Support int8 models.

[comfyanonymous]

This is pretty basic, lora support needs to be improved.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 7ac66646-4151-4ee5-bbb8-1774f355a8e4

📥 Commits

Reviewing files that changed from the base of the PR and between 639c8fa and 3507d3b.

📒 Files selected for processing (4)

• comfy/ops.py
• comfy/quant_ops.py
• requirements.txt
• tests-unit/comfy_quant/test_mixed_precision.py

---

📝 Walkthrough

Walkthrough

This PR adds int8_tensorwise quantization support, including TensorWise INT8 layout registration, load/save handling for weight_scale and optional convrot metadata, and changes to MixedPrecisionOps.Linear so quantized execution can skip input quantization when required. It also updates the comfy-kitchen dependency and adds a unit test covering metadata loading, forward output, and state-dict serialization.

🚥 Pre-merge checks | ✅ 3 | ❌ 2

❌ Failed checks (1 warning, 1 inconclusive)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 44.44% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.
Description check	❓ Inconclusive	No description was provided, so the intent cannot be assessed from it.	Add a short description of the int8 model support and key implementation details.

✅ Passed checks (3 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly matches the main change: adding support for int8 models.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[monnky]

thank you so much for int8 @comfyanonymous

[Dazrock]

May i ask what node it is to get it to work?
I looked everywhere, and i can't seem to find a node that uses it.
My application is fully updated.

[krishna0354]

May i ask what node it is to get it to work? I looked everywhere, and i can't seem to find a node that uses it. My application is fully updated.

The regular load diffusion model should wrork if you the right INT8 format

[Dazrock]

May i ask what node it is to get it to work? I looked everywhere, and i can't seem to find a node that uses it. My application is fully updated.

The regular load diffusion model should work if you the right INT8 format

I downloaded the Int8 fast loader. It runs, but i get images with just static noise.
I think it's because i didn't download the right branch of ComfyUI maybe? Idk...
thx for the reply tho.
Appreciate it!

[krishna0354]

I use comfyUI portable, and with the int8fast node never had any problem. Now Bob the creator of the int8 fast node created a script to convert his int8 convrot to the confyUI native format and is working even better. Yahoo Mail: Search, Organize, Conquer On Tue, Jun 30, 2026 at 9:41 AM, ***@***.***> wrote: Dazrock left a comment (Comfy-Org/ComfyUI#14636) May i ask what node it is to get it to work? I looked everywhere, and i can't seem to find a node that uses it. My application is fully updated. The regular load diffusion model should work if you the right INT8 format I downloaded the Int8 fast loader. It runs, but i get images with just static noise. I think it's because i didn't download the right branch of ComfyUI maybe? Idk... thx form the reply tho. Appropriate it! — Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications, keep track of coding agent tasks and review pull requests on the go with GitHub Mobile for iOS and Android. Download it today! You are receiving this because you commented.Message ID: ***@***.***>

[Dazrock]

I use comfyUI portable, and with the int8fast node never had any problem. Now Bob the creator of the int8 fast node created a script to convert his int8 convrot to the confyUI native format and is working even better. Yahoo Mail: Search, Organize, Conquer On Tue, Jun 30, 2026 at 9:41 AM, @.> wrote: Dazrock left a comment (Comfy-Org/ComfyUI#14636) May i ask what node it is to get it to work? I looked everywhere, and i can't seem to find a node that uses it. My application is fully updated. The regular load diffusion model should work if you the right INT8 format I downloaded the Int8 fast loader. It runs, but i get images with just static noise. I think it's because i didn't download the right branch of ComfyUI maybe? Idk... thx form the reply tho. Appropriate it! — Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications, keep track of coding agent tasks and review pull requests on the go with GitHub Mobile for iOS and Android. Download it today! You are receiving this because you commented.Message ID: @.>

Yeah. The regular Model Loader only has the older variants of FP8 quantization. There isn't any options for Int8.
If i attempt to run the regular Model Loader in default mode (Because that's the only option i have which isn't selecting a old fp8 format) it just gives me a error. Saying the quantization type is invalid.
For the models tested i downloaded rdbtAnima Int8, and Anima base v1 Int8, and AnimayumeInt8 v10BaseFinal.
All of them fail to run. One of them was showing a Tensorwise error?
I've confirmed that my ComfyUI ver is 0.26.2 portable.
Thx for the help.