Workflow A
Method Text Workflow
Use the main page when you want AutoFigure-Edit to generate the first-stage image
for you.
Choose the Right Workflow
Start from method text if you want the full pipeline. Start from import mode if you
already have the stage-1 academic raster figure and only want SAM + SVG.
Workflow B
Import Existing Figure
Use the import page when you already have the academic raster figure and want to
continue directly from segmentation and SVG reconstruction.
Recommended Presets
Preset 1: OpenAI Main Route
SVG / Reasoning Provider: OpenAI Responses. Step 1 Image Provider: Same as SVG path.
Image Model: gpt-image-2. SVG Model: gpt-4.1.
Preset 2: Gemini + OpenAI Images
SVG / Reasoning Provider: Gemini. Step 1 Image Provider: OpenAI Images. Image Model:
gpt-image-2. Use this if you prefer Gemini SVG reconstruction but OpenAI image generation.
Preset 3: Custom Relay
Choose Custom and fill Custom API URL when you use an OpenAI-compatible relay or your
own gateway. Keep the default Custom endpoint if it already works for you.
What the Pipeline Actually Does
Step 1
Generate or Import figure.png
The system either generates the academic raster figure from method text, or accepts
your uploaded stage-1 figure directly.
Step 2
Run SAM3 segmentation
SAM3 detects icon-like regions and creates labeled placeholders plus box metadata.
Step 3
Crop icons and remove background
Each detected icon is cropped and cleaned so later replacement in SVG becomes easier.
Step 4
Rebuild as SVG
The multimodal model reconstructs the figure as editable SVG while respecting the
placeholder layout from SAM.
Step 5
Replace placeholders and finalize
Placeholder boxes are replaced by processed icons and the final SVG is written for
editing or export.
Main Page: Step-by-Step Filling Guide
1. Paste method text
Start with the method section, not the abstract. Include the pipeline logic,
components, arrows, stages, and notable visual entities that should appear in the figure.
2. Choose SVG / Reasoning Provider
This decides how SVG reconstruction works. If you do not want to think too much, use
OpenAI Responses or Gemini first.
3. Decide whether step 1 should follow or override
Keep Step 1 Image Provider linked unless you specifically want a different image
model or a different service for the raster generation stage.
4. Fill API key and Custom URL only when needed
For OpenAI Responses + linked OpenAI Images, one compatible API key is often enough.
Fill Custom API URL only if you selected Custom on that route.
5. Tune image model, SVG model, and SAM settings
Leave the defaults first, then only adjust model ids or SAM prompt/backend if you
know what is failing or what visual style you need.
Import Page: Step-by-Step Filling Guide
1. Upload the stage-1 academic figure
This should be the raster figure that normally would have been produced by step 1.
Do not upload the reference image or a final SVG here.
2. Choose only the SVG / reasoning route
Import mode skips image generation, so there is no step 1 image provider to fill.
You only need to decide how SAM and SVG reconstruction should continue.
3. Fill SVG model and API key
Use the default SVG model first. Change it only if you know your provider exposes a
better model for multimodal SVG reconstruction.
4. Configure SAM backend
SAM still runs in import mode. You must choose whether it uses local SAM3, fal.ai,
or Roboflow, and provide the corresponding key if the backend requires one.
What Each Field Means
Method Text
Paste the method section of your paper. The cleaner and more structural it is, the
better the generated figure tends to be.
SVG / Reasoning Provider
Controls the text reasoning and the multimodal SVG reconstruction stage. This is the
most important provider selector on the page.
Step 1 Image Provider
Controls only the first-stage raster image generation. Leave it linked if you do not
need to separate the image path from the SVG path.
Custom API URL
Used only when the route is Custom. Fill the OpenAI-compatible base URL provided by
your relay or gateway.
Image Model
Default is gpt-image-2 for OpenAI Images. You can manually replace it with any
compatible image model id if needed.
SVG Model
Default follows the selected reasoning route. Use gpt-4.1 or gpt-5.5 on OpenAI
Responses, and the Gemini defaults on Gemini/OpenRouter/Custom unless you know you
need a different id.
Auto Upscale
Enabled by default. It enlarges figure.png to a 4K long edge while preserving aspect
ratio. Keep it on unless you specifically want the original resolution.
SAM Settings
SAM Backend selects how segmentation runs. SAM Prompt controls what objects the model
should try to detect, such as icons, people, robots, or animals.
SAM3 Backend Guide
Local (SAM3)
Best when you already installed SAM3 locally and want everything on your own
machine. No external API key is needed, but local dependencies must be ready.
fal.ai API
Good if you do not want to install SAM3 locally and you have a FAL key. Usually
stable, but it is an external paid API route.
Roboflow API
Often the easiest hosted SAM option. Use this when you want a remote backend and
your environment can reach the Roboflow endpoint.
How to Fill SAM Prompt
Think of SAM Prompt as the object vocabulary. Use comma-separated words such as
`icon,person,robot,animal` or add domain words like `diagram,cell,molecule,arrow`.
When to Change SAM Backend
If local SAM3 is unavailable, switch to fal.ai or Roboflow. If remote APIs are slow
or inaccessible, local becomes the fallback if your environment supports it.
When a SAM API Key Is Required
Local does not need a SAM API key. fal.ai needs a FAL key. Roboflow needs a
Roboflow key. If the SAM backend is local, leave the SAM API key blank.
Common Filling Examples
I only want the easiest stable setup
Main page. Provider = OpenAI Responses. Image Provider = Same as SVG path. Image Model = gpt-image-2. SVG Model = gpt-4.1. Fill one API key.
I already have the stage-1 figure
Import page. Upload the figure. Choose Provider = OpenAI Responses or Gemini. Fill SVG Model and API Key. Leave image settings alone because step 1 is skipped.
I use a relay / private API gateway
Choose Custom on the route you want to redirect. Fill Custom API URL with your gateway base URL, then fill the matching API key.