SciDraw AI Help Docs

SciDraw AI turns a written description, a rough sketch, an existing image or a data file into a figure you can put in a paper, a poster or a slide deck. This page covers the whole product: where to click, what each option does, what it costs in credits, and what you get back. Use the table of contents to jump to a topic, or start with the quick start below — it takes about five minutes and the free sign-up credits cover it.

Quick start

To make your first figure, open the AI drawing workspace, keep the Text to Image mode, describe the figure in the Image Description box and click Generate Image. After about 40 seconds the figure appears in a conversation. Type a follow-up such as "make the arrows thicker" to change it, then click the download icon to save it.

Sign up and get free credits

You can create an account with Google, GitHub, or an email address and password. No credit card is needed.

• Every new account gets 10 credits at sign-up.
• Every account, free or paid, gets 5 credits each day, added automatically. There is nothing to claim.
• Both gifts expire after 1 day, so they are for trying things out and light daily use, not for saving up.

A 2K Text to Image figure costs 5 credits, so the daily gift covers one standard figure a day. The Free plan has no monthly credits; for more, subscribe to a plan or buy a credit package (see Credits and pricing).

If you sign up with email, open the verification email and click the link; you are signed in automatically. The sign-up gift starts its one-day clock when you submit the form, not when you verify, so verify and use it the same day.

Your first figure, step by step

1. Open the workspace. Go to /ai-drawing. If you already have work there, click + (New Conversation) at the top of the History sidebar so the new figure gets its own conversation. You can type a prompt before signing in; the button reads Sign in to generate and your draft is kept through sign-in.

2. Keep Text to Image. Under Generation Mode, the Text to Image tile is selected by default.

3. Describe the figure. In Image Description, say what the figure shows, which parts it has, how they are arranged and which labels you want. Plain scientific language works best; SciDraw AI rewrites your description into a detailed drawing brief before generating. For example:

   Graphical abstract of a nanoparticle drug-delivery study: on the left, PEGylated
   lipid nanoparticles carrying siRNA circulate in a blood vessel; in the middle they
   accumulate in a tumor through leaky vasculature; on the right, a tumor cell takes
   them up by endocytosis and the siRNA silences a target gene. Label each stage.
   Clean white background, consistent blue and orange palette.

4. Optional: pick a template and style. Open Advanced Options and choose, for example, TOC Graphical Abstract as the template, Flat Illustration as the style and 16:9 as the aspect ratio.

5. Generate. Leave Image Resolution at 2K and Images per Run at 1. The line under the button shows the cost (5 × 1). Click Generate Image. If the progress card says the service is busy, wait rather than clicking again — a second click starts a second job.

6. Refine. Type one change into the box under the image, such as "Make the tumor cell larger and move the gene-silencing label below it", and press Enter. The new version appears below the original.

7. Download. Hover the image and click the download icon. You get the original PNG, with no watermark and at no cost. For PowerPoint, SVG or Visio versions, use Convert (see Export formats).

After a download, a small notice may ask Publishing this figure? (at most once a day). Copy credit line copies a ready-made acknowledgement such as "Created with SciDraw AI (https://sci-draw.com)." for your figure legend. It is optional.

If something goes wrong

Features

SciDraw AI has two workspaces: AI drawing for illustrations (graphical abstracts, mechanisms, workflows, pathways, covers) and SciVis for charts drawn from your data. Both run in the browser; there is nothing to install. Everything you make can be downloaded for free and converted into editable formats for credits.

Three drawing modes

The AI drawing workspace has three modes, shown as tiles under Generation Mode. Pick the mode by what you already have: words, a sketch, or a finished image.

Switching tiles clears the prompt and the upload, so choose the mode before you start writing. You can also link straight to a mode: /ai-drawing?mode=sketch opens Sketch to Image and /ai-drawing?mode=replicate opens Edit Image.

