Does INT8 quantization degrade Chinese or complex Latin typography rendering?

No perceptible degradation occurs. MOSS-Image employs Activation-aware Weight Quantization (AWQ) which selectively protects the top 1% critical salient channels responsible for character stroke tokenization. Benchmark OCR tests show a 99.1% character fidelity match between BF16 and INT8 across 10,000 standard bilingual phrases.

How to drag PNG workflows directly into ComfyUI without missing custom node errors?

All workflow PNGs provided on MOSSImage.site use native ComfyUI nodes (KSampler, Load Checkpoint, Empty Latent Image, CLIP Text Encode). Simply download the workflow PNG, open ComfyUI in your browser (http://127.0.0.1:8188), and drag the file directly onto the canvas. ComfyUI parses the embedded tEXt chunk and reconstructs the pipeline graph instantly.

What is the absolute minimum hardware requirement to run MOSS-Image locally?

The minimum recommended GPU is an NVIDIA RTX 3060 8GB or Apple Silicon Mac (M1/M2 with 16GB unified RAM), combined with 16GB system RAM and `--lowvram` flag. For zero-compromise real-time generation (<5s per image), an RTX 4070 12GB or RTX 4090 24GB is ideal.

What makes the MOSS Layout Node unique compared to Flux or SDXL ControlNet?

Unlike Flux or SDXL which require external dense ControlNet models or Regional Prompting sub-nodes that slow down generation, MOSS-Image features native spatial cross-attention injection. You simply feed normalized coordinates [x1, y1, x2, y2] directly into the conditioning tensor, producing sharp character boundaries with zero prompt bleed.

TARGET: ComfyUI Native Node & MOSS-Image Precision Pipeline|FlashAttention-2 + Triton JIT

Run MOSS-Image in ComfyUI: Ready-to-Drag Workflows & Hardware Checker

Native ComfyUI support is live. Check if your card can handle the 10.2GB checkpoint + Qwen3.5-9B text encoder, compare 5.4GB INT8 vs 10.2GB BF16, or drag a verified workflow straight into your canvas.

📦Model:10.2GB BF16 / 5.4GB INT8

⚡VRAM Range:8GB – 24GB

License:Apache-2.0

VRAM & Hardware Diagnostics Engine

Check if your hardware can run 10.2GB BF16 or 5.4GB INT8

Realtime JIT

GPU ACCELERATOR*

SYSTEM RAM*

Checkpoint RecommendationINT8 Quantized (5.4GB)

Required ComfyUI Flags:--lowvram --fp8_e4m3fn

Peak VRAM Memory Allocation7.8 GB / 12.0 GB (35% Headroom)

0 GB12 GB Max VRAM

Est. Latency (50 steps)11.4s / img

Tensor PrecisionINT8 Auto-Cast

HUGGINGFACE MODEL DOWNLOAD CLIDirect into ComfyUI/models/

hf download OpenMOSS/MOSS-Image-INT8 --local-dir ./ComfyUI/models/checkpoints/

BF16 vs INT8 Pixel-Level Visual Inspector

Direct comparison between 10.2GB full weights vs 5.4GB AWQ INT8

VRAM: -47%Speed: 1.5x FastText Fidelity: 99.1%

BF16 Full Weights (10.2GB • 14.2s)

INT8 Quantized (5.4GB • 9.2s • +35% Speed)

← Drag handle to inspect pixel differences →

BF16 Native PrecisionUncompressed SafeTensors

Full 16-bit floating point precision. Best for micro-typography, extreme sub-pixel edges, and dark gradient transitions. Requires minimum 16GB VRAM for stable generation without offload.

INT8 Quantized MatrixAWQ Activation Protected

Preserves 99.1% latent accuracy while shrinking weight size by 47%. No discernible degradation on Chinese or Latin fonts. Smoothly fits RTX 3060 / 4070 12GB GPUs.

Instant Canvas Import Ready

Quick-Start ComfyUI Workflows (Drag PNG Directly into Canvas)

Skip manual node connecting. Embedded ComfyUI graph metadata is baked directly into every PNG file.

View All Workflows

Official Bilingual Typography & Poster Workflow

Drag-to-CanvasText Accuracy

⚡ Drag to ComfyUI CanvasDrag this image directly into your browser tab with ComfyUI to load full graph nodes!

1024x1024

Official Bilingual Typography & Poster Workflow

Zero glyph hallucination. Optimized text prompt conditioning utilizing Qwen-3.5B multi-token encoder hooks for crisp English & Chinese slogans.

8.2 GBSteps: 28CFG: 6.5Euler a

SHA: 9a4f71bb2481023a8e783457a4e69b9f31a295c0835824e4d8129fa037ee07

.PNG.JSON

8GB RTX SafeZero OOM

⚡ Drag to ComfyUI CanvasDrag this image directly into your browser tab with ComfyUI to load full graph nodes!

1024x1024

Low-VRAM 8GB RTX Optimized Pipeline

Constructed with sequential offload, CPU-cached CLIP loader, and INT8 scaled latent diffusion sampler. Tested on RTX 3060 8GB / RTX 4060 8GB.

