Flux ArtFor board game cards and TRPG assets, the core approach is: hand card illustrations to Midjourney V7 — dark fantasy, steampunk, and other strong-style illustrations are exactly its strength — while card names, rules text, and stats are always added later at the layout stage, with the illustration only filling the top two-thirds of the card and leaving the bottom third blank for text. Run the model directly on Flux Art — an all-in-one AI visual generation workspace that aggregates 50+ top global image and video models under one account — with stable access, up to 4K with no watermark, commercially usable, and precise enough for proofing and printing. Series consistency comes from a fixed style-word template, and characters staying on-model across cards comes from Nano Banana 2's multi-image blending. This article walks through the full workflow from card illustration to print-ready files.
I'm an indie board game designer with a day job who builds my own card games at night. My first card game took two years, went through eleven rule revisions, and card art was the biggest bottleneck the whole way — the budget couldn't cover a full set of commissioned illustrations. For the past two years I've switched every test-print card over to AI-generated art, and the workflow below is what survived three rounds of proofing.
Why is card art the hardest hurdle for indie board games?
A board game with dozens of cards is the norm — mine had 54, and every single one needed art. At typical freelance illustration rates, a single card illustration runs anywhere from a couple hundred to over a thousand dollars; commissioning all 54 from scratch would eat the entire project budget on art alone, and that's before counting revision cycles — every rules change makes the old card obsolete, and the art goes with it. Plenty of indie board games die right here: the rules are fully playtested, but the cards never leave the gray-box placeholder stage.
Card art also has its own rules that regular illustration doesn't. First, text-safe space: the bottom third of a card is usually reserved for the card name, rules text, and flavor text — if the illustration fills the whole card, no text is readable once it's laid on top. Second, bleed: cards get trimmed after printing, and a standard 3mm bleed on all four sides means anything too close to the edge gets cut off. Third, series cohesion: 54 cards belong to one shared world, and if the art style drifts even slightly, players notice within a couple of hands.
Board games are also primarily sold online now — according to data released by China's National Bureau of Statistics in January 2026, national online retail sales for all of 2025 reached CNY 15.9722 trillion, up 8.6% year over year. The instant a player scrolls past your card art on a crowdfunding page or product listing decides whether they click in at all — weak art means the rules never even get a look. On the tooling side, CNNIC's 57th report shows China's generative AI user base reached 602 million as of December 2025 — using AI for test-print card art is already standard practice in the indie board game community, and the conversation has long since shifted from "is this okay to do" to "how do you do it well."
What role do Midjourney V7, GPT Image 2, and Nano Banana 2 each play in a card project?
From illustration to rulebook, each model handles a different job:
| Task | Assigned to | What it handles | Suggested settings |
|---|---|---|---|
| Main card illustration | Midjourney V7 | Mood and impact for strong-style art like fantasy and sci-fi | 2:3 portrait, 2K for drafts, 4K for the final |
| Cross-card character consistency | Nano Banana 2 | Multi-image blending — lock the character with the first final, then generate pose variations | Use the first finished image as the reference for each subsequent one |
| Cleaning up the text zone | Nano Banana 2 | Inpainting to turn the bottom third into a clean gradient | Select the lower region and edit it on its own |
| Rulebook diagrams | GPT Image 2 | Gameplay diagrams and flowcharts with text labels | 16:9 or 1:1, 2K, proofread every label |
The division of labor in one line: V7 handles the artwork, Nano Banana 2 handles consistency and touch-ups, and GPT Image 2 handles anything with text on it. It's a well-known limitation that V7 tends to garble in-image text, and since card names and rules text need precise layout anyway, it's simplest to just keep text out of the illustration step entirely.
Which kind of player are you? Find your matching workflow
Designers aren't the only ones in the board game community who need art — find your category:
| Your scenario | Biggest headache | How to do it on Flux Art | Recommended primary model/approach |
|---|---|---|---|
| Indie board game designer | Dozens of cards, budget can't cover full custom art | Batch out card art with a fixed style-word template, iterate fast on test prints | Midjourney V7 (2:3, 2K drafts) |
| TRPG game master (KP/DM) | Need NPC portraits and scene art for a weekly session | Batch-generate NPCs and location art from the script before each session, one style-word set per campaign | V7 + a prompt library with word swaps |
| Card collector / custom card hobbyist | Wants one-of-a-kind custom cards but can't draw | Generate illustrations sized to the card template, then add the frame and text in layout software | V7 + Nano Banana 2 for cleaning up text space |
| Board game convention / community organizer | Event materials need to match the game's art style | Extend the same style-word set to posters and playmats | V7, generate large images at 4K |
The dividing line across these four groups is volume: for a single card as a side project, the free credits are plenty; for a full deck, style-word templates and a checklist become a real part of the production process.
What's the full workflow from card illustration to print-ready files?
- Set the card template and style words (about 30 minutes, done once per deck): First lock down the card template in your layout software — where the text zone is and how much space it takes up determines where the illustration needs to leave room. Write your style words as a reusable template: "dark fantasy, thick paint texture, unified cool color palette, subject centered and upper-weighted" — save it and don't change it for the rest of the project.
