Models · Edges and lines

AnyLine line art

AnyLine (TheMisto.ai): sharp line maps for line-conditioned ControlNets such as MistoLine, from the MTEED edge network blended with a line art of the image. MIT code; OpenRAIL++-M weights (commercial use with use restrictions).

experimental Fork of huggingface/controlnet_aux@b1aac31 · backends: portable, kernels

Checkpoint anyline

Paper 2023 Added in 0.2.0

Portable, RTX 5060 Ti 16 GB 50.7 ms 1080x1920, fp16
vs the faster original 11.5× original fp16: 585 ms
Peak memory 654 MiB original: 909 MiB
Accuracy vs original fp32 – no test-set run

Use it

pip install fastforks

Lean install with only what this model needs: pip install fastforks-core

from fastforks import edges

model = edges.load("anyline")  # kernels on a supported GPU, else portable PyTorch
result = model(image)
result.edges
result.backend  # what ran: mode, device, dtype, kernels
fastforks bench anyline

Times the model on this machine. load(…, mode="portable") or mode="kernels" picks a backend; usage guide.

Speed and memory

Every backend on one GPU (RTX 5060 Ti 16 GB) and the same input, the authors’ code included. Kernels runs fp16. Portable runs fp16 on CUDA.

BackenddtypeTime per call (ms)vs original fp16Peak memory
allocated / reserved
Load (s)First call (s)
Original fp32fp32not measured ––––
Original fp16torch.autocast(float16) around the call; huggingface/controlnet_aux@b1aac318b15c anyline, host OpenCV resize, float post-processingfp16 585–909 MiB/1,002 MiB0.10.59
Portable fp32fp32not measured ––––
Portablefp16 50.712×654 MiB/722 MiB0.20.05
Kernelsnot measured ––––
Portable + torch.compilenot measured here ––––
TensorRTnot measured here ––––

Measured on RTX 5060 Ti 16 GB, torch 2.14.1+cu130, 2026-11-14. First call, 3 warm-up, peak reset, median of 3 rounds x median of 5. Ratios divide two times of this table (same GPU, same input). Original fp16 is the authors’ own precision option, else torch.autocast. torch.compile and TensorRT: build time instead of load time. TensorRT memory (engine, not comparable) is the GPU memory the engine took, read from the driver because TensorRT allocates outside PyTorch, so it can't be compared with the PyTorch peaks of the other rows.

Other devices

DeviceBackenddtypems / callPeak memoryOutput vs referenceInput
Apple M1 Pro 16 GB MPS, torch 2.14.1, 2026-11-14Originalfp32259 1,557 MiB process–1024x1024
Portablefp16117 1,319 MiB processmatches1024x1024
Portablefp32130 2,346 MiB processmatches1024x1024
Kernelsfp32not implemented–1024x1024

CUDA memory is torch’s peak allocated / reserved; Apple has no such counter, so the Mac rows show the process footprint (weights, Python and torch included). “Output vs reference” compares the device’s output with the stored golden output under the card’s tolerance; – = not checked on that device. Kernels “not implemented” = no kernels for this model on that backend yet: it runs portable there.

Licence and weights

Commercial use allowed

Code
MIT LICENSE
Weights
LicenseRef-CreativeML-OpenRAIL-PlusPlus-M source
Training data
Not stated for MTEED (TheMistoAI/MistoLine `Anyline/MTEED.pth`, the TEED network).
  • The MTEED weights use the OpenRAIL++-M licence: commercial use is allowed but with use restrictions that must be passed on; the training data terms are not checked.

Our reading of the licences, not legal advice.

Weights

MTEED.pth 248 KB· hosted on Hugging Face

fastforks downloads and checks these files on first use (fastforks download anyline).

Accuracy

The test set through the authors’ code and through ours: same checkpoint, same protocol. The distance between the original’s fp32 and fp16 runs is what half precision alone costs; our runs should stay within it.

Accuracy not measured for anyline.

Closeness to the original

Beyond the test-set scores: the raw output of ours against the original’s fp32 output on the same input, pixel by pixel. CPU, float32: max |diff| <= 0.002 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 0.02 and p99 <= 0.25 (absolute error of the output edges).