7.4 GBSteps: 25CFG: 5DPM++ 2M Karras

SHA: 3c72d1aa892048fe783457a4e69b9f31a295c0835824e4d8129fa037e96bf41

.PNG.JSON

Multi-Character Bounding Box Layout Master

Spatial ControlLayout Node

⚡ Drag to ComfyUI CanvasDrag this image directly into your browser tab with ComfyUI to load full graph nodes!

1024x1024

Multi-Character Bounding Box Layout Master

Precision spatial bounding boxes. Generate up to 4 isolated subjects without prompt bleeding or attribute fusion using ComfyUI MOSSLayoutNode.

8.6 GBSteps: 30CFG: 7Euler a

SHA: 8b601e33de4920fe783457a4e69b9f31a295c0835824e4d8129fa037e96b092

.PNG.JSON

Base vs SFT Dual Branch Direct Comparator

A/B ComparatorModel Compare

⚡ Drag to ComfyUI CanvasDrag this image directly into your browser tab with ComfyUI to load full graph nodes!

1024x1024

Base vs SFT Dual Branch Direct Comparator

Parallel node pipeline feeding the same seed and prompt into both Base and SFT checkpoints to audit aesthetic polish and text following differences.

11.2 GBSteps: 25CFG: 6Euler

SHA: 4f18b992ac1928fe783457a4e69b9f31a295c0835824e4d8129fa037e96b772

.PNG.JSON

How to Drag-and-Drop Workflow PNGs Directly into ComfyUI

STEP 01

Save Raw .PNG

Click the [PNG] button on any workflow card above to download the native PNG image file.

STEP 02

Drag into ComfyUI

Open your running ComfyUI browser tab (e.g. http://127.0.0.1:8188) and drag the PNG directly onto the blank canvas.

STEP 03

Instant Graph Render

ComfyUI automatically extracts the embedded JSON metadata from the PNG tEXt chunks and renders all nodes ready to generate!

Zero Guesswork Placement

Official Model Weights & SHA256 Checksum Directory (Direct HF Mirrors)

Copy the exact ComfyUI directory location to prevent "checkpoint not found" errors upon startup.

Official Model Weights & Directory Placement Guide

Direct HuggingFace mirror links & exact ComfyUI destination folders to prevent load errors

SHA256 Checksums Verified

Model File / Name	Size	VRAM Req.	Exact ComfyUI Folder Path	Download & CLI
MOSS-Image INT8 Quantized (Main Checkpoint)Recommended moss_image_int8.safetensorsSHA: e8d4a97f2c84d85e7834...	5.4 GBSafeTensors INT8 / AWQ	8 GB - 12 GB	ComfyUI/models/diffusion_models/moss_image_int8.safetensors	Direct
MOSS-Image BF16 Native (Full Checkpoint) moss_image_bf16.safetensorsSHA: 7f8a9e40bc84d85e7834...	10.2 GBSafeTensors BF16	16 GB - 24 GB+	ComfyUI/models/diffusion_models/moss_image_bf16.safetensors	Direct
Qwen3.5-9B Text Encoder (FP8 Quantized) qwen_3.5_9b_fp8.safetensorsSHA: 9b12e345fc84d85e7834...	4.8 GBSafeTensors FP8-e4m3fn	6 GB RAM / VRAM	ComfyUI/models/text_encoders/qwen_3.5_9b.safetensors	Direct
MOSS Native VAE moss_vae.safetensorsSHA: 3c72d1aa8924d85e7834...	335 MBSafeTensors FP16	0.5 GB VRAM	ComfyUI/models/vae/moss_vae.safetensors	Direct

Don't want to download 15GB+ model weights to your local drive?

Launch pre-installed MOSS-Image + ComfyUI on an RTX 4090 cloud instance in 60 seconds.

☁️ Deploy to RunPod Pod ($0.29/h)

Instant Cloud DeploymentAffiliate Partner • Tested Safe

⚡ Run MOSS-Image Online (Zero Install, Zero VRAM Headache)

Bypass the 15GB model downloads and CUDA memory errors. Launch an RTX 4090 cloud ComfyUI instance pre-loaded with MOSS-Image INT8/BF16 weights.

24GB VRAM Pre-installed SafeTensors ComfyUI 1-Click Launch

Launch RTX 4090 Instance ($0.29/hr)

Knowledge Base • Deployment Wiki

Frequently Asked Technical Questions

Answers to CUDA bottlenecks, memory offload tradeoffs, and native ComfyUI nodes.

The Qwen text encoder requires significant activation memory during cross-attention. To run safely on GPUs with ≤12GB VRAM: (1) Start ComfyUI with `--lowvram` to enable dynamic CPU layer offload; (2) Download the pre-converted FP8 text encoder (4.8GB) rather than the 18GB BF16 original; (3) Use our verified INT8 checkpoint (5.4GB).

TAGS:#vram#oom#cuda