Text to Image takes a description of up to 5,000 characters, in any language. What makes the biggest difference:

• Name the figure type first: "a graphical abstract", "a three-step experimental workflow", "a signaling pathway diagram".
• Describe the layout: "left to right", "three panels labelled A, B, C", "a central cell with arrows to four outcomes".
• List the labels exactly as they should appear, in quotes: labelled "TNF-α", "NF-κB", "nucleus". Fewer, shorter labels come out more accurately.
• Say what matters visually: colour coding, a white background, the level of detail.
• One figure per prompt. Generate panels in very different styles separately.

Image Text Language sets the language of the labels drawn in the figure, independently of the language you write the prompt in. It defaults to the language of the site. All 11 site languages are available.

Sketch to Image redraws a rough sketch as a clean figure and keeps the positions of what you drew. Upload a JPG, PNG or WEBP of up to 10 MB; there is no prompt box in this mode, so write the labels legibly on the sketch itself. Photograph it flat and evenly lit, and crop away the desk and shadows. SciDraw AI has no drawing canvas — draw on paper, a whiteboard or any drawing app, then upload.

Edit Image changes an image you upload according to a written instruction: restyle it, change colours, translate or remove text, add or remove elements. The image can come from anywhere — an old paper figure, a screenshot, a figure from another tool. This mode has no template, style or text-language options; put any style or language change in the instruction. To change an image you just generated in SciDraw AI, you don't need this mode: use the follow-up box under the image (see Refining and editing).

For figure-by-figure guidance and ready-made example prompts (graphical abstracts, pathways, PRISMA diagrams, lab equipment, teaching diagrams) browse the tools directory.

Templates, styles and image settings

Templates and styles are optional presets under Advanced Options, available in Text to Image and Sketch to Image. A template says what kind of figure you are making; a style sets the look. Each adds matching keywords to your prompt and neither changes the cost.

• Templates: TOC Graphical Abstract, Journal Cover, Experimental Workflow, Mechanism Illustration, Signaling Pathway, System Architecture, Comparison Analysis.
• Styles: Flat Illustration (the most common look for papers and slides), Isometric 3D, Photorealistic, Schematic Diagram, Artistic Abstract, Scientific Illustration.

Pick a template that matches what you describe, or none; a Signaling Pathway template with a prompt about a lab workflow pulls the image in two directions.

Three more settings shape every run:

• Image Resolution. 2K is the default and fine for drafts, slides and most journal figures. 4K costs 20 credits per image in any mode, takes longer, and renders text more clearly — useful for text-heavy figures, posters and covers. Every account can use 4K. A practical approach: iterate at 2K, then make the final version at 4K.
• Aspect Ratio. Auto, 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, 5:4, 4:5 or 21:9. Text to Image starts at 4:3. In Sketch to Image and Edit Image, Auto keeps the proportions of your upload. Use 16:9 for a slide, 1:1 for a TOC graphic, 3:4 for a journal cover.
• Images per Run. Generate 1, 2, 4, 8 variants at once to compare. Results appear as thumbnails; click one to select it for Edit or Convert.

The cost line under Generate Image reads per image × images = total. Your balance is checked for the whole batch before the run starts, and only the images that are actually produced are charged: if 3 of 4 succeed, you pay for 3.

Refining and editing

After the first image, the workspace becomes a conversation. Type what to change into the box under the image ("Describe how you want to modify the image..."), press Enter (Shift + Enter for a new line), and a new version appears below the previous one. Earlier versions stay in the conversation, so you can always go back.

Which image is edited. Your message edits the image you selected with Edit (top-left of each image; a thumbnail appears above the input box), otherwise the newest image. To branch, click Edit on an earlier version and describe a different change. A follow-up uses the settings currently shown, so switch to 4K before sending if you want the final version at 4K.

What an edit costs. The same as generating one image in the mode the conversation started in: 5 credits at 2K after Text to Image, 10 after Sketch to Image or Edit Image, 20 at 4K. Failed edits cost nothing.

