Bria Fibo 1.5 on WaveSpeedAI: Generate 1.5 and Edit 1.5 Versus the Fibo Family

Bria’s Fibo 1.5 arrived on WaveSpeedAI on August 31, 2026 as two endpoints: bria/fibo-generate-1.5/text-to-image and bria/fibo-edit-1.5/edit. Both cost a flat $0.04 per image, the same as the eight earlier Fibo endpoints. What you get for that price is a distilled model that samples in 4 to 6 steps, a 4MP output tier, reference-image guidance on generation, and multi-image input on edits. What you lose is the negative prompt.

This post lays out the parameters of both new endpoints, what 1.5 changes relative to FIBO, how they sit against the existing bria/fibo family, and which endpoint to reach for. All parameters and prices are from the live WaveSpeedAI schemas on September 5, 2026.

What Fibo 1.5 is

FIBO is Bria’s 8B-parameter, JSON-native text-to-image model, announced in November 2025 and published as open source on GitHub. Its distinguishing idea is structured prompting: long, schema-shaped captions that separately describe subject, lighting, composition, color, and camera, so a change to one attribute does not disturb the others.

Fibo 1.5, per the Bria model card on Hugging Face, is a two-stage post-training of that model. A DMD (distribution matching distillation) stage collapses the sampler to 4 to 6 guidance-free steps; a DMD-R stage then refines the distilled student to recover realism and texture. The original ran 50 steps at guidance scale 5. The card also notes that negative prompts have no effect on 1.5, that the weights are not intended for fine-tuning, and that the JSON-native structured prompting carries over unchanged.

On training data, Bria’s position is consistent across its product page and the model card: FIBO is trained exclusively on licensed data, curated from a pool of roughly one billion images, with each asset verified for commercial use and attribution traceability. That is Bria’s claim, and it is the reason the family is positioned for brand and enterprise work. That is Bria’s claim; WaveSpeedAI does not independently verify the training set.

Fibo Generate 1.5: text-to-image

bria/fibo-generate-1.5/text-to-image, $0.04 per image.

Parameter	Required	Values	Default	Notes
prompt	yes	string		Natural-language description.
aspect_ratio	no	1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9	1:1
resolution	no	1mp, 4mp	1mp	New in 1.5. Same price at either tier.
images	no	array, max 1 URL		New in 1.5. One reference image to guide subject, composition, or style.
style_id	no	photoreal	unset	New in 1.5. The only preset currently exposed.
structured_prompt	no	JSON string		Bria’s VGL/JSON schema for precise control.
seed	no	integer	-1	-1 means random.
enable_sync_mode	no	boolean	false	Wait for the result in the same response when it is ready inside the sync window.

Compared with the original bria/fibo endpoint, the additions are resolution, images, and style_id; the removal is negative_prompt. bria/fibo also exposes enable_base64_output, which 1.5 does not.

A minimal request:

{
  "prompt": "studio product shot of a matte black ceramic mug on a walnut table, soft window light from the left, shallow depth of field",
  "aspect_ratio": "4:5",
  "resolution": "4mp",
  "style_id": "photoreal",
  "seed": 42
}

Fibo Edit 1.5: image editing

bria/fibo-edit-1.5/edit, $0.04 per edit.

Parameter	Required	Values	Default	Notes
images	yes	array, 1 to 4 URLs		Ordered. The first is the image being edited; the rest are visual context.
prompt	no	string		The edit instruction. Optional because references alone can drive an edit.
mask_image	no	URL		Single-image edits only. White is edited, black is preserved.
aspect_ratio	no	same nine values as generate		New in 1.5. Lets the output canvas differ from the input.
structured_instruction	no	JSON string		Structured edit instruction for detailed control.
seed	no	integer	-1
enable_sync_mode	no	boolean	false

The meaningful change from bria/fibo/edit is the input count. The original schema states that images must contain exactly one item; 1.5 takes one to four ordered images, so you can hand it a product, a background reference, and a style reference in one call. aspect_ratio on an edit endpoint is also new. As with generation, negative_prompt is gone, and structured_prompt has been renamed structured_instruction.

{
  "images": [
    "https://example.com/product.jpg",
    "https://example.com/beach-reference.jpg"
  ],
  "prompt": "place the product on wet sand at golden hour, matching the light and color of the second image",
  "aspect_ratio": "16:9"
}

The whole Fibo family on WaveSpeedAI

Every Fibo endpoint is $0.04 per image with no formula, so the choice is about capability, not cost.

Endpoint	Input	What it does	Prompt	Notes
bria/fibo-generate-1.5/text-to-image	prompt, optional 1 reference	Generate	free text or JSON	1MP or 4MP, photoreal preset
bria/fibo	prompt	Generate	free text or JSON, negative prompt	Original FIBO
bria/fibo-edit-1.5/edit	1 to 4 images	General edit	free text or JSON	mask optional, aspect ratio
bria/fibo/edit	1 image	General edit	free text or JSON, negative prompt	mask optional
bria/fibo-edit/genfill	image, mask, prompt	Fill a masked region	free text	all three required
bria/fibo/relight	image	Change lighting	none	light_type from 13 presets, light_direction front, side, bottom, top-down
bria/fibo/reseason	image	Change season	none	season: spring, summer, autumn, winter
bria/fibo/restore	image	Restore old or damaged photos	none	no other parameters
bria/fibo/colorize	image	Colorize or recolor	none	style: contemporary color, vivid color, black and white colors, sepia vintage
bria/fibo/image-blend	image, prompt	Blend or composite by instruction	free text

The five single-purpose endpoints (relight, reseason, restore, colorize, image-blend) are worth keeping in mind precisely because they take no prompt, or a very short one. They are deterministic switches, which makes them easy to put behind a UI control and easy to batch.

When to use which

New generation work: Fibo Generate 1.5. It is the same price as bria/fibo, adds the 4MP tier and a reference image, and Bria’s model card says the distilled sampler runs in 4 to 6 steps rather than 50, which is the kind of difference that shows up in queue time. The one reason to stay on bria/fibo is if your prompts depend on negative_prompt; 1.5 has nowhere to put it and Bria says it would be ignored anyway. Move that guidance into the positive prompt or into a structured_prompt.

Edits that need more than one image: Fibo Edit 1.5. Product-in-scene composites, style transfer from a reference, and anything where you would otherwise have to describe a look in words all benefit from passing the look as an image.

Simple single-image edits with a mask: either. bria/fibo/edit and bria/fibo-edit-1.5/edit both accept a mask and both cost $0.04. If you need a negative prompt, use the original.

Filling a region with something new: bria/fibo-edit/genfill. It is stricter than a masked edit: image, mask, and prompt are all required, and the prompt describes only what goes inside the mask.

One-control adjustments: the single-purpose endpoints. Do not write a relighting prompt when bria/fibo/relight with light_type: "blue hour light" will do it repeatably.

Reproducibility across the family

Every generate and edit endpoint accepts seed, and the structured prompt fields exist so that a campaign can hold composition constant while changing one attribute at a time. On 1.5, pair a fixed seed with structured_prompt on generation, or structured_instruction on edits, and change one field per iteration. That workflow is where FIBO’s JSON-native design earns its keep, and it works identically on 1.5.

For background on the editing endpoint this builds on, see Introducing Bria Fibo Edit on WaveSpeedAI, and for the relight preset list in practice, Introducing Bria Fibo Relight on WaveSpeedAI.

Try the new endpoints from their model pages: Fibo Generate 1.5 text-to-image and Fibo Edit 1.5. The original bria/fibo and bria/fibo/edit remain available at the same price.