Quickstart

Three lines

from nobg import AutoModel

model = AutoModel.from_pretrained("feyninc/FeyNobg")
model.process("input.jpg").save("output.png")

process runs the whole pipeline — load, preprocess, forward under no_grad in eval mode, post-process, composite — and builds the processor the model's own config implies, so there is nothing else to load. The result is an RGBA PIL.Image at the input's original resolution.

image takes anything loadimg accepts: a path, a URL, a base64 string, a numpy array or a PIL image.

Several images

Pass a list, get a list back — each matte at its own original resolution:

for cut, path in zip(model.process(["a.jpg", "b.jpg"]), ("a.png", "b.png")):
    cut.save(path)

Useful keywords:

alpha = model.process("input.jpg", return_type="alpha")  # (H, W) tensor in [0, 1]

With an explicit processor

predict is the same call with the processor passed in. Reach for it when you already have one, or when a checkpoint's preprocessor_config.json disagrees with its config.image_size — process trusts the model config.

from nobg import AutoProcessor

processor = AutoProcessor.from_pretrained("feyninc/FeyNobg")
model.predict(processor, "input.jpg").save("output.png")

The processor comes first, then the inputs that vary: predict(processor, image, prompt, boxes), each optional after image. BiRefNet takes neither prompt nor boxes; SAM3 takes both. model.default_processor() returns the one process would build, if you want it without the Hub round-trip.

Driving the steps yourself

When you need the intermediates — the matte before compositing, the model outputs, the tensors on a particular device:

import torch
from loadimg import load_img

image = load_img("input.jpg").convert("RGB")
inputs = processor(image, return_tensors="pt")

with torch.no_grad():
    outputs = model(pixel_values=inputs["pixel_values"])

alpha = processor.post_process_alpha_matting(
    outputs, target_sizes=[(image.height, image.width)]
)[0]
processor.cutout(image, alpha).save("output.png")

post_process_alpha_matting takes one (height, width) per image, so a batch comes back at each image's original resolution. See Batched inference.

Text-promptable cutouts

FeyNobg finds the foreground subject. MultiMatte lets you say which one:

from nobg import AutoModel

model = AutoModel.from_pretrained("feyninc/multimatte")
model.process("input.jpg", "the dog").save("dog.png")

With no prompt it falls back to "the main foreground subject", which is what makes prompt-free background removal work. See Text and box prompts.

Next steps

Or try it in the browser first: 🤗 FeyNobg Space.