How editing works. Every edit is a written instruction applied to the whole image; there is no brush, mask or region selection, and the model redraws the full picture. Small details elsewhere can shift, so say what must stay the same:

Make one change per message, quote exact text, and rephrase rather than repeat an instruction that failed twice. When a figure has drifted after many edits, start a new conversation with a better prompt — it is usually faster and cheaper than ten fixes. For pixel-exact changes to one label or arrow, convert the figure and finish it in PowerPoint or a vector editor.

SVG and PPTX results are final files: they can't be edited with AI or converted again. Make content changes first, then convert.

History. All your conversations are listed in the History sidebar of the workspace; click one to reopen it with its images and conversions. Search conversations... filters by title, and the trash icon deletes a conversation. The pricing page lists 30-day history retention on the Free plan and permanent retention on paid plans, so download any figure you need to keep.

Data charts (SciVis)

SciVis makes charts from data. Upload a CSV, TSV, TXT or Excel file (or paste a table), optionally say what you want to show, and SciVis writes and runs plotting code to draw a publication-style chart. Because the chart is drawn from your numbers, bars, points and axes match the values exactly — SciVis does not invent data. Each generation and each follow-up change costs 10 credits, charged only when a chart is produced.

1. Add your data. Sign in, then drag a file onto Data Input or use Paste a table (copy the cells including the header row). Files can be up to 5 MB; only the first worksheet of an Excel file is read. Check the Data Preview: the first row must hold unique column names, and numeric cells must not contain units ("12 mg") — put units in the column name.
2. Say what you want (optional). For example, "box plot of expression by treatment with individual points and significance brackets". Leave it empty and SciVis picks a suitable chart.
3. Set Advanced Options (optional). Choose one or more of 46 Chart Types (several types plus a Subplot Layout make a multi-panel figure), a Journal Style (Default, Nature, Science, Cell, PLOS or ACS), the Image Size (single or double column) and a Color Scheme (Okabe-Ito, the default, and Wong are colour-blind safe).
4. Generate Chart. If the code fails, SciVis fixes it and retries automatically; if it still fails, you see "No credits were charged."
5. Refine and download. Ask for changes in the chat box ("log10 scale on the x-axis", "move the legend outside on the right", "translate labels to Chinese"). Use Edit on an earlier version to change that one instead of the latest.

Charts download for free as PNG, PDF, EPS or TIFF, at 300 dpi (Default, PLOS) or 600 dpi (Nature, Science, Cell, ACS) and the column width you chose. There is no SVG download. Convert to Excel (.xlsx) turns a single-chart figure of a supported type into a workbook with your data and a native, editable Excel chart for 10 credits.

The 46 chart types are grouped as Basic & Trends (bar, line, scatter, time series, waterfall, radar …), Distribution & Statistics (box, violin, histogram, raincloud, Q–Q …), Relationships & Models (regression, error bars, correlation matrix, Bland–Altman, standard curve …), Biomedical (volcano, Kaplan–Meier, forest, ROC, Manhattan, dose–response, enzyme kinetics …) and Flows & Structures (heatmap, dendrogram, UpSet, Sankey …). Every type costs the same. Tidy "long" data — one row per observation, with columns such as group, time, value — gives the most predictable charts.

Common chart types also have their own pages with example data, such as the volcano plot generator, the Kaplan–Meier plot generator and the box plot maker. They cost the same and can continue in the full SciVis workspace.

Export formats

The download button always saves the original PNG for free. When you need to edit the figure in another program, use Convert on the image. Conversions are charged only when the file is ready; a failed conversion costs nothing.

Which one to choose:

• You need to change some words or translate: Editable PPTX is the quickest. Shapes and icons stay one picture; only the text is editable.
• You work in Illustrator, Inkscape or Figma: SVG. Extract Text Only is enough for labels and suits photorealistic or shaded figures. Convert All Elements suits flat figures — icons, arrows, boxes — when you want to recolour, move parts or enlarge for a poster without blur.
• It is a flowchart, pipeline or architecture diagram you will keep editing: Visio. It takes about a minute; leave the tab open and don't resubmit. Photorealistic or shaded illustrations are a poor fit.
• The journal wants TIFF or a minimum pixel size: use the image converter below.

