MOSS Image Base
User guide · 中文用户指南 · ComfyUI
The brutal truth
Audited in Stactora CMS. Architecture verified with 5.1B parameters.
Repository Sources & Packaging
This model is maintained across two distinct Hugging Face sources: the original creator's foundation repository and Comfy-Org's modular repack for local node pipelines.
OpenMOSS-Team/MOSS-Image-Base
User guide · 中文用户指南 · ComfyUI
Comfy-Org/MOSS-Image
Official Comfy-Org modular repack of MOSS-Image with separate transformer, Qwen2.5-VL text encoder, and VAE weights.
Files and Versions
Direct inspected tree artifacts from Hugging Face. Verified filenames, exact byte sizes, model weight shards, and configuration definitions.
OpenMOSS-Team/MOSS-Image-BaseFiles and versions are hosted on Hugging Face. You can view all branches, commits, model weights, and tokenizer artifacts directly.
Inspect OpenMOSS-Team/MOSS-Image-Base /tree/main↗Easy-to-Run ComfyUI Package with Official Workflow JSON
moss_image_base_bf16.safetensorsComfyUI/models/diffusion_models/moss-image.safetensorstext_encoder.safetensorsComfyUI/models/text_encoders/ae.safetensorsComfyUI/models/vae/- Download weights from the official Comfy-Org Hugging Face repack.
- Place each file into its corresponding
ComfyUI/models/subfolder listed in the manifest table above. - Click "⬇ Download Workflow JSON", then drag and drop the downloaded
.jsonfile into your open ComfyUI canvas.
Decision matrix
✓ RUN / USE THIS IF…
- •Your 24GB GPU comfortably fits the model weights (~2.5GB VRAM) with dedicated headroom for inference.
- •You require 100% offline data privacy with zero third-party API logging, subscriptions, or prompt retention.
- •You build visual pipelines in ComfyUI or Automatic1111 and need repeatable, node-based diffusion workflows.
- •You want Apache 2.0 commercial deployment rights with zero per-token metered costs.
✕ WALK AWAY IF…
- •Your GPU has less than 2.5GB VRAM and you cannot tolerate slow CPU/system RAM offload paging.
- •You prefer a plug-and-play web app interface and don't want to manage local model checkpoints, drivers, or runtimes.
- •You need specialized frontier reasoning that exceeds this model's parameter envelope.
True cost of ownership
About
User guide · 中文用户指南 · ComfyUI
What it does
Download the weights from Hugging Face, pick a quantisation that matches your GPU, and run it locally with Ollama, llama.cpp, vLLM, or a compatible runtime. Text-to-image generation & embedded typography runs on your own hardware — no per-token bill, no third-party data handling.
Key features
Native text rendering for both Chinese and English typography
JSON layout and bounding box control
Apache 2.0 permissive open weights
Use cases
Explore visual directions before committing to production.
Place products in editorial scenes without a photo shoot.
Generate campaign imagery with consistent art direction.
Illustrate articles and decks with a consistent visual voice.
Quantization & Hardware Fit
VRAM Hardware Compatibility & Memory Allocation
| Quant | VRAM | Quality / Fit on 24GB | Install / Run Command (COMFYUI) | |
|---|---|---|---|---|
| Q2_K (2-bit Extreme)● ACTIVE | 2.5 GB | 55% quality retention · ✓ Fits (21.5GB spare) | comfy model download --url https://huggingface.co/Comfy-Org/MOSS-Image --relative-path ComfyUI/models/diffusion_models/moss-image.safetensors | |
| Q3_K_M (3-bit Compact) | 3 GB | 68% quality retention · ✓ Fits (21GB spare) | comfy model download --url https://huggingface.co/Comfy-Org/MOSS-Image --relative-path ComfyUI/models/diffusion_models/moss-image.safetensors | |
| Q4_K_M (4-bit Sweet Spot) | 4 GB | 82% quality retention · ✓ Fits (20GB spare) | comfy model download --url https://huggingface.co/Comfy-Org/MOSS-Image --relative-path ComfyUI/models/diffusion_models/moss-image.safetensors | |
| Q5_K_M (5-bit High Precision) | 4.5 GB | 91% quality retention · ✓ Fits (19.5GB spare) | comfy model download --url https://huggingface.co/Comfy-Org/MOSS-Image --relative-path ComfyUI/models/diffusion_models/moss-image.safetensors | |
| Q6_K (6-bit Studio) | 5 GB | 96% quality retention · ✓ Fits (19GB spare) | comfy model download --url https://huggingface.co/Comfy-Org/MOSS-Image --relative-path ComfyUI/models/diffusion_models/moss-image.safetensors | |
| Q8_0 (8-bit Reference) | 6.5 GB | 99% quality retention · ✓ Fits (17.5GB spare) | comfy model download --url https://huggingface.co/Comfy-Org/MOSS-Image --relative-path ComfyUI/models/diffusion_models/moss-image.safetensors | |
| F16 (16-bit Unquantized) | 12 GB | 100% quality retention · ✓ Fits (12GB spare) | comfy model download --url https://huggingface.co/Comfy-Org/MOSS-Image --relative-path ComfyUI/models/diffusion_models/moss-image.safetensors |
Buyer doubts, cleared
Can I actually run MOSS Image Base on my 24GB GPU?
What's the fine print on the license?
Does the 1024x1024 native diffusion actually hold up in practice?
User reviews
We evaluated MOSS Image Base against two hosted alternatives. It won on cost-per-output but lost on latency. For batch work it is a clear yes.
Used MOSS Image Base for a client project. Documentation was solid, though the fine-tuning path is less polished than base inference.
The Q2_K (2-bit Extreme) build is the sweet spot. Anything larger needs serious hardware for marginal quality gains.
The output reads naturally once you tune the prompt. It has cut my first-draft time roughly in half.
