I created this custom node because i wasn't getting the results I needed for my usecase when using the ComfyUI-ToSVG node by Yanick112 and the Flux Text to Vector Workflow by Stonelax. A big thanks to them for their work and inspiring me to try my own node.

Not getting the results I want is NOT a complaint; ComfyUI-toSVG implements the VTracer logic, and this works great for multicolor and detailed images.
For logo's, text, etc. I found Potrace SVG conversion by Peter Selinger better suited, with the caveat that it only handles 2 colors; a Foreground and Background.
I also found Potrace optimizes shapes differently, creating fewer vector-points and needing almost to no cleanup when loading the SVG into vector design software.

Potracer to SVG node traces a raster image (IMAGE) into an SVG vector graphic using the 'potracer' pure Python library for POTRACE by Tatarize. (I may mix up the terms 'potrace' and 'potracer' at times)

Comparing Vtracer and Potrace

While each user and route will have their specific usecase, my usecase is creating designs for Vinylcutters and logos. This usecase requires sharp images, fluid shapes and clear separation of fore and background. It is also vital to have the lines and curves smooth, with as few vectors as possible while staying true to the form.


Artifacts	Vector points

Vtracer (Green, Top) shows significantly more artifacts, hard edges or straight lines, and irregular traces compared to Potracer (Blue, Bottom), which stays true to the original form.	Vtracer has significantly more vector points compared to the optimized Potracer version.

For usecases where Vtracer suits best, go check out https://github.com/Yanick112/ComfyUI-ToSVG/, where Yanick explains the workings in detail.

In short; As Potrace converts the image to 1 foregroundcolor and 1 backgroundcolor, it is pretty much unusable for any image requiring more than one color, especially photos.

Installation

Due to both usecases for SVG export are very valid when working with ComfyUI, I opted to keep the naming and category alike for easier navigation.

note: This node requires Requires LATEST VERSION of Yanick112's ComfyUI-ToSVG --> SaveSVG node to save the SVGfile and VectorToRaster node if you want to reconstruct the SVG to an image if you want to see the result in ComfyUI. See his repository for install instructions

note 2025-05-08: With the new API node a default ComfyUI node SaveSVG is built. This node conflicts with Yanick112's ComfyUI-ToSVG --> SaveSVG node. The new SaveSVG node is currently under the API nodes, but will eventually be part of base ComfyUI functionality. See this ComfyUI issue. I have duplicated and renamed Yanick112's SaveSVG node to keep compatibility with older versions. The node now also outputs attaching to the NEW SaveSVG node of ComfyUI v0.3.32 and up.

Using ComfyUI Manager --> Custom Node Manager --> Search for:
```
  ComfyUI-ToSVG-Potracer
  Comfyui-ToSVG
```

Using comfyregistry manual installation:

  comfy node registry-install ComfyUI-ToSVG-Potracer
  comfy node registry-install ComfyUI-ToSVG

Manual Installation (ToSVG-Potracer only)
1. Navigate to your /ComfyUI/custom_nodes/ folder.
2. Run the following command to clone the repository:
```
 git clone https://github.com/ImagineerNL/ComfyUI-ToSVG-Potracer
```
3. Navigate to your ComfyUI-ToSVG-Potracer folder.
  - Command for Portable/venv:
```
path/to/ComfUI/python_embeded/python.exe -s -m pip install -r requirements.txt
```
  - Command for system Python:
```
pip install -r requirements.txt
```
4. Visit https://github.com/Yanick112/ComfyUI-ToSVG for installation of ComfyUI-ToSVG

Intended Usage

The input image should only use 2 distinct colors (e.g. black and white). If other pixel values appear in the input, they will be converted to black and white using a simple threshold method. You can set the Foreground and Background colors for the svg.

Outputs svg strings as 1 flat shape (as a compound path). Should you want to adjust the shapes by hand, currently, when you release the path in vector design software, the 'internal openings' become filled shapes. This is easy to fix using the boolean tools and substract. My main usecase is Silhouette Studio for vinylcutting and especially with 'zero_sharp_corners' set to true, i don't need to do any postprocessing on the shape.

Parameter	Usage	Default
threshold	Brightness cutoff (0-255) for binarization to B/W.	128
input_foreground	Defines if the input image is a Black object on White background or White on Black background	Black on White
turnpolicy	How to resolve ambiguities in path decomposition	minority
turdsize	Suppress speckles of up to this many pixels	2
corner_treshold	Smaller values = sharper corners	1
zero_sharp_corners	Forces al corners to be fluid (sets corner_treshold = 1.34 in backend)	false
opttolerance	Curve optimization tolerance	0.2
optimize_curve	Curve optimization, joins adjacent Bezier curve segments where possible. Reduces filesize and points	true
foreground_color	Defines foreground color after trace #rrggbb	#000000
stroke_color	*Defines stroke color after trace #rrggbb / none (only used when stroke_width > 0.0)	#ff0000
stroke_width	*Sets a stroke width/outline for the traced shapes (only valid when stroke_color <> none)	0.0
background_color	Defines background color after trace #rrggbb / none	#ffffff
no_background	Removes the background color (sets background_color to 'none' in backend)	false
output_scale	Scale the output SVG string by a factor X for resizing	1.0

svg_strings	A set of strings that can be converted to svg shape by connecting it to the ComfyUI-ToSVG --> SaveSVG node