Text recognition works line by line, so very small, curved or overlapping labels may be missed; check the labels after converting. To convert an image made elsewhere into PPTX, SVG or VSDX, upload it at /vectorize-image — same modes, same prices.

Print resolution. The image converter takes a JPG, PNG or WebP up to 10 MB and saves it as PNG, PDF or TIFF (LZW-compressed). The 300 DPI, 600 DPI and 1000 DPI presets make sure the longer side is at least 1024, 2048 or 4096 px, using AI upscaling when the image is smaller; larger images are never shrunk. The PDF holds a raster picture, not vector artwork.

The DPI a figure has in print depends on how wide it is printed: DPI = pixels ÷ printed width in inches. A 2048 px image printed 89 mm wide (a single column) is about 585 DPI; printed 178 mm wide it is about 292 DPI. Check your file with the Figure DPI calculator and look up your journal in Journal figure requirements.

Before you submit

Three tools help with the last step before a journal submission:

• Figure Checker rates one raster figure (PNG, JPEG, WebP or TIFF, up to 4 MB) on effective DPI, file format, transparency, colour space, grayscale readability and colour-blind safety, each as Pass, Warn or Fail. 10 credits per report.
• Figure Legend reads an uploaded figure and drafts a short, manuscript-style legend as plain text and as a LaTeX caption, in any of the 11 site languages. 5 credits per legend. Always check the draft against your actual methods and results.
• Free reference pages, no account needed: Figure DPI calculator, Journal figure requirements, Figure accessibility kit and the AI figure policy tracker.

To credit or cite SciDraw AI in a paper, copy the wording from /cite.

Credits and pricing

Everything that calls an AI model is paid for with credits. Every plan is metered: a paid plan adds a fixed number of credits each month, and each generation, conversion or check spends some of them. The tables below are generated from the same configuration as the pricing page, so they always show current prices.

Plans

Prices are in US dollars. Yearly plans are charged once a year; the per-month figure is that yearly price divided by twelve.

• Free has no monthly credits. Free accounts run on the sign-up gift and the daily gift.
• Monthly plans renew on the day you subscribed. Each successful renewal adds the plan's monthly credits.
• Yearly plans are paid up front, but the credits arrive month by month: the first month's credits when the payment goes through, then one month's allowance at a time for the rest of the year.
• Lifetime is a single payment with no renewals. It adds its monthly credits when you buy it and again at the start of each calendar month.
• Changing plans. On a monthly plan, the pricing page shows Upgrade or Downgrade; the new plan's full monthly allowance is added straight away and credits you already have keep their expiry date. On a yearly plan you can't switch during the subscription year; buy a credit package for extra credits instead.

There are no team seats or shared balances; each researcher uses their own account. Students and educators can get a discount (see the FAQ).

Credit packages

A credit package is a one-time purchase with no subscription. Purchased credits never expire, so packages suit occasional use, a deadline crunch, or topping up a plan without upgrading. You can buy them on any plan, including Free, from the pricing page or Settings → Credits.

What each operation costs

Every operation has a fixed cost, shown before you run it. Downloading a result you already have is always free.

A drawing run costs the per-image cost times Images per Run. The largest single run, 8 images at 4K, costs 160 credits. A follow-up edit costs the same as one image in the mode the conversation started in.

Expiry and spending order

Your balance is made of separate batches — a sign-up gift, today's daily gift, this month's plan credits, a purchased package — each with its own expiry:

When you spend credits, SciDraw AI always takes them from the batch that expires soonest, and spends credits that never expire last. So gifts and plan credits are used before a package you bought, and nothing that is about to expire goes to waste.

Settings → Credits shows your balance and how many credits expire soon; the Transactions tab lists every credit added, spent or expired.

