Skip to content

MagicQuill

Intermediate

Development line: project:magicquill · thread magicquill-development
Last event: 2024-11-18 · 1 dated since 2024-11-18 · Researched: 2026-09-04 · confidence: medium

What it is

MagicQuill is a research image editor for artists who need local control beyond prompt-only inpainting.

  • V2 controls content, placement, structure, and color through visual layers, adding a local-edit brush, foreground props, and SAM segmentation.
  • V1 controls additions, subtractions, and color brush edits, with Draw&Guess suggesting prompts from sketches.

Development line

  • 2024-11-18 — MagicQuill's public demo, source repository, and hosted Space were linked. Project page, source code, and Hugging Face demo formed the public practitioner entry points.

What changed

  • 2024-11-14 — MagicQuill V1: the arXiv preprint defined the three-brush editor and Draw&Guess prompt assistant.
  • 2024-11-18 — MagicQuill V1: project page, source code, and Hugging Face demo formed the public practitioner entry points.
  • 2024-11-21 — MagicQuill V1: Windows path handling, save behavior, and .bat/.sh setup helpers were updated.
  • 2024-11-25 — MagicQuill V1: drag-and-drop image upload and a download button were added.
  • 2024-12-06 — MagicQuill V1: auto-save and resolution controls were added.
  • 2024-12-07 — MagicQuill V1: a ComfyUI custom node was released.
  • 2024-12-16 — MagicQuill V1: a ModelScope deployment was added.
  • 2025-01-02 — MagicQuill V1: a Docker container path was added.
  • 2025-02-24 — MagicQuill: a Pinokio launcher route was referenced; its historical package state cannot be reconstructed from the supplied endpoint.
  • 2025-02-27 — MagicQuill V1: the paper was accepted to CVPR 2025.
  • 2025-12-02 — MagicQuill V2: the layered-visual-cues paper was submitted to arXiv.
  • 2025-12-03 — MagicQuill V2: the released system moved from V1’s brush-only interaction to content, spatial, structural, and color layers on Flux Kontext.

How to use this

From 2024-11-18, practitioners could evaluate MagicQuill through its public demo and hosted Space, while using the linked repository to inspect the implementation before adoption.

  1. Use the running V2 Hugging Face Space when a local GPU with about 40 GB VRAM is unavailable. — https://huggingface.co/spaces/AI4Editing/MagicQuillV2
  2. For local V2, use the active zliucz/MagicQuillV2 repository, create a Python 3.10 environment, and install its requirements. — https://github.com/zliucz/MagicQuillV2
  3. Download LiuZichen/MagicQuillV2-models into models/, then start the UI with python app.py. — https://huggingface.co/LiuZichen/MagicQuillV2-models
  4. In V2, isolate a foreground prop with SAM, place it through the Visual Cue Manager, then use local-edit, edge, color, or removal controls for the relevant layer. — https://magicquill.art/v2/
  5. For an 8GB GPU, use V1: clone recursively so the LLaVA submodule is present, fetch the checkpoints, then run python gradio_run.py. — https://github.com/magic-quill/magicquill

Best practices

Superseded by this

  • 2025-12-03 — MagicQuill V2 supersedes V1 as the active capability line, but V1 remains the documented lower-VRAM path rather than being universally obsolete.
  • 2026-09-04 — The V2 README’s old clone target is stale: the observed magic-quill/MagicQuillV2 route returns 404; use the active zliucz/MagicQuillV2 repository.
  • 2026-09-04 — Pinokio should not be treated as official MagicQuill model guidance; its current installer is a third-party wrapper around a noncanonical fork.

Still unknown

  • The original 2025-02-24 Pinokio page content is not recoverable from the supplied URL today: it redirects to Pinokio’s general site, leaving installer revision and behavior on that date unconfirmed.
  • The V1 /demo/ route returned no inspectable page body, and its Hugging Face Space shows “Running on Zero”; live edit completion was not independently tested.
  • Pinokio is a launcher path rather than a second MagicQuill model, leaving its precise connection to the 2025-02-24 event unconfirmed.
  • V1 and V2 GitHub pages show no immutable software releases, so README update-log dates serve as the available release evidence.
  • The stated V1 and V2 hardware figures are author-reported requirements, not independent benchmarks.

Sources

source title read
https://magicquill.art/demo/ MagicQuill demo 2026-09-04
https://github.com/magic-quill/magicquill MagicQuill V1 repository 2026-09-04
https://huggingface.co/spaces/AI4Editing/MagicQuill MagicQuill V1 Hugging Face Space 2026-09-04
https://pinokio.computer/ Pinokio 2026-09-04
https://arxiv.org/abs/2411.09703 MagicQuill: An Intelligent Interactive Image Editing System 2026-09-04
https://github.com/pinokiofactory/MagicQuill Pinokio MagicQuill repository 2026-09-04
https://raw.githubusercontent.com/pinokiofactory/MagicQuill/main/install.js Pinokio MagicQuill installer script 2026-09-04
https://github.com/magic-quill/MagicQuillV2 MagicQuillV2 legacy GitHub clone route 2026-09-04
https://github.com/zliucz/MagicQuillV2 MagicQuill V2 repository 2026-09-04
https://arxiv.org/abs/2512.03046 MagicQuillV2: Precise and Interactive Image Editing with Layered Visual Cues 2026-09-04
https://magicquill.art/v2/ MagicQuill V2 project page 2026-09-04
https://huggingface.co/LiuZichen/MagicQuillV2-models MagicQuill V2 model card 2026-09-04
https://huggingface.co/spaces/AI4Editing/MagicQuillV2 MagicQuill V2 Hugging Face Space 2026-09-04

Agent brief

  • Subject: project:magicquill, thread magicquill-development, 1 dated events 2024-11-18 → 2024-11-18.
  • Practical note: From 2024-11-18, practitioners could evaluate MagicQuill through its public demo and hosted Space, while using the linked repository to inspect the implementation before adoption.
  • Confidence: medium. Dated supersedes above are the authority for what is obsolete.