新規登録でプレミアムプランが割引に割引を受け取る
eroq

ブログ/models·2026年10月8日·6分·執筆:eroqチーム

Flux 3 Image guide — natural light and clean typography on eroq

Flux 3 Image is Black Forest Labs' latest FLUX on eroq. How to prompt natural light, unstaged photos and reliable in-image text, with prices and examples.


Flux 3 Image is the latest FLUX from Black Forest Labs, and on eroq it is the generalist with good taste in light. It renders natural photoreal scenes, with window light that falls off the way real light does and skin that looks like skin, and it handles in-image text reliably. Two jobs suit it best: photography that should not look over-styled, and typography on real surfaces such as labels, signs and packaging. If you are unsure which model to use for a photograph, it is a sensible default.

The numbers

  • Price: 32 credits / image; at the entry pack's rate of about a cent a credit, that is roughly $0.32.
  • Speed: typically about 20 seconds a take (the studio's planning estimate).
  • Takes: batch one to four; 128 credits buys a full run of four, and failed takes refund on their own.
  • Formats: 1:1, 3:4, 4:3, 9:16 and 16:9, at the same price.
  • Inputs: text only. Flux 3 has no reference-picture slot and no Edit mode on eroq.
  • Plan: any plan; nobody needs a subscription to render it.
  • Content: moderated upstream by its provider, with refusals returned as content_blocked at no cost.
  • API id: flux-3-image.
  • The cheaper neighbor: Flux Image, eroq's high-fidelity photoreal engine, at 12 credits / image.

Describe the light before the subject

The most useful habit with Flux 3 is to open with the light and the setting, then bring in the subject. When light comes first, the scene tends to get built around it, and the result reads as a photograph rather than a subject pasted into a backdrop. Compare:

  • "A man reading a newspaper in a café, soft window light."
  • "Soft light from a café's tall front window falls across a corner table where a man reads a newspaper."

Same content; the second one tells the model where the light comes from, what it touches and where the shadows go. Treat light as a physical thing with four properties:

  • Source: a window, a bare bulb, the sun, a phone screen, a fridge left open.
  • Direction: from behind, from the left, from overhead, low and raking.
  • Quality: hard or soft, diffused by curtains, broken by blinds.
  • Color: warm tungsten, cool overcast, the orange of the last half hour of sun.

Photographs that do not look staged

The fastest way to an over-styled image is to ask for one. For a natural result:

  • Ask for ordinary moments: "candid", "unposed", "mid-laugh", "caught turning toward the door".
  • Name the camera and lens: "35mm", "50mm at f/2". They anchor a photographic register.
  • Name the register: "documentary photograph", "lifestyle photograph", "film photo with slight grain".
  • Add a small imperfection: a crumpled napkin, a fingerprint on the glass, a chipped mug.
  • Skip the hype words. "Ultra-detailed, masterpiece, 8k" push toward the glossy look you are trying to avoid.

Typography that sits on a surface

Flux 3 is one of the lineup's reliable text models, and the trick is to treat text as an object in the scene, not as a caption floating over it:

  • Quote the exact words: the label reads "WILD HONEY".
  • Name the type category: geometric sans, slab serif, script, stencil, hand-painted.
  • Name the surface and the process: printed on a kraft-paper label, painted on brick, embossed in leather, bent in neon tubing, chalked on a board, foil-stamped on a box.
  • Keep it short: one or two text elements per image give the cleanest result.
  • Light the text. A sign in deep shadow is a sign nobody can read.

For prompt sets built around this, see Flux 3 prompts for signage, logos, labels and packaging, and for how it stacks up against OpenAI's text specialist, the GPT Image 2.5 vs Flux 3 typography showdown.

Example prompts

  1. Morning portrait, 3:4. Light first, an unposed moment, a lens and a film register.

    Soft early light through a bedroom window falls across a woman in a grey knit sweater sitting on the edge of the bed, holding a mug in both hands and looking outside, unposed, natural skin, 50mm, film photo with slight grain.

  2. Backlit street, 16:9. Golden hour from the end of the street, long shadows and a candid subject.

    Late golden-hour light from the end of the street backlights a cyclist riding past shuttered shops, long shadows across the pavement, a dog trotting beside the bike, candid documentary photograph, 35mm.

  3. Honey jar label, 1:1. Text on a real label, with the type style and the binding named.

    Warm side light from a farmhouse window on a glass jar of honey on a wooden table, a kraft-paper label tied with twine reading "WILD HONEY" in a hand-stamped serif, a honey dipper resting on a linen cloth, shallow depth of field.

  4. Diner neon, 4:3. Lettering made of light, which also lights the scene.

    A rainy night outside a small diner, the window glowing, a red neon sign in the window spelling "OPEN LATE" in script tubing, reflections in the puddles on the sidewalk, a waitress wiping the counter inside, 35mm photograph.

  5. Boatworks wall, 4:3. Faded hand-painted letters under flat overcast light, which keeps them legible.

    Overcast daylight on an old brick workshop wall with a hand-painted sign reading "HARBOR BOATWORKS" in faded cream letters, a wooden rowboat on trestles below it, coils of rope on the ground, documentary photograph.

  6. Can lineup, 16:9. The hardest one here, with four strings on three objects: render four takes and keep the cleanest.

    Three cans of sparkling water in mint green, coral and lemon yellow standing on a white plinth, each printed with "FIZZ" in a bold geometric sans, with the flavor below in small capitals: "MINT", "BERRY", "CITRUS", soft even studio light, product photography.

Flux 3 Image or Flux Image?

Flux Image is eroq's high-fidelity photoreal engine at 12 credits / image, 20 credits less per image than Flux 3. For photoreal detail from a prompt with no lettering in the frame, it remains a strong, cheaper choice. Flux 3 is the pick when the light has to feel natural and when there are words to render. The full comparison is in Flux 3 Image vs Flux Image.

The rest of the neighborhood: Seedream 5.0 Pro and Nano Banana Pro go further on raw detail, GPT Image 2.5 Sunburst is the pick for strict layouts and posters, and Eroq One is the answer when the product or face has to be yours, from reference pictures. When you are ready, open the image studio, choose Flux 3 Image and start your prompt with the light.

FAQ

Is Flux 3 Image better than Flux Image?

For natural light and typography, yes: that is what eroq positions it for. Flux Image is cheaper and still a good choice for photoreal detail without text in the frame.

Can Flux 3 Image render text on packaging and signs?

Yes, it renders in-image text reliably. Quote the exact words, name the type style and the surface (a label, a painted wall, a neon tube), keep it to one or two short text elements and make sure the text is lit.

How long does Flux 3 Image take per image?

Typically about 20 seconds a take. That is a planning estimate: a busy moment can add queue time.

Does Flux 3 Image do image-to-image on eroq?

No. On eroq it is text-to-image only, with no reference pictures and no Edit mode. To change an existing picture or match your own product, use Eroq One, which edits from a plain-language instruction and reads up to ten references.

タグfluxblack-forest-labsphotorealtypographyimage-models

この記事で使ったモデルで作ってみましょう。無料の50クレジットで始めるか、全エンジンと料金をご覧ください。