.ai

les meilleurs modèles d'IA open source

Rechercher

FLUX.2 Dev: prompting the newest Black Forest Labs model

Q: Fine text has swapped or missing letters

Too many words or very small text relative to the canvas. Limit each text element to a few words, increase the canvas size and describe the text as large or prominent.

Q: A FLUX.1 LoRA produces noise or no effect

FLUX.1 and FLUX.2 weights are not compatible. Pick a LoRA from the Flux2 family in the catalog.

Sujet : Modèles

Par Captain

Publié le 2026-10-03

FLUX.2 Dev brings larger prompts, better text, stronger realism and multi-reference editing. This guide covers the turbo workflow, guidance and resolution settings, prompt structure and ideas that use its new strengths.

Cette page n'est pas encore traduite ; elle est affichée en anglais.

Aperçu

FLUX.2 Dev

is Black Forest Labs' second-generation model. It is substantially larger than

FLUX.1

, uses a modern language-model text encoder, renders text and fine detail more reliably and understands long structured prompts. The catalog workflow runs it with a turbo

LoRA

so a generation finishes in a handful of steps.

Where FLUX.1 was the first model that read whole sentences,

FLUX.2

reads whole paragraphs: lists of materials, lighting setups and layout instructions are followed closely, which makes it suited to product, poster and architectural work.

Référence

Nom	Type	Rôle
Prompt	text	Long structured descriptions work. Separate subject, materials, lighting, camera and layout with full stops.
Guidance	2.5-4	Around 3 is a good start; lower for photographic softness.
Steps	turbo	The catalog workflow is sped up with a turbo LoRA; it needs only a few steps and more do not help.
Resolution	1-4 MP	Larger canvases than FLUX.1 are stable; 1536x1024 is a good working size.
LoRAs	Flux2 family	Use LoRAs tagged Flux2 in the catalog, not FLUX.1 LoRAs.

Aucune installation requise – exécutez les modèles IA dans le cloud

BitVector Prism est la façon la plus simple de commencer : choisissez un modèle, tapez un prompt et générez dans une application web épurée, sans rien configurer. BitVector est aussi disponible sur Discord et sur le web (SpyGlass).

Essayer BitVector

Pas à pas

Describe the image in three or four full sentences: subject and action, materials and colors, light, then camera or medium.
Generate at about 1.5 megapixels; the model keeps detail at that size without tiling.
For text, quote the exact words and state where they sit (on the top third of the poster).
Compare guidance 2.5 and 3.5 on the same seed and keep the one whose colors you prefer.
Add one Flux2 LoRA at 0.8 if you want a particular photographic or illustrative signature.

Exemples

Poster with text

A concert poster for a jazz trio. Bold cream serif letters read "MIDNIGHT TRIO" across the top third. Below, a double bass, trumpet and drum kit drawn in loose ink wash on indigo paper. Small white text at the bottom reads "FRIDAY 9PM". Flat graphic lighting, printed texture.

Architectural visualization

Exterior view of a small timber library on a lakeshore at dusk. Warm interior light glows through floor-to-ceiling glass, charred cedar cladding, a thin steel roof, reflections in still water. Wide-angle architectural photography, 24mm, long exposure.

Creative idea: catalog of invented objects

A clean studio catalog photo of three ceramic teapots shaped like sleeping animals (a fox, a hare, a hedgehog), matte glaze in sage, rust and cream, arranged in a row on a pale oak table, soft overhead light, product photography.

Astuces

FLUX.2 understands layout language: foreground, midground, background, left third, negative space.
Mention the light source and its color temperature; the model renders mixed lighting convincingly.
For photoreal people, add a camera body and lens and a word about skin texture; it reduces the polished look.
Keep the Flux2 LoRA count low; the turbo LoRA already occupies part of the budget.

Dépannage

Fine text has swapped or missing letters

Pourquoi cela arrive

Too many words or very small text relative to the canvas.

Comment le corriger

Limit each text element to a few words, increase the canvas size and describe the text as large or prominent.

A FLUX.1 LoRA produces noise or no effect

Pourquoi cela arrive

FLUX.1 and FLUX.2 weights are not compatible.

Comment le corriger

Pick a LoRA from the Flux2 family in the catalog.

Aucune installation requise – exécutez les modèles IA dans le cloud

PirateDiffusion est réservé à Telegram et conçu pour les pros : des milliers de modèles, LoRA et workflows pilotés par commandes de chat, avec génération illimitée à prix fixe.