Bria Fibo 2.0: API Pricing, Dev vs Edit Turbo and Licensing

Bria announced Fibo 2.0 on October 20, 2026, bringing image generation and editing into one model family. Developers can use its hosted API or explore downloadable weights. The important distinctions are which variant handles which task, how structured editing works and what commercial use requires.

The model is also live in Bria Create.

Fibo 2.0 Dev vs Edit Turbo

The two checkpoints serve different workflows. Fibo 2.0 Dev is the non-distilled base model for generation, editing and fine-tuning. Edit Turbo is the distilled editing model.

Variant	Main use	Inference approach
Fibo 2.0 Dev	Text-to-image, editing and fine-tuning	Full model with classifier-free guidance
Fibo 2.0 Edit Turbo	Image editing	Four steps, without classifier-free guidance

Edit Turbo is editing-only. Someone looking for fast text-to-image generation should not treat it as a drop-in replacement for Dev.

Bria says the family is trained exclusively on licensed data. Its model cards also report first place in image editing among open-weight models on Artificial Analysis as of October 20. That is an attributed, dated category claim, not evidence that Fibo leads every closed model or every editing task.

For a product team, a preference ranking is only a starting point. An ecommerce workflow may care more about keeping a particular label, material or silhouette unchanged than about which output looks better in a general comparison.

How the Fibo editing API works

Bria's editing API documentation describes two stages. A vision-language bridge converts an instruction into structured JSON, then Fibo renders the edit. Developers can supply their own structured instruction instead of relying on the bridge each time.

The endpoint accepts one to four source images. Native masking is available for a single image, but not for multi-reference requests. Masking generates replacement content within the selected region; it is not a conventional pixel-adjustment operation.

To recreate an output, Bria instructs developers to retain the original images, any mask, the returned structured instruction and the seed. Keeping only the user's sentence is therefore an incomplete record of the request.

This separation creates a useful debugging path. If an edit goes wrong, a team can inspect what the bridge asked the renderer to do before deciding whether the rendering model failed. It also makes it possible to review or constrain instructions before an expensive generation step.

Fibo API pricing

Bria's Fibo 2 release notes map the new model to the existing generation and editing endpoints, without a code migration. The current Bria pricing page lists Fibo generation and Fibo Edit at $0.032 per image. It lists Fibo Structured Prompt separately at $0.02 per call. These are the published API line items, not a quote for running the downloadable models on your own hardware.

At the listed image rate, 1,000 image operations cost $32 before any separately billed operations. A workflow that makes additional structured-prompt calls needs to account for those too. The platform's general “starting from” price is not the Fibo rate.

The page advertises access to all models in its free tier, with 100 generations and ten requests per minute. Pay-as-you-go lists 60 requests per minute. These are published platform-tier limits, not measured Fibo rendering throughput; check account-specific limits before planning a batch.

Open weights do not mean unrestricted commercial use

The model cards provide access for noncommercial use under Bria's model terms. Commercial use requires the appropriate Bria agreement; a downloadable checkpoint is not a blanket commercial license. Enterprise pricing is custom.

There is also a clear release boundary. Both model cards describe multi-turn editing with pixel-stable preservation as coming soon. It should not be promised as an available capability simply because the current model accepts editing requests.

Fibo 2.0 is most interesting where a team needs an inspectable editing process and can evaluate preservation on its own assets. The decision should include the whole workflow: instruction quality, rendering quality, repeated-call cost and deployment terms.

I’m Paras. I work at the intersection of AI, engineering and distribution.

I help technical founders turn product launches and implementation details into clear stories that developers and buyers can understand and share.

Book a 30-minute call.