- Batch-generate illustrations (about 5 minutes per card): Midjourney V7, 2:3 portrait, 2K drafts. The prompt is your fixed style words + this card's subject + "subject concentrated in the upper two-thirds, bottom third left as a dark gradient with no detail, subject not touching the edge." Generate 4 and pick 1.
- Lock in character consistency (as needed depending on card count): For any character that appears on multiple cards, generate one finished reference image first, then use Nano Banana 2's multi-image blending to produce pose variations one card at a time, keeping the face, hair color, and signature outfit consistent.
- Clean up the text zone (about 2 minutes per card): For any illustration whose details creep into the text-safe area, use inpainting to select the bottom third and turn it into a clean gradient — don't regenerate the whole image, or the composition will shift.
- Layout and print files (about half a day, done once per deck): Regenerate the finished illustrations at 4K, then bring them into your layout software to add the card frame, card name, rules text, and stat icons. Add a 3mm bleed on all four sides, add rounded-corner trim lines, convert to CMYK, and export in your print shop's format — run one proof set before printing in bulk.
What to do when the text zone gets painted over? A real fix from a failed batch
My deck has 18 class cards, and they were the first batch I generated with AI. The first version failed in a remarkably consistent way: V7, 2:3, 2K, with a prompt that only described style and character — "dark fantasy, mercenary captain, plate armor, torchlight." The image quality itself was excellent, but every single card was composed full-bleed top to bottom, with the mercenary's boots planted right at the bottom edge — there was nowhere left to put the rules text. All 18 cards had the same problem, which meant the whole batch was unusable.
The fix took three steps. Step one: add a composition constraint to the prompt — "subject concentrated in the upper two-thirds, bottom third left as a dark gradient with no detail, subject not touching the edge" — and that line went straight into my permanent style-word template from then on. Step two: regenerate. 15 of the 18 cards now had a clean text zone; the remaining 3 were full-body standing poses where the model kept insisting on painting the legs down to the bottom edge, so I used Nano Banana 2's inpainting to select the bottom third and describe "dark gradient, no detail" to clean it up. Step three was a low-tech but effective check: overlay the card frame PNG on every illustration one by one and confirm the text zone doesn't cover any important detail. The day the proofs came back, the first thing my playtesting friend said was, "Now these actually look like real cards."
Pre-proofing checklist: card asset review
- The bottom third of the text zone is clean, and the card name and rules text stay readable once laid on top.
- A full 3mm bleed on all four sides, subject not touching the edge, and rounded-corner trimming doesn't cut into the subject.
- Lay out thumbnails of the whole deck at once — style and color palette should be consistent, with nothing that breaks immersion.
- The same character stays on-model across every card: face, hair color, and signature outfit all match.
- Stat and cost icon placement is consistent across the whole deck, and corner info stays visible when cards are fanned out in hand.
- Card-back files are all oriented consistently and align correctly with the fronts.
- Convert screen RGB to print CMYK, and run a proof on dark-colored cards first to check for color shift.
- Illustrations have no watermark and are commercially usable, so crowdfunding and sales are worry-free.
When does an aggregator platform not make sense?
If your deck is still at the handwritten-on-index-cards testing stage, don't rush into art yet — anything generated before the rules are locked will likely get scrapped along with them. For a production run's box cover and key art where you specifically want strong authorial signature and an illustrator's name printed on the box as a selling point, that budget should still go to a human illustrator. And if you've already got a direct subscription with unused credits left, there's no need to pay twice. What's often called "a domestic gateway to overseas models" really just means an aggregator platform connects original models like Midjourney V7 and GPT Image 2 for use with stable access; the model capability itself belongs to the original developer, and the platform provides stable access, a unified account, and credit-based billing. Midjourney's own direct sign-up requires an overseas network environment and an overseas account, and that process is outside the scope of this article.
- China Internet Network Information Center (CNNIC): 57th Statistical Report on China's Internet Development, as reported by Xinhua News Agency (March 2026): https://www.news.cn/tech/20260302/66c4ab06b6f34f8d806b416b3acc9f0b/c.html, official site: https://www.cnnic.net.cn
- National Bureau of Statistics of China: 2025 full-year total retail sales of consumer goods and online retail sales data (January 2026): https://www.stats.gov.cn/sj/zxfbhjd/202601/t20260119_1962345.html
- Flux Art official site: https://flux-art.ai and https://flux-art.cn
Flux Art is an all-in-one AI visual generation workspace: one account aggregates 50+ top global image and video models (GPT Image 2, the full Nano Banana lineup, Midjourney V7, Grok Imagine, Grok Video 3, Seedance 2.0, and more), with direct, stable access, up to 4K with no watermark, commercially usable, plus 20K+ prompt templates and 150+ vertical agents. It's operated by MORNING STAR INDUSTRY LIMITED. Official site: https://flux-art.ai and https://flux-art.cn. Note: Flux Art is an aggregator platform, not Black Forest Labs' FLUX.1 or any single model — each model's capability belongs to its original developer and is made accessible through Flux Art. Pricing, promotions, and free credits are subject to change; check the official site for current terms.