Workflows

1. Txt2Img to SVG Potracer Example Workflow


The above image is just a visualisation, does not contain workflow Workflow based on the Flux Text to Vector Workflow by Stonelax

drag/drop in ComfyUI You can now also find the workflow in the ComfyUI workflow Templates. Example Workflow JSON Example Workflow PNG Example Workflow on OpenArt.ai

Workflow details

The node is model independent, but I'm getting great results using FLux. However, Flux Dev sometimes has trouble with white backgrounds and creates blurry images. Defining a different color background can help a lot with that. Make sure your input image is as clean as possible. Describe your prompt accordingly

Model: Flux_dev (t5xxl_fp8_e4m3fn / clip_l)

Settings:
Euler - Simple
Steps - 25-30
Latent Image size - 1024x1024
Distilled CFG Scale - 3.5
CFG - 1
Comfyui-various - Image Contrast: 1.5 - 2

Lora: Simple_Vector_Flux_v2_renderartist
Trigger keyword: v3ct0r , vector
Recommended strengths: 0.6 - 0.9

Lora: Simple_Vectors_Flux_by_Sarcastic_TOFU
Trigger keyword: Simple_Vectors_Flux

Lora: Textimprover-FLUX-V0.4
Trigger: aidmaTextImprover
Strength: 0.3 - 1

LORA: v3ctora (Vector art & Line art (Flux))
Trigger: v3ctora style

Prompt tips:
Style: Bold, clean, smooth letters with rounded corners, solid fills (no texture), black on white, high contrast
Theme: Typography, large bold lettering, playful, dynamic
Text: With the text: "YOUR TEXT"
colors: black, white
Background: Clean white, simple design, no extra details

Alternative Route

For the second option in this workflow; you load an existing 2 color image of an icon/text and can convert it to SVG. You can use an upscale node for the input image, but I found it rarely improves quality; in fact, it is usually worse than the original.
It usually helps more to clean up the contrast, especially with highly compressed jpg images. The output svg can be scaled using the output_scale parameter

zoomed in original jpg vs default settings Potracer SVG conversion reference

Sources, Shoutouts, Love and Inspiration

Potrace by Peter Selinger: https://potrace.sourceforge.net/
Potrace Pure Python by Tatarize: https://github.com/tatarize/potrace
ComfyUI-ToSVG by Yanick112: https://github.com/Yanick112/ComfyUI-ToSVG
StabilityMatrix: https://github.com/LykosAI/StabilityMatrix
Flux Text to Vector Workflow by Stonelax@odam.ai
Gemini AI

CHANGENOTES (BREAKING!!!)

V1.0.0 Initial release
V1.2.0 BREAKING / FIXING CHANGE: Output LIST --> Output String
Yanick112's ToSVG Node switched recently from LIST to String. V1.0.0 didn't account for that.
ComfyUI-ToSVG has the 'nightly' version in ComfyUI Custom Node Manager, and therefor doesn't do versioning.
- Update package; in your workflow: rightclick recreate node or Reload example workflow (V1.2).; It should connect now to ToSVG nodes.
- If it DID work before and now it doesn't, update the ComfyUI-ToSVG Package.
V1.3.1 FIXING CHANGE: SaveSVG With the new API node a default ComfyUI node SaveSVG is built. This node conflicts with Yanick112's ComfyUI-ToSVG --> SaveSVG node. The new SaveSVG node is currently under the API nodes, but will eventually be part of base ComfyUI functionality. See this ComfyUI issue. I have duplicated and renamed Yanick112's SaveSVG node to keep compatibility with older versions. The node now also outputs attaching to the NEW SaveSVG node of ComfyUI v0.3.32 and up.

To Do

Deploy V1 to Github
Deploy to Comfyregistry
Add Output Scale factor (v1.1)
Real life testing & feedback

Disclaimer

This is my First ever (public) ComfyUI node. While tested thoroughly, and as with all custom nodes, USE AT YOUR OWN RISK.

While tested a lot and I have IT knowledge, I am no programmer by trade. This is a passion project for my own specific usecase and I'm sharing it so other people might benefit from it just as much as i benefitted from others. I am convinced this implementation has its flaws and it will probably not work on all other installations worldwide. I can not guarantee if this project will get more updates and when.

License

"Potrace" is a trademark of Peter Selinger. "Potrace Professional" and "Icosasoft" are trademarks of Icosasoft Software Inc. Other trademarks belong to their respective owners.

This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2, or (at your option) any later version.

Furthermore, this program calls the Potrace Python Library by Tatarize, which is permitted to be relicensed under any terms the Peter Selinger's original Potrace is licensed under. If he broadly publishes the software under a more permissive license the Potrace Pyton port should be considered licensed as such as well, as well as this program.

Further, if you purchase a proprietary license for inclusion within commercial software under Peter Selinger's Dual Licensing program your use of this software shall be under whatever terms he permits for that. Any contributions to this port must be made under equally permissive terms.

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.