Images

Flux 3 Text-to-Image: posters, labels and scenes that read

Flux 3 Text-to-Image is built for detailed composition and readable lettering. Five habits, three prompts, and when to pick something cheaper.

4 min read
A clear glass light bulb standing on a dark stone, its filament glowing violet, with a faint purple haze in the near-black background

Look at the two examples on the Flux 3 Text-to-Image page before you write anything. One is a robot repair shop on a rainy night, with a neon sign that reads "SECOND CHANCE". The other is a perfume campaign with a bottle labelled "SOL" and, in smaller type, "EAU DE PARFUM". Both prompts are a full paragraph long. That is the model's personality in one glance: Black Forest Labs pitch it on detailed composition and typography, and it rewards a prompt written like a brief.

So instead of a feature tour, here are five habits that suit it, each with what goes wrong without it.

1. Write the words in quotes, and keep them short

Put every piece of text you want in the picture inside quotes and say where it lives: on a sign, a label, a book spine. Short phrases are the safest bet: a title, a name, a two-line tagline at most. Without quotes, the model may treat your title as a description rather than as letters to draw.

Prompt
A vintage travel poster for a mountain railway. A red train crosses a stone viaduct above a deep green valley at golden hour, snowy peaks behind. Bold cream lettering across the top reads "ALPINE LINE", and a small line at the bottom reads "SINCE 1912". Flat screen-print colours, visible paper grain.
Two text elements, each with a place and a style.

2. Describe the type, not just the words

"Elegant gold lettering", "hand-painted sign letters", "condensed sans serif in white": the font's character is part of the scene, and the model listens. If you need room for real text set later in a layout tool, ask for empty space instead, as in our post on leaving space for text.

3. Compose like a photographer

The examples spell out who sits where, what is in the foreground, what the light does. Do the same: subject, then setting, then foreground and background, then light, then style. The field's own hint asks for exactly that order of things.

Prompt
A cosy independent bookshop at dusk, seen from the pavement through the front window. Inside, a grey cat sleeps on a stack of hardbacks on the display table. Warm lamps glow between tall shelves. Hand-painted gold letters on the glass read "NIGHT OWL BOOKS". Light rain on the pavement reflects the window. Shot on a 35mm lens, shallow depth of field, cinematic colour.
A scene with one line of text, painted on the glass.

4. Draft cheap, finish once

Resolution has three steps, and the smallest is the default. Higher steps cost more and the top one can take several minutes, so do your exploring at the default and only re-run the winner bigger. For rough ideas about layout and colour you can go cheaper still: a few drafts on Fast Image cost less than one run here (our Fast Image post explains the trade-off). Bring the composition you liked back to Flux 3 for the version with real lettering.

Prompt
Product shot of a matte black coffee bag standing on a pale concrete counter, morning side light, a few roasted beans scattered in front. The bag's front label reads "DARK HOURS" in bold white condensed letters, with "single origin" in small type below. Clean, minimal, plenty of empty counter on the right.
A packaging mock-up. The empty counter leaves room for a price or a headline later.

5. Choose the frame and the format before you run

  • Aspect ratio defaults to square. The list goes from very wide banners to tall phone-screen shapes; a poster wants a tall ratio, a website hero a wide one. Choosing it first saves a crop that cuts your lettering. Our aspect ratio guide has the common sizes per use.
  • Output format (Advanced): JPEG by default, PNG when you will edit the picture further.
  • Prompt expansion (Advanced, off by default) fleshes out a short prompt. Turn it on for a one-line idea; keep it off when your quoted text and layout are already exact, so nothing gets reinterpreted.

Where it falls short

It takes text only: there is no field for a reference picture, so it cannot keep a specific face, product or character the same across a series. Use the edit model for that. And readable is not the same as perfect. Long sentences, tiny print and unusual names are where letters still slip, so zoom in and check every word before anything goes to print; our post on fixing hands and text covers the repair. For a printed piece, the event poster guide walks through the rest. The price for your settings shows on the Generate button in the studio.

Questions

Can Flux 3 Text-to-Image write text in a picture?

Yes, typography is one of its strengths. Put the words in quotes, keep each phrase short, and check the spelling before you use the picture.

Can I give it a reference image?

No, it takes a text prompt only. To change or combine existing pictures, use Flux 3 Image Edit.

Which resolution should I choose?

Start at the default step while you explore, then re-run your favourite at a higher one. Higher steps cost more and take longer.

#Images#Models#Typography

Keep reading

All articles →