CheckToleranceMeasured
CPU fp32, max |diff|≤ 0.002 not recorded
cpu_outliers≤ 0.0001 not recorded
GPU default dtypes, mean |diff|≤ 0.02 0.00008
GPU default dtypes, p99 |diff|≤ 0.25 0.00002

Recorded for anyline on RTX 5060 Ti, torch 2.10.0+cu130, 2026-11-17, with these kernels: edge_components

Details

AnyLine preprocessor of MistoLine (TheMisto.ai; controlnet_aux `AnylineDetector`, ComfyUI-Anyline): the MTEED edge network (TEED architecture, weights by TheMisto.ai) blended with a Gaussian-difference line art whose small specks are removed; sharp line maps for line-conditioned ControlNets.

Input / output
Input: one RGB image. Output: a line / edge map [H, W] in [0, 1] at the input size (1 = line), the ControlNet preprocessors (line art, HED, PiDiNet, TEED / Anyline); M-LSD also returns straight line segments.
Input size
any H x W >= 32, padded to a multiple of 32; tested up to 1080x1920 (654 MiB allocated)
Options
  • short_side=1280>= 64· changes the output
Outputs
edges
CUDA kernels
  • edge_components
Apple (MPS)
fp32, fp16

Limitations

  • Layers are blended in float as TheMistoAI/ComfyUI-Anyline; controlnet_aux's `AnylineDetector` blends uint8 arrays, where `1 - (1 - top) * (1 - base)` wraps around, so its maps differ wherever line art is drawn.
Full reference, generated from the repository

AnyLine (TheMisto.ai): sharp line maps for line-conditioned ControlNets such as MistoLine, from the MTEED edge network blended with a line art of the image. MIT code; OpenRAIL++-M weights (commercial use with use restrictions).

Commercial use: yes.

variantinputbackendtime (ms)peak memory (MiB, allocated / reserved)
anylinesynthetic (1080x1920)original
portable
kernels
585
50.7
–
909 / 1002
654 / 722
–
anylinesynthetic (1024x1024)original
portable
kernels
portable + torch.compile
TensorRT fp16
299
24.0
–
11.6
12.9 (full line map only; memory estimated)
509 / 562
362 / 384
–
306 / 346
528 / 542
Details
  • Original code: https://github.com/huggingface/controlnet_aux at b1aac318b15c
  • Code licence: MIT (permissive); files: LICENSE
  • Weights licence: LicenseRef-CreativeML-OpenRAIL-PlusPlus-M
  • Licence evidence: https://github.com/TheMistoAI/ComfyUI-Anyline/blob/40d7b290fe829d50e4e12e7f34498c2d30913b53/LICENSE ; https://huggingface.co/TheMistoAI/MistoLine/blob/1d9d0b3d48b295cf70d80a0f839a3055672f8393/README.md
  • Training data: Not stated for MTEED (TheMistoAI/MistoLine Anyline/MTEED.pth, the TEED network).
  • Licence note: The MTEED weights use the OpenRAIL++-M licence: commercial use is allowed but with use restrictions that must be passed on; the training data terms are not checked.
  • Options: short_side = 1280 (>= 64; also resolution; also detect_resolution)
  • Kernel libraries: edge_components
  • Variants: anyline
  • Outputs: edges
  • Input size: any H x W >= 32, padded to a multiple of 32
  • Tested up to (portable, fp16, RTX 5060 Ti 16 GB): anyline 1080x1920 (0.6 GiB peak allocated, 0.7 reserved)
  • anyline 1024x1024, first call: portable + torch.compile 40 s (cold Inductor cache), TensorRT fp16 30 s (ONNX + engine).
  • anyline original 1080x1920, original 1024x1024: torch.autocast(float16) around the call; huggingface/controlnet_aux@b1aac318b15c anyline, host OpenCV resize, float post-processing
  • anyline load / first call: original 0.11 s / 0.622 s, portable 0.17 s / 0.053 s
  • Protocol, first call, 3 warm-up, peak reset, median of 3 rounds x median of 5

Parity

CPU, float32: max |diff| <= 0.002 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 0.02 and p99 <= 0.25 (absolute error of the output edges).

variantkernels error vs original, mean / p99GPU, torchkernels used
anyline8e-05 / 2e-05RTX 5060 Ti, torch 2.10.0+cu130edge_components@d6becc089ab3:sm_120

Limitations

  • Layers are blended in float as TheMistoAI/ComfyUI-Anyline; controlnet_aux's AnylineDetector blends uint8 arrays, where 1 - (1 - top) * (1 - base) wraps around, so its maps differ wherever line art is drawn.

Example output

No example rendered for this checkpoint yet.