Failed jobs and refunds

• You only pay for results. Your balance is checked before a job starts; credits are deducted after the image, file, chart, check or legend has been created. A failed or filtered job costs nothing, and in a batch only the successful images are charged. Some jobs reserve credits when they start and release them if they fail, so you won't see a separate refund entry.
• Refunds of purchases. The Terms of Service allow a refund of a first subscription or a Lifetime purchase within 7 days if less than 10% of its credits were used, of a renewal within 48 hours if none of the new period's credits were used, and of an unused credit package within 7 days. Approved refunds go back to the original payment method minus a 10% processing fee (no fee for duplicate charges or billing errors on our side). To ask, email support or use the contact page with your account email, the order or invoice, the purchase date and the reason. Requests are handled within 5 business days.
• Cancelling. Go to Settings → Billing and click Manage Subscription and Billing. The plan runs until the end of the period you paid for; credits already issued keep their own expiry, and purchased packages are unaffected.

Credits have no cash value and can't be exchanged for cash.

FAQ

Credits

Why did an image cost more credits than I expected?

5 credits is the price of one 2K Text to Image picture. Sketch to Image and Edit Image cost 10 at 2K, any mode at 4K costs 20, and follow-up edits cost the same as the mode the conversation started in. The total is multiplied by Images per Run.

Do failed generations or conversions cost credits?

No. Credits are only taken for images and files that are actually produced. If your balance was too low, the job didn't run and nothing was deducted.

Do I have to claim the daily credits, and do they add up?

No to both. 5 credits are added to every account once a day, at about 16:10 UTC, without signing in. Whatever you don't use expires before the next ones arrive.

Do subscription credits roll over? Do yearly plans give all credits at once?

Plan credits expire 31 days after they are added and don't roll over. Yearly plans give the same monthly allowance as monthly plans, one month at a time for the whole year. Purchased packages never expire.

Do the API and MCP use the same credits?

Yes. The API and the MCP connector spend credits from the same balance at the same prices as the website.

Billing, refunds and invoices

Which payment methods are accepted?

All payments go through Stripe checkout in US dollars. Cards always work. Alipay and WeChat Pay may also be offered, depending on your region and what you buy; the checkout page shows exactly which methods are available for your order. If your card is in another currency, your bank converts the amount.

Can I get a refund?

Within the windows in the Terms of Service: 7 days for a first subscription or Lifetime purchase with less than 10% of its credits used, 48 hours for a renewal with none of the new credits used, and 7 days for an unused credit package. See Failed jobs and refunds for how to ask.

Where are my invoices and receipts? Can the invoice show my university?

Invoices and receipts are in the Stripe customer portal: Settings → Billing → Manage Subscription and Billing. If you need an invoice in your institution's name (with address, VAT number and an optional PO number), send the details on Institutional invoicing before you pay and wait for the confirmation email. Payment is still made online by card.

Is there a student or educator discount?

Yes: 40% off one order for students and 20% off for teachers and academic staff. Apply once on the Education discount page with proof of your status; the discount is then applied automatically at checkout, with no code. It works on any plan or credit package, and a yearly plan usually saves the most. Only one discount applies per order.

How do I cancel my subscription?

In Settings → Billing, click Manage Subscription and Billing and cancel in the portal. You keep the plan until the end of the paid period; there is no pro-rata refund for the rest of it.

Using your figures

Who owns the figures I make?

Under the Terms of Service you own the images you generate and keep all rights to what you upload. You grant SciDraw AI a limited licence to store, display and process your content to run the service. Free users' works may be shown anonymously in the community gallery; paid users' creations are kept private.

Can I publish a figure in a journal or use it commercially?

The paid plans include journal publication and commercial use, as listed on the pricing page. If you use the Free plan and plan to publish, read the current Terms and pricing page, and contact us if in doubt. Whatever you use, follow your journal's AI policy (see the AI figure policy tracker) and never present a generated image as experimental data.

Can I get a licence document to show an editor?

