Scifig Review: strong product, clear room to improve (80/100)

Scifig is a freemium scientific illustration AI service for researchers.

Fact	Value
Domain	scifig.ai
Category	scientific illustration AI for researchers
Pricing	Freemium; 0–40 USD
Pages crawled	38
Crawl date	2026-08-31

Scifig earns 80/100 by combining clear researcher-focused positioning with strong design execution and sound technical foundations. First impressions and design execution both score 92/100, while audience and messaging score 90/100; together, they make the product’s purpose easy to understand.

The weaker areas are practical. Performance scores 65/100, with the homepage’s 2.1s LCP affected by a large JavaScript bundle and an unoptimized image. Writing quality scores 67/100, and editorial QA scores 64/100, reflecting useful topical guidance alongside malformed paragraph extraction and other quality signals. Decision-support surfaces are the lowest-scoring area at 58/100: comparison content does not give readers explicit recommendations or tradeoffs.

Usability, accessibility, documentation, risk and stability, and technical SEO remain solid to strong. The result is a capable, focused service whose next gains come from helping visitors compare options, improving content quality, and reducing homepage friction.

SciFig makes its researcher focus and product category clear immediately. The hero claim, ‘Any input. One AI scientific figure generator.’ is specific and commits to seven creative paths, editable output, journal-style rendering, and domain awareness. Logos from 20+ universities and tiered pricing with credit counts provide concrete proof points. The distinction from generic AI image generators rests on accuracy, editability, and scientific output. Keep the positioning, but use one consistent brand form—SciFig or scifig.ai—across the experience to strengthen recognition.

SciFig’s messaging answers the practical questions researchers bring to a figure workflow. Clear H1s and feature descriptions explain the product; university logos and data privacy assurances address trust; a free tier, paid plans, and credit counts address cost; and direct CTAs explain how to begin. Terms such as ‘vector export’, ‘publication-ready’, and ‘journal-style output’ fit the audience’s expertise without unnecessary simplification.

SciFig is generally easy to navigate, but decision-making takes an avoidable extra step. The homepage links directly to Text-to-Figure and blog content, while pricing is two clicks deep through the home-to-pricing path. Add Pricing to the main navigation. The contact page also labels one response area as ‘快速响应’ without clearly identifying whether it is live chat or a support ticket, and the email address sits below multiple sections. Label each contact method plainly and standardize links that currently alternate between buttons and text.

SciFig has a strong accessibility base at 85/100, with no critical P0 failures reported in the available findings. Keyboard and screen-reader navigation still need two structural improvements: there is no skip link and no <main> landmark. Add a skip link to every page and identify primary content with <main>. The pricing page also has a reported color-contrast failure, while some icon-only buttons lack aria-label; resolve both in the next accessibility pass.

SciFig’s visual system is disciplined and works across screen sizes. Spacing follows a 16px/24px grid, pages use one H1 with clear section breaks, and mobile checks found no horizontal scroll. Tap targets are at least 44px and inputs are 16px, while contrast checks pass AA requirements in the design review. The remaining work is refinement: the Free Plan card uses 16px top padding while adjacent cards use 24px, and blog headings at 28px can feel undersized on mobile.

The homepage’s 2.1s LCP is within the supplied 2.5s reference threshold, but the page still carries avoidable render friction. Lab data records 2,101ms LCP, 31ms TTFB, and 415ms total blocking time, with the initial JavaScript bundle estimated above 300KB. The LCP image has no preload and multiple images lack width and height, increasing shift risk. Reduce the initial bundle, defer non-critical scripts, preload the critical image, and declare image dimensions. Network logs also show 27 blocking third-party requests.

SciFig’s strongest copy names the researcher’s workflow: ‘From any input to a ready-to-use scientific figure — editable at every step.’ Text-to-Figure, Figure Enhancer, Sketch-to-Figure, export options, and publication rights make the promise tangible. The color-palette article adds practical guidance, including no more than six muted categorical colors and pairing color with labels or shapes. However, the homepage repeats ‘Any input’ in three headings. Assign separate jobs to input, editing, and export sections and remove the repeated headline phrasing. Then fix extraction before judging scannability: N1 reports one paragraph for 3,426 homepage words and one for a 1,106-word app page.

