feat: add TAEQI2.1 support for Qwen Image 2.1 by leejet · Pull Request #2123 · leejet/stable-diffusion.cpp

leejet · 2026-10-10T15:33:55Z

Summary

Add TAEQI2.1 encoding, decoding, and previews for Qwen Image 2.1, with 64-channel latents, 16x spatial compression, and RGBA output.

Related Issue / Discussion

N/A

Additional Information

N/A

Checklist

I have read and confirmed this PR follows the contribution guidelines.

Summary by CodeRabbit

New Features
- Added TAE support for Qwen Image 2.1, including image encoding, decoding, and previews.
- Supports 64-channel latents, 16× spatial scaling, and RGBA images. Official safetensors weights can be used without conversion.
Documentation
- Added setup and usage instructions for encoding, decoding, and generating previews with Qwen Image 2.1.

coderabbitai · 2026-10-10T15:34:20Z

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository UI
Review profile: CHILL
Plan: Advanced
Run ID: e33cf889-4087-49d6-b35d-6b6a45f02e04

📥 Commits

Reviewing files that changed from the base of the PR and between b7a74e6 and 690d504.

📒 Files selected for processing (4)

docs/qwen_image_2.1.md
docs/taesd.md
src/model/vae/tae.hpp
src/pipeline/model_builders.cpp

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 7 remain after this review.

📝 Walkthrough

Walkthrough

Adds Qwen Image 2.1 support to TAESD with a 16-channel image representation, 64 latent channels, and patch size 2. The model builder selects the image autoencoder path. Documentation describes weight download and CLI usage for encoding, decoding, and previews.

Changes

Qwen Image 2.1 TAE Support

Layer / File(s)	Summary
TAESD f16 model support `src/model/vae/tae.hpp`	`TinyEncoder` and `TinyDecoder` accept an `f16` mode. For Qwen Image 2.1, `TAESD` enables that mode, sets 64 latent channels, patchifies encoder input, and unpatchifies decoder output.
Model selection and usage instructions `src/pipeline/model_builders.cpp`, `docs/taesd.md`, `docs/qwen_image_2.1.md`	`create_tae` selects `TinyImageAutoEncoder` for Qwen Image 2.1. The documentation adds TAEQI2.1 setup and CLI instructions for encoding, decoding, and previews.

Priority: ⬇️ Low

Estimated code review effort: 3 (Moderate) | ~20 minutes

Change: Feature

Suggested reviewers: stduhpf, carlgao4

Merge Risk | ⚪ Minimal · up to 690d5

Merge Risk: ⚪ Minimal · up to 690d5

No actionable issue was established in the Qwen Image 2.1 TAE changes. The PR is mergeable after normal checks, though loading the official checkpoint remains untested.

Security Architecture Review

Security architecture risk: 🔵 Low · up to 690d5

The change remains within the existing local processing boundary, with no demonstrated expansion of privileges. Risk is low rather than minimal because malformed-input behavior and exposure through external integrations are not fully established.

Retained concerns
No architecture-level concerns identified.

Security review details

Security Blast Radius

inferred — The demonstrated exposure is image and latent processing within the caller's existing inference process and backend. Malformed-input failures could affect that process's availability, but the inspected construction and execution paths establish no new privilege or cross-tenant transition. Deployment-specific tenant sharing and external access to low-level C++ entrypoints remain unknown.

Trust Boundaries and Controls

observed — The inspected init-image path normalizes image channels before encoding. Model metadata validation rejects missing required sources and mismatched registered tensor dimensions. These controls predate the PR and remain on the selected processing path; they do not establish complete validation of arbitrary direct decode inputs or checkpoint authenticity.

Resilience and Maintainability Implications

inferred — Ordinary execution failures retain existing cleanup and retry containment rather than introducing a new persistent transition. Reentrant graph execution is rejected by the existing active-state guard, but this is not evidence of thread-safe concurrent use. Stronger cancellation and interruption guarantees remain unestablished.

Pre-merge checks | 4 | 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage		Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 7 functions across 2 files. (2 skipped: 2 …	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check		The title clearly and concisely describes the main change: adding TAEQI2.1 support for Qwen Image 2.1.
Description check		The description includes all required sections, explains the change, identifies related issues as N/A, provides additional information as N/A, and confirms the contribution checklist.
Linked Issues check		Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check		Check skipped because no linked issues were found for this pull request.

Full details: Docstring Coverage

Explanation

Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 7 functions across 2 files. (2 skipped: 2 unsupported.)

Fix all pre-merge checks with AI

✨ Finishing Touches 💡 1

📝 Generate docstrings 💡

Commit to this branch
Create a new PR

🧪 Generate unit tests (beta)

Commit to this branch
Create a new PR

Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Warning

Some tools did not complete. Review the errors below.

🔧 Cppcheck (2.22.0)

src/pipeline/model_builders.cpp

Cppcheck timed out; analysis of this file is incomplete

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

stduhpf · 2026-10-11T00:58:52Z

Oh.
I think I started working on this right before you merged it, I was confused why I had conficts when trying to rebase my implementation on lastest master.

CMay · 2026-10-11T20:45:41Z

Having trouble getting this to work using the instructions provided using the latest release build. Either keeping the VAE and using the TAE for decoding only, or using it exclusively. I get metadata errors.

