Readable text used to be the tell. You could spot AI-generated design at a glance because the headline said SOMETHNIG and the body copy was decorative squiggle.
FLUX.2 treats typography as a headline capability rather than a known weakness, and Black Forest Labs publishes actual syntax for it. Applied properly you get product ads, magazine covers, posters and UI mockups with type that survives being read.
Sourcing note: every technique below is documented by Black Forest Labs in their typography, JSON prompting and prompting guides. The step-count trade-off is BFL's own description of FLUX.2 [flex]. Last verified July 30, 2026.
What this guide solves
The pain point: people prompt for design the way they prompt for photos — describing a vibe and hoping the text lands. Typography needs specification, not description, and the specification has a documented format.
The differentiator: the syntax rules together with the two things that decide design output beyond wording — which model tier to use, and when to switch from prose to a JSON-structured prompt.
The five text rules
BFL's documented technique for readable text:
- Quote the exact string — "The text 'OPEN' appears in red neon letters above the door"
- Specify placement — where the text sits relative to everything else
- Describe the style — "elegant serif typography", "bold industrial lettering", "handwritten script"
- Set the size — "large headline text", "small body copy", "medium subheading"
- Set the colour — including hex codes for brand text
The quotation marks matter most. They mark a literal string rather than a description of a concept, which is the difference between the model rendering your headline and the model rendering the idea of a headline.
Hex colours, bound to objects
FLUX.2 supports precise colour matching by hex code. The documented syntax signals the code with a keyword:
"the logo text 'ACME' in color #FF5733"
Gradients work by naming both ends:
"the color of the vase is a gradient, starting with color #02eb3c and finishing with color #edfa3c"
The documented constraint: hex codes work best when clearly associated with a specific object. A vague "use #FF0000 somewhere" produces inconsistent results.
For brand work that means binding every code to a named element — the logo text, the background panel, the accent bar — rather than listing a palette and hoping.
Switch to JSON for structured layouts
For infographics and layout-driven design, BFL documents structured prompts. The pattern:
- a
typeandtitle - a
subtitlefor supporting context sections, each with a heading, content and a visual- a
color_schemearray of hex values - a
styledescriptor such as "modern, clean, corporate"
The same shape works for product flat-lays: a scene, a subjects array where each subject has a description and colours, plus style and lighting.
When to use it: when the output has structure — defined regions, repeated section shapes, data laid out in a grid. For a single photographic subject, plain natural language is better. The dividing line is whether you would draw a wireframe for it.
Trying these on real work? Flux 3 AI is a browser workspace for poster concepts, ad variants and campaign frames, with upscaling for delivery. Open the image generator or see the plans.
Pick the right tier
This is the part most typography guides skip entirely.
FLUX.2 [flex] exposes a steps parameter, and BFL describes the trade explicitly: steps trade typography accuracy against latency. Their published comparison runs 6, 20 and 50 steps — more steps, better type, slower generation.
That makes [flex] the tier for design work specifically. BFL positions it as specialised for typography and preserving small details, at $0.05/MP.
| Tier | When to use it for design |
|---|---|
| FLUX.2 [flex] | Typography-critical work; tune steps up for accuracy |
| FLUX.2 [max] | Highest overall fidelity, $0.07 first MP |
| FLUX.2 [pro] | Volume design production, $0.03 first MP |
| klein 4B | Fast iteration on layout before committing, $0.014 |
A practical loop: rough the layout on klein, finalise on flex with steps raised. Cheap exploration, accurate delivery.
What it is good for
BFL documents FLUX.2 generating clean typography for product marketing materials, magazine layouts and posters, plus infographics and UI mockups.
Their own example prompt for a product ad carries a headline string, a subtext string, a described product close-up, background treatment and an aesthetic — all in one medium-length prompt. That is the shape to copy for any marketing asset:
[Product] advertisement, '[HEADLINE]' headline, '[subtext sentence]' subtext, [product description and framing], [background treatment], [aesthetic]
Prompt in the target language
A documented behaviour that matters for localised campaigns: FLUX.2 understands multiple languages, and prompting in the native language of the content often produces more culturally authentic results — local architecture, atmosphere and market detail rendered more accurately.
For a French campaign, write the prompt in French. This is documented model behaviour, not superstition.
Length still applies
The general prompt rules do not stop applying because you are doing design work: 30–80 words for most projects, subject first, no negative prompts. A typography prompt that balloons past 80 words to specify eight text elements is usually a sign the job wants a JSON-structured prompt instead.
FAQ
How do I get exact text in a FLUX image? Put the string in quotation marks and specify placement, style, size and colour.
Can I use my brand hex colours? Yes, bound to a named element rather than mentioned loosely.
Which FLUX model is best for typography? FLUX.2 [flex] — it exposes a steps parameter that trades typography accuracy against latency.
Do more steps really improve text? BFL's published comparison shows typography accuracy improving from 6 to 20 to 50 steps, at the cost of speed.
When should I use a JSON prompt? When the output has structure — infographics, data layouts, defined regions. Not for a single photographic subject.
Does prompting in another language help? For culturally specific content, BFL documents that it often does.
Can FLUX do UI mockups? Yes, it is one of BFL's documented use cases alongside infographics and magazine layouts.
Bottom line
Typography with FLUX is specification work. Quote the string, place it, style it, size it, colour it — and bind hex codes to objects. When the layout has structure, stop writing prose and send JSON.
Then pick the tier deliberately: rough on klein, finish on [flex] with steps raised. That combination is what turns "AI made a poster" into something you would actually put a brand name on.
Try it in the Flux 3 AI workspace with your own headline and brand colours.
Sources
- Typography & Design (BFL documentation) — text rendering tips, documented design use cases
- Official FLUX.2 prompting guide — quoted strings, hex binding, length guidance, multilingual behaviour
- JSON Structured Prompting (BFL docs) — structured prompt format
- Infographics (BFL docs) — layout-driven generation
- HEX Color Code Prompting (BFL docs) — colour binding rules
- UI Mockups (BFL docs) — interface layout generation
- Multi-Language Prompting (BFL docs) — native-language prompting behaviour
- FLUX.2 pro / flex model page — flex positioned for typography and fine detail
- FLUX.2: Frontier Visual Intelligence (BFL, November 25, 2025) — steps parameter trading typography accuracy against latency
- BFL API pricing — per-tier rates for design workloads
Scope note: techniques and tiers as of July 30, 2026. Flux 3 AI is an independent creator workspace, not affiliated with Black Forest Labs.
