Text-to-Image
Diffusers
Safetensors
MageFlowPipeline
ajh
mage-flow
mage-flow-nvfp4-ajh
nvfp4
blackwell
qwen3-vl
quantization
Instructions to use ajh-code/Mage-Flow-NVFP4-AJH with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use ajh-code/Mage-Flow-NVFP4-AJH with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("ajh-code/Mage-Flow-NVFP4-AJH", torch_dtype=torch.bfloat16, device_map="cuda") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- Draw Things
- DiffusionBee
Third-party notices
Microsoft Mage and Mage-Flow
This package targets microsoft/Mage-Flow and includes the minimal Python
inference sources under vendor/mage_flow/.
- Upstream: https://github.com/microsoft/Mage
- Pinned source commit:
df7f84d9f8fc991d189d929f03cff623b430a4a2 - License: MIT
- License copy:
LICENSEandlicenses/MAGE-MIT.txt
The Microsoft base checkpoint is not duplicated in this package. The
portable loader downloads the required transformer, VAE, scheduler, and
configuration files from microsoft/Mage-Flow, or accepts a local copy.
Qwen3-VL and the mixed NVFP4 checkpoint
The packaged text encoder derives from Qwen/Qwen3-VL-4B-Instruct and is an
exact copy of the learned-rounded mixed checkpoint published at
InsecureErasure/Qwen3-VL-4B-Instruct-NVFP4.
- Quantized upstream: https://huggingface.co/InsecureErasure/Qwen3-VL-4B-Instruct-NVFP4
- Pinned quantized revision:
cf080d6af667f5f4949295bd09a1f72a218fe0ee - Base model: https://huggingface.co/Qwen/Qwen3-VL-4B-Instruct
- License: Apache-2.0
- License copy:
licenses/QWEN-APACHE-2.0.txt - Packaged artifact SHA-256:
719906b435800757d22013d3d475a4853d59b779b022669fa8b8a193b85d0f41
The quantized checkpoint uses the ComfyUI comfy_quant representation and
executes packed projections through comfy-kitchen.