[V] validating model metadata --- diffusion_engine.cpp:1213
[E] VAE tensor 'tae.decoder.1.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.1.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.10.conv.0.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.10.conv.0.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.10.conv.2.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.10.conv.2.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.10.conv.4.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.10.conv.4.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.11.conv.0.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.11.conv.0.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.11.conv.2.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.11.conv.2.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.11.conv.4.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.11.conv.4.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.13.conv.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.14.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.15.conv.0.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.15.conv.0.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.15.conv.2.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.15.conv.2.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.15.conv.4.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.15.conv.4.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.16.conv.0.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.16.conv.0.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.16.conv.2.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.16.conv.2.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.16.conv.4.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.16.conv.4.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.17.conv.0.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.17.conv.0.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.17.conv.2.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.17.conv.2.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.17.conv.4.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.17.conv.4.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.19.conv.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.20.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.22.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.22.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.3.conv.0.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.3.conv.0.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.3.conv.2.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.3.conv.2.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.3.conv.4.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.3.conv.4.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.4.conv.0.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.4.conv.0.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.4.conv.2.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.4.conv.2.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.4.conv.4.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.4.conv.4.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.5.conv.0.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.5.conv.0.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.5.conv.2.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.5.conv.2.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.5.conv.4.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.5.conv.4.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.7.conv.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.8.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.9.conv.0.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.9.conv.0.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.9.conv.2.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.9.conv.2.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.9.conv.4.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.decoder.9.conv.4.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.0.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.0.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.10.conv.0.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.10.conv.0.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.10.conv.2.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.10.conv.2.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.10.conv.4.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.10.conv.4.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.11.conv.0.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.11.conv.0.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.11.conv.2.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.11.conv.2.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.11.conv.4.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.11.conv.4.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.12.conv.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.13.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.14.conv.0.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.14.conv.0.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.14.conv.2.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.14.conv.2.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.14.conv.4.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.14.conv.4.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.15.conv.0.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.15.conv.0.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.15.conv.2.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.15.conv.2.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.15.conv.4.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.15.conv.4.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.16.conv.0.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.16.conv.0.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.16.conv.2.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.16.conv.2.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.16.conv.4.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.16.conv.4.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.17.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.17.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.2.conv.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.3.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.4.conv.0.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.4.conv.0.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.4.conv.2.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.4.conv.2.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.4.conv.4.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.4.conv.4.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.5.conv.0.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.5.conv.0.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.5.conv.2.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.5.conv.2.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.5.conv.4.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.5.conv.4.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.6.conv.0.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.6.conv.0.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.6.conv.2.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.6.conv.2.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.6.conv.4.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.6.conv.4.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.7.conv.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.8.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.9.conv.0.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.9.conv.0.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.9.conv.2.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.9.conv.2.weight' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.9.conv.4.bias' not in model metadata --- model_manager.cpp:763
[E] VAE tensor 'tae.encoder.9.conv.4.weight' not in model metadata --- model_manager.cpp:763
[E] model metadata validation failed --- diffusion_engine.cpp:1272

That is with this file: https://huggingface.co/madebyollin/taeqi2_1/resolve/main/taeqi2_1.safetensors?download=true

Is there something I overlooked?

stduhpf · 2026-10-11T21:18:43Z

@CMay I think that's because the lastest published release is master-956-1b0ba10, on commit 1b0ba10, which predates this PR. The release build implementing this feature seems to have been canceled: https://github.com/leejet/stable-diffusion.cpp/actions/runs/38065961786. So you'd need to build it locally to use taeqi2_1

CMay · 2026-10-11T21:47:41Z

@CMay I think that's because the lastest published release is master-956-1b0ba10, on commit 1b0ba10, which predates this PR. The release build implementing this feature seems to have been canceled: https://github.com/leejet/stable-diffusion.cpp/actions/runs/38065961786. So you'd need to build it locally to use taeqi2_1

Thanks! Actually, the builds succeeded there so I downloaded the build artifact for my backend and taeqi2.1 now works. :)

https://github.com/leejet/stable-diffusion.cpp/actions/runs/38065961786#artifacts

feat: add TAEQI2.1 support for Qwen Image 2.1

690d504

leejet merged commit 2988060 into master Oct 10, 2026
11 checks passed

leejet deleted the feat/taeqi2-1-support branch October 10, 2026 16:01

Navigation Menu

Repository navigation

feat: add TAEQI2.1 support for Qwen Image 2.1 - #2123

feat: add TAEQI2.1 support for Qwen Image 2.1#2123
leejet merged 1 commit into
masterfrom
feat/taeqi2-1-support

leejet commented Oct 10, 2026 •

edited by coderabbitai Bot

Loading

Uh oh!

coderabbitai Bot commented Oct 10, 2026 •

edited

Loading

Walkthrough

Changes

❌ Failed checks (1 warning)

Uh oh!

Uh oh!

stduhpf commented Oct 11, 2026

Uh oh!

CMay commented Oct 11, 2026

Uh oh!

stduhpf commented Oct 11, 2026

Uh oh!

CMay commented Oct 11, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

leejet commented Oct 10, 2026 • edited by coderabbitai Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Related Issue / Discussion

Additional Information

Checklist

Summary by CodeRabbit

Uh oh!

coderabbitai Bot commented Oct 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Walkthrough

Changes

❌ Failed checks (1 warning)

Uh oh!

Uh oh!

stduhpf commented Oct 11, 2026

Uh oh!

CMay commented Oct 11, 2026

Uh oh!

stduhpf commented Oct 11, 2026

Uh oh!

CMay commented Oct 11, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

leejet commented Oct 10, 2026 •

edited by coderabbitai Bot

Loading

coderabbitai Bot commented Oct 10, 2026 •

edited

Loading