← Back to AI Insights
Gemini Executive Synthesis

Incompatibility between MOSS-Image's ComfyUI loader and Comfy-Org checkpoints due to fused vs. split QKV projections.

Technical Positioning
Seamless integration with ComfyUI, checkpoint compatibility, and efficient model loading.
SaaS Insight & Market Implications
This issue details a critical integration failure between MOSS-Image's ComfyUI loader and its own Comfy-Org checkpoints, stemming from a fundamental mismatch in QKV projection handling (split vs. fused). This `AttributeError` directly prevents users from utilizing the model within a popular workflow environment. Such compatibility issues are significant developer pain points, leading to immediate operational blockers and eroding trust in the product's robustness. The detailed root cause analysis and environment specifics provided by the user indicate a sophisticated user base expecting seamless interoperability. Market implication: In the B2B SaaS AI space, integration friction with established tools and frameworks is a major adoption deterrent. Products must ensure robust, tested compatibility across common deployment environments to secure enterprise usage.
Proprietary Technical Taxonomy
MOSSImageUNETLoader native single-file loader Comfy-Org/MOSS-Image single-file checkpoints AttributeError: 'NoneType' object has no attribute 'dtype' _Block proj.q proj.k

Raw Developer Origin & Technical Request

Source Icon GitHub Issue Oct 19, 2026
Repo: OpenMOSS/MOSS-Image
ComfyUI native single-file loader fails on all Comfy-Org checkpoints: _Block expects split q/k/v but checkpoints ship fused proj.qkv (AttributeError: 'NoneType' object has no attribute 'dtype')

## Summary

`MOSSImageUNETLoader` (the native single-file loader) **cannot load any of the four `Comfy-Org/MOSS-Image` single-file checkpoints**. It fails immediately with:

```
AttributeError: 'NoneType' object has no attribute 'dtype'
```

**Root cause:** `_Block` in `comfyui/moss-image/model.py` declares **separate** `proj.q` / `proj.k` / `proj.v` / `proj.o` linears, but every Comfy-Org repackaged checkpoint stores a **fused** `proj.qkv.weight` `[3*dim, dim]`. Nothing populates `proj.q/k/v`, so under ComfyUI's DynamicVRAM lazy init they stay `None` and line 96 blows up.

Note this is **inconsistent with `_TextBlock` in the same file**, which already uses fused `self.qkv` — so it looks like a simple oversight, not an intentional design difference.

## Environment

| | |
|---|---|
| ComfyUI | `0.39.0` (`d49e888`) |
| comfy-aimdo | present → `DynamicVRAM` enabled (`comfy.memory_management.aimdo_enabled = True`) |
| Custom node | `comfyui/moss-image` copied to `custom_nodes/moss-image/` |
| Python | 3.12 |
| GPU | 24 GB VRAM laptop (CUDA 13.x, torch 2.14.1+cu130) |
| Repo revision | `95559c9` (docs: use official arXiv BibTeX citation) |

## Steps to reproduce

1. Download `Comfy-Org/MOSS-Image` → `diffusion_models/moss_image_base_int8_convrot.safetensors` (5.41 GB), `text_encoders/qwen3.5_9b_w4a8_gptq.safetensors`, `vae/flux2-vae.safetensors` into the standard `ComfyUI/models/{diffusion_models,text_encoders,vae}/` folders.
2. Load the node **MOSS-Image UNET Loader** (`MOSSImag...

Developer Debate & Comments

No active discussions extracted for this entry yet.

Adjacent Repository Pain Points

Other highly discussed features and pain points extracted from OpenMOSS/MOSS-Image.

Extracted Positioning
Ease of installation and execution of MOSS-Image within ComfyUI.
User-friendly deployment, simplified integration, and reduced setup complexity.
Extracted Positioning
Hosting MOSS-Image's LongLongTextBench (LLTB) and RealLayoutArena (RLA) datasets on Hugging Face.
Open-source community engagement, dataset discoverability, and standardized access via Hugging Face Hub.
Extracted Positioning
Detailed prompt for ultra-realistic image generation, focusing on cultural specificity and avoiding AI artifacts.
High-fidelity, culturally accurate, and artifact-free image generation.
Extracted Positioning
Model file download location management.
User control over resource storage.
Extracted Positioning
MOSS-Image's caption generation pipeline, prompt engineering, and multimodal text handling.
Transparency in annotation processes, prompt engineering best practices, and multimodal capability validation.

Frequently Asked Questions

Market intelligence mapped to Incompatibility between MOSS-Image's ComfyUI loader and Comfy-Org checkpoints due to fused vs. split QKV projections..

What is the technical positioning of Incompatibility between MOSS-Image's ComfyUI loader and Comfy-Org checkpoints due to fused vs. split QKV projections.?
Based on our AI analysis of the original developer request, its primary technical positioning is: Seamless integration with ComfyUI, checkpoint compatibility, and efficient model loading.
How is the developer community reacting to Incompatibility between MOSS-Image's ComfyUI loader and Comfy-Org checkpoints due to fused vs. split QKV projections.?
Yes, we have tracked 1 direct responses and active debates regarding this specific topic originating from GitHub Issue.
Which technical concepts are associated with Incompatibility between MOSS-Image's ComfyUI loader and Comfy-Org checkpoints due to fused vs. split QKV projections.?
Our proprietary extraction maps Incompatibility between MOSS-Image's ComfyUI loader and Comfy-Org checkpoints due to fused vs. split QKV projections. to adjacent architectural concepts including MOSSImageUNETLoader, native single-file loader, Comfy-Org/MOSS-Image, single-file checkpoints.

Engagement Signals

1
Replies
open
Issue Status

Cross-Market Term Frequency

Quantifies the cross-market adoption of foundational terms like GPU and _Block by tracking occurrence frequency across active SaaS architectures and enterprise developer debates.