Yes. The License Certificate button next to Download issues a one-page PDF for that figure or chart, with a certificate number and a QR code that opens a verification page. It is free and available to any account that has made a purchase (a plan or a credit package), even after a subscription ends. See Publication license.

How do I credit or cite SciDraw AI?

Add "Created with SciDraw AI (https://sci-draw.com)." to the figure legend, or a sentence in Methods or Acknowledgments. APA and BibTeX entries are on /cite.

Exports and editing

Can I change the text inside a generated figure?

Not by typing into the image. Either ask for the change in a follow-up message, or convert the figure to Editable PPTX or SVG (Extract Text Only), which turns the labels into editable text.

Can I edit only one area of an image?

No. There is no mask or region selection; every edit applies to the whole image. Describe the part to change precisely and say what to keep.

Do my images have a watermark?

No. Images you generate have no watermark. Only the example previews on tool pages are watermarked.

Can I get an SVG of a SciVis chart?

No. Charts download as PNG, PDF, EPS or TIFF. PDF and EPS are vector formats most journals accept, and single charts can be converted to an editable Excel workbook.

Account

I signed up with email and now signed in with Google. Do I have two accounts?

No, if both use the same email address: Google and GitHub sign-ins are linked to the existing account, so your credits and history stay in one place.

How do I change my password, email or delete my account?

Change or add a password in Settings → Security, or use Forgot password on the login page. Changing the account email and deleting an account are done by support: contact us from the account's email address.

I signed up with email but can't sign in.

Email accounts must be verified first. Open the registration email and click the link; you are signed in automatically. Check your spam folder, or use Resend verification email on the login page.

How do I change the language?

The language menu in the site header changes the interface. The language of the text drawn inside a figure is set separately, with Image Text Language in the drawing workspace.

Can my lab share one account?

Accounts are personal; there are no team seats or shared balances. Each researcher should use their own account. For an institution-named invoice, see Institutional invoicing.

Use SciDraw AI in ChatGPT and Claude (MCP)

SciDraw AI has an MCP (Model Context Protocol) server, so you can ask ChatGPT, Claude, Cursor or Codex to draw a figure, edit or convert one, or turn data into a chart without leaving the assistant. You add one URL, sign in to SciDraw AI in your browser, and the assistant spends credits from your existing balance at website prices. No API key is needed.

Connect your assistant

Paste one line into your assistant and sign in to SciDraw AI when the browser opens.

ChatGPT app / Codex (Codex in the ChatGPT desktop app, or the Codex CLI):

/goal Read https://sci-draw.com/mcp/install.md to install SciDraw AI for Codex and run its first-time setup.

Claude (Claude Code connects itself; on claude.ai and in the desktop app, add https://sci-draw.com/api/mcp under Settings → Connectors instead):

Connect the SciDraw AI MCP server at https://sci-draw.com/api/mcp

Manual steps for ChatGPT on the web, Claude, Claude Code, Codex and Cursor are on sci-draw.com/mcp. Clients that can't do a browser sign-in can send an API key as the header Authorization: Bearer sd_…; create keys in Settings → API Keys.

What you can ask for

The assistant can generate a figure from a description (2K or 4K, any aspect ratio, up to 4 images), edit a figure from an image link, convert a figure to SVG, PPTX, PNG, PDF or TIFF, make a chart from CSV or JSON data, and report your credit balance. Visio conversion is not available through MCP. For example:

• "Use SciDraw AI to draw a graphical abstract: tumour cells release exosomes carrying miR-21 to fibroblasts. Flat style, white background, English labels."
• "Convert the last figure to an editable PowerPoint file."
• "Here is my CSV of dose and response by group. Make a dose–response chart in Nature style."
• "How many SciDraw AI credits do I have left?"

A request waits up to 50 seconds. If the job takes longer, the assistant gets a job ID and checks it again until the result is ready. Don't ask it to send the same request again — a repeated request is a new job and is charged again. Failed jobs are not charged.

To disconnect, remove the connector in your assistant's settings (or delete the API key). For the HTTP API, see the developers page.