SciFig’s comparison content does not yet give readers a defensible choice. The ‘10 Best Scientific Illustration Tools in 2026’ page gives SciFig a dedicated section without explicit reasoning about why it wins or where another tool is better. Its table includes ‘AI-Powered?’ but omits integration with lab software, export formats such as SVG/PDF, and support for multi-domain figures; BioRender and MindTheGraph are absent from the detailed sections. Add these decision axes, state tradeoffs, and explain the recommendation. Pricing should also state annual-billing savings clearly and keep schema prices aligned with display prices.

SciFig’s review content needs a visible method and a clear commercial disclosure before readers can fully assess its recommendations. The ‘10 Best Scientific Illustration Tools in 2026’ page has no ‘How We Chose’ or ‘Methodology’ section, so selection criteria are not explained. Add criteria such as AI-powered generation, ease of use, and export formats, then place a plain-language affiliate disclosure near the recommendations. This is a fixable transparency gap: the content can retain its usefulness while showing readers how conclusions and financial relationships are handled.

SciFig shows sound operational foundations: robots.txt allows public paths while blocking app internals, the sitemap contains 731 URLs, and sampled canonicals are self-referencing. HTTPS uses single-hop 301 redirects for HTTP and www variants, with HSTS preload, nosniff, and a comprehensive CSP. Raw and rendered text parity is exact, with 0% JavaScript-dependent content across sampled templates. The remaining stability picture is incomplete because the crawl bundle has no Lighthouse or PSI measurements for LCP, INP, or CLS and no structured-data extraction. Run those probes on key templates before closing the assessment.

SciFig has useful, specific product and blog material, but publication QA is weakened by visible template artifacts and incomplete provenance. The anchors ‘0110 best scientific illustration tools in 2026’ and ‘05how to make a graphical abstract in 6 clear steps’ expose ordering tokens and should be normalized before publication. The claim ‘受到全球 100,000+ 研究者的信赖：’ (‘trusted by 100,000+ researchers worldwide’) also needs a source or qualification. Finally, repair paragraph segmentation: N1 reports one paragraph for 3,426 homepage words and one for 2,706 words in the color-palettes article, making structural QA provisional. Prioritize repairing paragraph segmentation before making stronger claims about scannability or structure.

SciFig has strong documentation ingredients. llms.txt lists tutorials for text-to-figure, enhancer, sketch-to-figure, reference-to-figure, PDF-to-figure, photo-to-figure, and vector canvas workflows. Turn those tutorials into a searchable help center organized by user job: create, improve, edit, export, and troubleshoot. Each guide should state prerequisites, expected output, credit usage, and next steps, with links from the app and pricing pages.

Technical SEO is strong on crawlability, rendering, and security. Robots.txt blocks app internals while allowing public paths; the sitemap contains 731 URLs; and sampled canonical signals are self-referencing. Raw and rendered text parity is exact, with 0% JavaScript-dependent content, so primary content is immediately available to crawlers. HTTPS redirects are single-hop and the sample includes HSTS preload, nosniff, and a comprehensive CSP. The evidence set cannot confirm Core Web Vitals or structured data because PSI/Lighthouse and JSON-LD extraction were not supplied. Capture both on the app, pricing, and blog templates.

Scifig is a strong fit for researchers who want publication-quality scientific figures quickly and prefer a focused SaaS workflow. Its clearest strengths are positioning at 92/100, design execution at 92/100, and audience and messaging at 90/100.

Three improvements would most strengthen the buying experience. The homepage needs performance work around its 2.1s LCP and the large JavaScript bundle. Writing and editorial QA need tighter output quality and clearer editorial standards. Comparison pages should state recommendations and tradeoffs directly, rather than leaving readers to infer them.

Scifig does not need a broader promise; it needs a more efficient path from promise to confident choice. For its intended research audience, the foundation is strong and the weaknesses are concrete.

This 13-dimension review uses SiteList’s crawl evidence and the public dimension score table for Scifig. The crawl covered 38 pages on 2026-08-31. Scores and article claims are limited to the supplied public findings; dimensions without a public score summary in this frame are not restated here.

Google Search Console was not connected, so search-performance enrichment was unavailable. The crawl was partial rather than a complete inventory of every possible page. Read How SiteList scores for the review methodology and dimension definitions.