How to Use Bria FIBO 2.0
Bria FIBO 2.0 generates photorealistic images from a text prompt and returns the structured JSON prompt it used, giving you controllable and reproducible results.
Basic Generation
Send a descriptive prompt covering subject, setting, lighting, and camera. FIBO expands it into a roughly 1,000-word structured prompt and then renders the image. Only prompt is required; every response also returns the structured_prompt and the seed.
Choosing Parameters
- •resolution:
1MP (fast) or 4MP (fine photoreal detail, about 30 seconds slower).
- •aspect_ratio:
1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, or 16:9 — use 3:4 or 4:5 for portraits and 16:9 for scenes.
- •style_id: only
photoreal is supported, and it applies to prompt-based generation only.
- •output_type:
png for quality or jpeg for smaller files.
- •seed: any integer from
0 to 2147483647 for deterministic output.
Working Modes
- •Generate: send
prompt alone for text-to-image.
- •Inspire: pass a single reference image (JPEG, PNG, or WEBP) via
image_urls to draw from an existing look.
- •Refine or Recreate: pass a
structured_prompt from a previous response with its seed; add a short prompt to change only specific attributes. A structured prompt cannot be combined with a reference image.
Recommended Settings
Tested configuration for a photoreal portrait:
- •prompt: detailed subject plus lighting and lens description
- •resolution:
4MP
- •aspect_ratio:
3:4
- •seed:
123456
- •output_type:
png
This produced a sharp 1536x2048 portrait with detailed skin texture, catchlights, and natural background bokeh. For faster drafts, drop to 1MP. Keep the seed and the returned structured prompt to reproduce a shot or refine one attribute at a time.
Examples to try
Each example below ran end to end through FIBO at the exact settings shown on its card. Every response also returns the structured_prompt and the seed, so any of these shots can be reproduced or refined one attribute at a time; resolution, aspect ratio, and framing are set per card.
- •Precise multi-object composition. Prompt: "A white wooden windowsill in soft daylight with five red apples in a row; the second apple from the left is replaced by a single green pear; a small brass bell sits on top of the third apple from the right; sheer curtain behind, shallow depth of field". The render places exactly the right objects in the right order — four red apples with a single green pear second from the left and a brass bell balanced on the middle apple — against a sheer-curtain window. Tip: state counts and placement explicitly ("second from the left", "on top of"); FIBO writes them into the structured prompt, so the layout replays with the returned seed.
- •In-image text. Prompt: "A wall-mounted chalkboard cafe menu with neat white chalk lettering reading "FLAT WHITE 4", "CORTADO 3.5" and "CROISSANT 3" on three lines, a small chalk drawing of a coffee cup, warm pendant light, slight film grain". All three menu lines spell correctly in clean chalk, next to a little chalk coffee-cup sketch, on a wood-framed board against brick. Tip: put the exact words in quotation marks and keep each line short; a 3:4 or 4:5 frame gives menus and signage room to stay legible.
- •Material and texture (macro). Prompt: "Extreme macro photograph of a brushed titanium dive watch resting on wet dark slate, water beads on the sapphire crystal, visible machining marks on the bezel, hard directional light, 100mm macro". Brushed-metal grain, crisp bezel knurling, and rounded water beads on the crystal all hold up under hard light on wet slate. Tip: name the material and the lens ("brushed titanium", "100mm macro") and set
style_id to photoreal for product-grade realism.
- •Anatomy — hands. Prompt: "A potter's hands shaping a spinning clay bowl on a wheel, wet clay slip on the fingers, clear knuckles and fingernails, side window light, 50mm, shallow depth of field". Both hands read correctly — five fingers each, natural knuckles, and clay-smeared nails cupping a freshly thrown bowl. Tip: when hands matter, describe what they are doing and call out "knuckles and fingernails", and keep the subject mid-frame so the model has room to resolve them.
- •Photoreal portrait at 4MP. Prompt: "A close-up head-and-shoulders portrait of an older Lisbon tram conductor in a dark uniform and flat cap, grey stubble and deep smile lines, standing in the open doorway of a bright yellow tram, 85mm lens, shallow depth of field, overcast daylight". At 4MP the face holds fine detail — skin pores, deep smile lines, and individual grey beard hairs — with catchlights in the eyes while the yellow tram falls into clean background bokeh. Tip: switch
resolution to 4MP and use a 3:4 or 4:5 frame for portraits (it adds about 30 seconds), then keep the returned seed and structured prompt to reproduce the shot or adjust one feature later.
Reference-guided
FIBO's Inspire workflow takes a single reference image through image_urls (JPEG, PNG, or WEBP), passed alongside a short prompt; a reference image cannot be combined with a structured_prompt.
- •Inspire from a reference image. Prompt: "An advertising hero shot of this perfume bottle standing on black volcanic rock with thin mist drifting at its base, dramatic rim light, dark background". The frosted-glass bottle and brushed-gold cap from the reference carry into a new scene — volcanic rock, drifting mist, and a rim-lit dark background — without redrawing the product. Tip: let the reference fix the object and let the prompt describe only how the new scene should differ.
Prompts this model is built for, worth trying next: "the same bottle on a pale marble ledge beside fresh eucalyptus", "a studio headshot of a ceramicist at 4MP in a 3:4 frame", "a chalkboard specials board reading 'SOUP OF THE DAY 6'", and "three labelled test tubes left to right: red, then clear, then blue".