Skip to content
ReviewRamble

The weekly brief

AI, reviewed, in your inbox.

  • New reviews of tools, models, MCP servers and repos, with the verdict up front
  • Re-checked prices and ratings when something changes
  • New analysis pieces as they are published

Models, agents, MCP servers and the tools actually worth your time.

One email a week. Unsubscribe with the link in any brief. How we use your email

Fibo 2.0

Bria's 8B open editor tops the open-weights editing board and is trained only on licensed images, but the weights are non-commercial.

Reviewed by Andrew Miller · Updated

Visit Fibo 2.0 Open source

Our rating

4.0/5

Fibo 2.0 edits as well as any model you can download, and it is the one whose training data a cautious legal team can trace.

What is Fibo 2.0?

Fibo 2.0 was released by Bria on 20 October 2026. It is an 8-billion-parameter flow-matching model that works in the FLUX.2 VAE's latent space, with Qwen3-VL-4B-Instruct as its text and image encoder. It is the first Fibo model to generate and edit in one set of weights: text-to-image, instruction editing, masked editing and edits from up to 4 reference images.

Two checkpoints are on Hugging Face: Fibo 2.0 Dev, the full 50-step base for fine-tuning, and Fibo 2.0 Edit Turbo, distilled to edit in 4 steps without guidance. Both are gated and released under CC BY-NC 4.0 for non-commercial use; commercial use needs an agreement with Bria or its hosted API. Bria says the model was trained only on licensed data from its library of about 1 billion images.

How much does Fibo 2.0 cost?

PlanPriceWhat you get
WeightsFreeCC BY-NC 4.0, non-commercial
fal$0.04 per imageCommercial use allowed
Bria API$32 per 1,000 imagesArtificial Analysis listing

Weights free for non-commercial use under CC BY-NC 4.0; commercial self-hosting by agreement with Bria (price not published). Hosted: $0.04 per image on fal; Artificial Analysis lists $32 per 1,000 images on Bria's API.

Is Fibo 2.0 worth it?

Fibo 2.0 is the strongest open image editor right now, by a statistically thin margin. Artificial Analysis puts it at 1085 Elo on its editing board, twenty-fifth of 79 and first among 22 open-weights models, ahead of Qwen-Image-2.1 at 1080 and FLUX 3 Image Klein at 1069. That is 173 points above Bria's previous FIBO Edit, though the board cannot yet separate it from Qwen-Image-2.1.

Its real distinction is provenance. Bria trains only on licensed images, so a brand or agency worried about where a model's pictures came from has a clearer answer than with Qwen, FLUX or Hunyuan. Hosted output carries commercial rights, and at $0.04 an image on fal it costs a little more than Nano Banana 2.1's $0.0336, which scores 67 points higher on the same board.

The open weights are for research. CC BY-NC 4.0 rules out commercial self-hosting without a deal, which puts it in the same position as Qwen-Image-2.1 and FLUX.2 [dev]; Qwen-Image-Edit-2511 remains the Apache-licensed fallback, 93 points lower. Output is about 1 megapixel, against 4 megapixels for FLUX.2 [dev], and references stop at 4 against 10 for Klein.

Two practical details stand out. The model was trained on long structured JSON captions, and Bria's bria-vgl tool that turns short prompts into them calls Google's Gemini, so prompts leave your servers unless you write the JSON yourself. Edit Turbo's 4 steps make it fast; Bria reports under 2 seconds for the model on its API, a figure fal says it has not verified.

Where it earns its place

  • An in-house creative team at a consumer-goods brand

    Edit campaign images with a model whose training data legal has signed off

    Call Bria's API or fal at about $0.04 an image, which carries commercial rights, and point legal to Bria's licensed-data claim. 10,000 edits a month cost about $400.

  • A computer-vision researcher

    Fine-tune an open editor for a narrow domain such as interior design

    Start from Fibo 2.0 Dev, which Bria says supports single-GPU LoRA training, under the non-commercial licence. The 8B size is far easier to train than 32B FLUX.2 [dev].

  • An e-commerce engineer replacing a background-swap pipeline

    Change only the backdrop of product photos and leave the product untouched

    Masked editing keeps pixels outside the mask, which suits catalogue work. Self-hosting it commercially needs a Bria agreement; until then, Edit Turbo on the hosted API is the licensed route.

Before you commit

Setup
Request access on Hugging Face, install diffusers 0.42.0 or later and load BriaFibo2Pipeline; Dev uses 50 steps at guidance 5.0, Edit Turbo 4 steps at guidance 1.0, at about 1024 x 1024.
Integrations
Diffusers, ComfyUI nodes, Draw Things, DiffusionBee, a Bria agent skill, Bria API, fal and WaveSpeed.
Data and privacy
Trained only on licensed data, per Bria. Weights CC BY-NC 4.0 and gated; commercial use needs a Bria agreement or the API. The bria-vgl prompt tool calls Gemini and needs a Gemini API key.
Cost at scale
At fal's $0.04 an image, 100,000 edits cost about $4,000 a month; at the $32 per 1,000 Artificial Analysis lists for Bria's API, about $3,200. Commercial self-hosting price not published.
Limits
About 1-megapixel output; up to 4 images per request; masks only on single-image edits; no VRAM figure published; multi-turn editing listed as coming soon.

What’s good

  • First among open models on Artificial Analysis's editing board at 1085 Elo
  • Trained only on licensed images
  • Masked, instruction and multi-reference editing in one 8B model
  • 4-step Edit Turbo checkpoint for fast editing

What’s not

  • Weights are CC BY-NC 4.0; commercial self-hosting by negotiation
  • About 1-megapixel output and at most 4 reference images
  • Prompt expansion tool sends prompts to Gemini
  • Lead over Qwen-Image-2.1 is within the board's margin

Best for

Brands and agencies that want the best open editor with traceable training data, using Bria's API in production and the weights for evaluation.

Fibo 2.0 alternatives

Fibo 2.0: common questions

Is Fibo 2.0 worth it?
Fibo 2.0 edits as well as any model you can download, and it is the one whose training data a cautious legal team can trace. We rate it 4.0 out of 5.
How much does Fibo 2.0 cost?
Weights free for non-commercial use under CC BY-NC 4.0; commercial self-hosting by agreement with Bria (price not published). Hosted: $0.04 per image on fal; Artificial Analysis lists $32 per 1,000 images on Bria's API.
What are the downsides of Fibo 2.0?
Weights are CC BY-NC 4.0; commercial self-hosting by negotiation; about 1-megapixel output and at most 4 reference images; prompt expansion tool sends prompts to Gemini; lead over Qwen-Image-2.1 is within the board's margin.
Who is Fibo 2.0 best for?
Brands and agencies that want the best open editor with traceable training data, using Bria's API in production and the weights for evaluation.
How much does Fibo 2.0 cost for a team or in production?
At fal's $0.04 an image, 100,000 edits cost about $4,000 a month; at the $32 per 1,000 Artificial Analysis lists for Bria's API, about $3,200. Commercial self-hosting price not published.
Is Fibo 2.0 safe for company data?
Trained only on licensed data, per Bria. Weights CC BY-NC 4.0 and gated; commercial use needs a Bria agreement or the API. The bria-vgl prompt tool calls Gemini and needs a Gemini API key.
What are the best Fibo 2.0 alternatives?
The closest alternatives we have reviewed are Qwen-Image-2.1, FLUX 3 Image Klein, Qwen-Image-Edit-2511 and FLUX.2 [dev].

Answers checked Nov 7, 2026.

Make Fibo 2.0? Show this rating on your siteFibo 2.0 rated 4.0 out of 5 on ReviewRamble