Skip to content
Images

DeFLOCNet

sketch- and color-guided image inpainting

2 min readupdated

DeFLOCNet handles image inpainting (filling in and restoring missing or masked-out regions) at a 256 × 256 resolution. Unlike standard mask-only filling, the model lets you guide contours with a sketch and specify target hues using color strokes. KumaPower's repository provides two checkpoints built on the sd_color_v2 architecture—one for close-up portraits and another for everyday scenes. The tool works well for targeted retouching, editing facial details, and refining compositions on small images.

Key features

  • Two specialized weights: deflocnet-hdtf-step10000.pt is trained on HDTF face crops (10,000 steps), while deflocnet-general-step23500.pt is trained on a mixed dataset of Places2, Unsplash, T2I, and HDTF (23,500 steps).
  • Triple guidance: The canvas combines a white mask (the generation region), black sketch lines, and sparse color hints.
  • Original pixel preservation: Pixels outside the white mask are copied directly from the 256 × 256 source image.
  • Inference-ready: The files are FP32 PyTorch generator checkpoints ready to run locally.

What you can do

  • Retouch and edit portraits: Use the portrait checkpoint with face auto-framing to redraw features or close-up details along custom contours.
  • Edit landscapes and architecture: Apply the general checkpoint to replace objects in building and nature photos while controlling the silhouette and palette.
  • Compare results in the studio: Test both checkpoints on the same image, mask, and seed in the DeFLOCNet Studio web demo before integrating locally.

Example

Upload an image to DeFLOCNet Studio, pick Auto-frame portrait (to automatically crop a single face) or Use full image (for general scenes and pre-cropped portraits), paint a white mask over the target area, and add a black sketch with color strokes—the model generates the missing region and blends it with the untouched source pixels.

Read moreDetails
Architecture: sd_color_v2 (the DeFLOCNet method was originally presented at CVPR 2021).
Format and resolution: FP32 generator checkpoints, fixed 256 × 256 input and output size.
Training: 10,000 steps (HDTF portrait) and 23,500 steps (General scenes).
Instructions for running locally via the command line are available in the author's GitHub repository.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.3 GB · ready files1.8 GB
  • FP8 / INT8recommendedweights 0.2 GB1.7 GB
  • 4-bitweights 0.1 GB1.6 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # KumaPower/DeFLOCNet
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download KumaPower/DeFLOCNet --local-dir ~/models/DeFLOCNet
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to DeFLOCNet
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • SageAttention 3v2.2.0from October 28, 2025

    even faster on RTX 50 / Blackwell

    FP4 attention for Blackwell. Faster than SageAttention 2, but the authors suggest v2 when accuracy matters.

    How to enable: Build from the sageattention3_blackwell folder of the repo; in ComfyUI via KJNodes.

    Earlier generations (2)
    • SageAttention 2++ — Quantised (INT8/FP8) attention with no visible quality loss. The main speed-up for image and video generation.
    • SageAttention 1 — First version: INT8 for Q·K. Still works, but 2++ is faster and more accurate.
  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (3)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
    • xFormers — The older way to speed up attention, works almost everywhere. Sage or Flash are faster where available.
  • cache-dit (DBCache)v1.5.6from November 20, 2026new release

    1.5–3× at nearly the same quality

    Reuses block outputs between diffusion steps. Supports most DiT models.

    How to enable: pip install cache-dit; one line cache_dit.enable_cache(pipe) for diffusers.

    Earlier generations (2)
    • MagCache — Skips steps by magnitude of change. More accurate than TeaCache, no calibration.
    • TeaCache — The first popular diffusion step cache. Works, newer methods are more accurate.
  • Nunchaku (SVDQuant 4-bit)v1.2.1from January 25, 2026

    ≈3× faster and 3.5× less memory

    4-bit versions of popular image models with almost no quality loss (FP4 on RTX 50).

    How to enable: Ready weights for supported models; in ComfyUI use ComfyUI-nunchaku.

  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of DeFLOCNet (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

KumaPower/DeFLOCNet
Author
KumaPower
Category
Images
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments