4m3fn) · Half-size T5 text encoder for FLUX.1 and SD3
t5xxl_fp8_e4m3fn.safetensors is the T5 v1.1 XXL text encoder stored in 8-bit float, half the size of fp16. It is the usual choice for FLUX.1 on machines with less than 32 GB of RAM. Put it in ComfyUI/models/text_encoders and load it in DualCLIPLoader with clip_l.
ComfyUI upcasts it as needed, so you do not need an fp8-capable GPU to use it.
t5xxl_fp8_e4m3fn.safetensors
The short version
This is the same T5 v1.1 XXL encoder as t5xxl_fp16, with its weights cast to the e4m3fn 8-bit float format. That halves the file to 4.89 GB and the memory it needs while encoding, for a small, usually invisible loss in prompt fidelity.
e4m3fn uses 4 exponent bits and 3 mantissa bits. The “scaled” variant, t5xxl_fp8_e4m3fn_scaled, stores extra scale factors to recover some precision for a slightly bigger file.
Install
ComfyUI’s own repackage. Recommended.
saves as t5xxl_fp8_e4m3fn.safetensors
Same SHA256. May ask you to sign in.
saves as t5xxl_fp8_e4m3fn.safetensors
ComfyUI/ └─ models/ ├─ … └─ text_encoders/ └─ t5xxl_fp8_e4m3fn.safetensors
$ cd ComfyUI/models/text_encoders$ wget -O t5xxl_fp8_e4m3fn.safetensors \$ https://huggingface.co/comfyanonymous/flux_text_encoders/resolve/main/t5xxl_fp8_e4m3fn.safetensors$ sha256sum t5xxl_fp8_e4m3fn.safetensors# 7d330da4816157540d6bb7838bf63a0f02f573fc48ca4d8de34bb0cbfd514f09
Safetensors files hold tensors only and cannot run code. Still compare the SHA256: a failed download often leaves an HTML error page saved under the model’s filename.
Use
FLUX.1 loads T5 next to clip_l in DualCLIPLoader (type flux). SD3 loads it in TripleCLIPLoader with clip_g and clip_l. Older video models such as CogVideoX load it alone in Load CLIP.
positive●
Order does not matter
DualCLIPLoader accepts T5 in either slot; the type setting decides how each is used.
Low on RAM?
t5xxl_fp8_e4m3fn is half the size of fp16 with a small quality cost.
Works with
t5xxl_fp8_e4m3fn.safetensors is one part of the stack. A workflow that reports a missing model usually needs these too. FLUX.1 dev is a gated download: accept the license on Hugging Face first.
Run it
We match workflows by the model files they load: this exact file first, then other versions of it, then workflows built for the same use, then other models from the same family.
19 workflows
These workflows already load t5xxl_fp8_e4m3fn.safetensors. Open one and run it as is.
DualCLIPLoader
Already loads t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
Already loads t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
Already loads t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
Already loads t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
Already loads t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
Already loads t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
Already loads t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
Already loads t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
Already loads t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
Already loads t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
Already loads t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
Already loads t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
Already loads t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
Already loads t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
Already loads t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
Already loads t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
Already loads t5xxl_fp8_e4m3fn.safetensors
24 workflows
These workflows load a different precision or format of the same model. Select t5xxl_fp8_e4m3fn.safetensors in their DualCLIPLoader node to use this version.
DualCLIPLoader
t5xxl_fp16.safetensors
Replace with t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
t5xxl_fp16.safetensors
Replace with t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
t5xxl_fp16.safetensors
Replace with t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
t5xxl_fp8_e4m3fn_scaled.safetensors
Replace with t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
t5xxl_fp8_e4m3fn_scaled.safetensors
Replace with t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
t5xxl_fp8_e4m3fn_scaled.safetensors
Replace with t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
t5xxl_fp8_e4m3fn_scaled.safetensors
Replace with t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
t5xxl_fp16.safetensors
Replace with t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
t5xxl_fp8_e4m3fn_scaled.safetensors
Replace with t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
t5xxl_fp8_e4m3fn_scaled.safetensors
Replace with t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
t5xxl_fp16.safetensors
Replace with t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
t5xxl_fp16.safetensors
Replace with t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
t5xxl_fp8_e4m3fn_scaled.safetensors
Replace with t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
t5xxl_fp16.safetensors
Replace with t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
t5xxl_fp16.safetensors
Replace with t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
t5xxl_fp16.safetensors
Replace with t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
t5xxl_fp16.safetensors
Replace with t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
t5xxl_fp16.safetensors
Replace with t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
t5xxl_fp16.safetensors
Replace with t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
t5xxl_fp16.safetensors
Replace with t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
t5xxl_fp16.safetensors
Replace with t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
t5xxl_fp16.safetensors
Replace with t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
t5xxl_fp16.safetensors
Replace with t5xxl_fp8_e4m3fn.safetensors
DualCLIPLoader
t5xxl_fp16.safetensors
Replace with t5xxl_fp8_e4m3fn.safetensors
Compare
| File | Relation | Size | Best for |
|---|---|---|---|
| t5xxl_fp8_e4m3fn.safetensors | This page · fp8 e4m3fn | 4.89 GB | Machines under 32 GB RAM |
| t5xxl_fp8_e4m3fn_scaled.safetensors | Same encoder · fp8 scaled | 5.16 GB | fp8 with better accuracy |
| t5xxl_fp16.safetensors | Same encoder · fp16 | 9.79 GB | Best prompt fidelity |
Troubleshoot
Why: The workflow asks for this exact filename, but ComfyUI cannot find it in models/text_encoders. Usually the file is missing, sits in a different folder.
Fix: Move the file into ComfyUI/models/text_encoders/, rename it to match the workflow (or pick it again in the dropdown), then refresh the node list or restart ComfyUI.
Why: The download is incomplete, or the browser saved an HTML sign-in or error page under the model’s filename.
Fix: Check that the file is exactly 4,893,934,904 bytes and that its SHA256 matches the value above. Download it again from Hugging Face · comfyanonymous/flux_text_encoders if not.
Rather skip the setup?
Run FLUX Kontext Face Swap in your browser on a cloud GPU. Nothing to download.
FAQ
t5xxl_fp8_e4m3fn_scaled.safetensors first.ComfyUI/models/text_encoders/. Then refresh the node list or restart ComfyUI so DualCLIPLoader lists the file.Sources: comfyanonymous/flux_text_encoders · Civitai model 704402
RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Models, enabling artists to harness the latest AI tools to create incredible art.