Image generation

FLUX.2 [dev]

Name: FLUX.2 [dev]
Author: Black Forest Labs

32-billion-parameter open-weights checkpoint that does text-to-image plus single- and multi-reference editing in one model, released 25 November 2025 on Hugging Face.

An image model by Black Forest Labs · read 2026-08-14 · sourcing high

Resolution: 4 MP for image editing (FLUX.2 family, per BFL launch post); no hard ceiling published
Weights: Downloadable
API: Yes
Commercial use: The licence forbids commercial use

Runs on consumer GPUs with fp8 optimisation, though the reference repo cites H100-class VRAM for the full model. Distributed under a non-commercial license, with a paid commercial license sold separately.

Commercial use — The licence forbids commercial use

FLUX Non-Commercial License. Commercial use of the weights or their outputs requires a paid licence from Black Forest Labs.

This entry was written by hand from the licence wording quoted above, because the source dataset's yes/no field was misleading here. Read the licence before you rely on any of this. Nothing on this page is legal advice, and this site does not tell you a model is cleared for commercial use.

Specification

Vendor figures, read 2026-08-14Read from Black Forest Labs's own documentation on 2026-08-14. Vendors change these between model versions without announcing it, so check theirs before you commit a shot to a number. Sourcing for this entry is strong: the figures came from pages we retrieved and read. Check the vendor's current documentation

What it takes in: text, image
What it makes: image
Maximum resolution: 4 MP for image editing (FLUX.2 family, per BFL launch post); no hard ceiling published
Aspect ratios: not established
Controls: image-to-image, reference-character-consistency, style-reference, text-rendering, editing-by-instruction, control-net-structure
Weights: Downloadable
API: Yes
Pricing: weights free to download; BFL docs list [dev] as local-only (no hosted per-image rate). Commercial licensing sold in tiers (e.g. bundles including 100K images/month).
Released or updated: 2025-11-25

Supported aspect ratios is not established here. That means no page we could retrieve stated it, not that the model lacks one.

Strengths

Frontier-level open weights for generation and reference editing in one checkpoint
Self-hostable for data-sensitive pipelines
Fine-tunable / LoRA-trainable via base checkpoints

Weaknesses

Non-commercial license blocks production use without a paid license
32B parameters means heavy VRAM requirements
Third-party host availability varies in licensing clarity

Best for

Research, prototyping and self-hosted experimentation before committing to a paid BFL commercial tier.

Where the sources disagree

Recorded rather than resolved. Picking the tidier number would hide the fact that the vendor's own pages do not agree.

The FLUX.2 launch post calls [dev] 'open-weight (commercial licensing available)' while the Hugging Face card and repo state the FLUX Non-Commercial License. Both are consistent if read as: default license is non-commercial, commercial use requires purchase.

What to understand before using it

Explainers naming the mechanisms this model's own documentation mentions.

Open weights versus closed API: renting or owning the model names “open weights”
Character and set consistency: the weakest link names “consistency”
Temporal consistency: flicker, drift and morphing names “consistency”
Camera control: a move without a camera names “control”

How to prompt it

Written for FLUX.2 [dev]. See prompt craft for the full vocabulary and cinematic language for the lens, light and mood tables.

Flux 2.0 prompt guidance
Precise prompt adherence; a creative alternative to the Banana family; key-pose generation in the sprite pipeline; cheap photorealistic drafts.

Key-pose prompts carry the STYLE FORMULA verbatim plus one pose instruction, and must demand full body in frame with empty margin above the head and below the feet plus a clean uniform key-colour background absent from the subject's palette.

MCP servers that front Black Forest Labs

Servers in the media MCP directory whose own description or homepage names this vendor. Absence of a server here is not evidence that none exists: most publish no tool list at all.

Adclip Image generation
Ad creative generation via MCP, copy and static images from a JSON brief. Keyless.
BananaBanana Image & Video Generation Video generation
Generate images & video (Nano Banana, Veo, Omni) pay-as-you-go. No subscription, crypto payments.
Civitai Image generation
Civitai MCP.
Clipia Image generation
Generate AI images, video, speech, music and presentations from Claude, ChatGPT and Cursor.
Cochlea Audio and voice
Render, analyze, and verify audio (WAV/FLAC/mp3/ogg) offline and deterministically via MCP tools.
ComfyUI Builder Image generation
ComfyUI MCP server: knowledge base, workflow builder, real-time execution, models, plugins.

See it from the vendor

Shotcraft hosts no sample output. Judging a model means watching its own showreel with your own eyes, and a clip re-hosted here would be a clip we chose.

Nearby in the directory

FLUX.1 Kontext [pro] / [max]

Instruction-driven image editing models from the FLUX.1 generation (2025), still exposed as live /flux-kontext-pro and /flux-kontext-max endpoints. Th

Image generation

FLUX.2 [flex]

Hosted FLUX.2 variant that exposes sampling steps and guidance scale as API parameters, letting callers trade quality and prompt adherence against spe

Image generation

FLUX.2 [klein] (4B / 9B)

Size-distilled open-weights members of the FLUX.2 family, released January 2026 in 4B and 9B sizes plus 'Base' checkpoints intended for fine-tuning an

Image generation

Sources

Read on 2026-08-14. A source that has changed since then will not be reflected here until the next pass.