Bria Best Practices

Q: What’s the simplest mental model for Bria image generation?

A: Think of it as a two-step system:Prompt / reference image → structured_prompt (JSON “blueprint”) → deterministic render.The big win is that you can store and reuse the structured blueprint (plus a seed) to get consistent results and controlled variations.

Q: Do the generation endpoints return images immediately?

A: By default, no. All Bria v2 endpoints are asynchronous — the API returns a request_id and a status_url immediately, and you poll that URL until the result is ready. If you need a synchronous response, pass sync: true in your request. This is useful for simpler integrations but not recommended for high-throughput workflows.

Q: What’s the best general way to write prompts for high-quality images?

A: Use a “shot spec” prompt, like a mini creative brief:

1. Subject (what)

2. Environment (where)

3. Composition (close-up/wide, centered, negative space)

4. Lighting (soft studio, golden hour, neon, etc.)

5. Style/medium (photoreal, illustration, 3D, etc.)

6. Constraints (what must stay the same; what to avoid; required text)

Q: Should I write long prompts with lots of adjectives (“ultra, 8K, masterpiece”)?

A: Usually no. Better results come from concrete visual controls: composition, lighting, camera feel, materials, and clear constraints, rather than stacks of vague superlatives.

Q: What’s the best way to refine without the image “drifting”?

A: This is the pro workflow:

1. Generate once from text (or text+image).

2. Save the returned structured_prompt and seed.

3. Refine using structured_prompt + prompt + seed, changing one thing at a time.

This keeps composition and style stable while you make targeted improvements. Bria explicitly supports refinement via structured prompt inputs.

Q: When should I lock the seed?

A: Lock the seed when you’re evaluating prompt changes and want apples-to-apples comparisons.Change the seed when the “blueprint” is good and you want variety.

Q: How do I get better photorealism?

A: Add specifics that photographers care about:

• lighting type and direction (“soft diffused key light from upper left”)

• composition (“3/4 angle, centered, negative space on right”)

• material cues (“matte ceramic, crisp reflections”)

• background (“white seamless, subtle shadow”)

• Resolution - set the ‘resolution’ parameter to ‘4MP’

These “visual controls” are commonly recommended across modern image generation prompt guides.

Q: How do I reliably generate images with text in them?

A: Put the exact text verbatim in quotes and specify:

• placement (centered/top-left/etc.)

• typography expectations (bold sans-serif, balloon lettering, etc.)

• clarity (“crisp edges, high contrast”)

Bria’s own examples show quoted text usage.

Q: When should I use /v2/image/generate?

A: Use it when you want best quality, nuance, and instruction adherence. Bria describes the standard pipeline as the higher-fidelity option.

Q: What input combinations does /v2/image/generate support?

A: The endpoint supports five distinct flows:

• Text to image — prompt only. Use when starting from scratch with a creative brief.

• Image to image — images only (single image). Use when a reference image should drive the visual direction.

• Image + text — images (single image) + prompt. Use when you want a reference image as a base with text to steer the output.

• Recreate — structured_prompt + seed. Use to reproduce a previously generated image exactly.

• Refine — structured_prompt + prompt + seed. Use to make targeted changes to a previous result without composition drift.

Q: What’s the best workflow to explore results for top results curation on /v2/image/generate?

A: Use a three-phase workflow:

Phase A — Explore quickly

• resolution: "1MP"

• fewer steps (e.g., 35–40)

• strong “shot spec” prompt

Phase B — Lock composition

• save structured_prompt and seed from the best result

• refine with structured_prompt + prompt (small deltas)

Phase C — Final render

• bump to resolution: "4MP" for the final

• increase steps to 50 if you need more detail

Bria documents 1MP vs 4MP and steps range, and supports structured_prompt-based refinement.

Q: When should I use structured_prompt instead of a fresh prompt?

A: Use structured_prompt when you want:

• reproducibility (save the blueprint)

• consistency across many images

• safe, minimal refinements without composition drift

Bria supports recreation with structured_prompt (and seed) and refinement with structured_prompt + prompt.

Q: How should I choose aspect_ratio for better composition?

A: Choose it based on the real output placement:

• 1:1 product tiles

• 4:5 social feed

• 9:16 stories/reels

• 16:9 banners/hero

In addition, the following aspect ratios are also supported: “2:3”, “3:2”, “3:4”, “4:3”, “5:4”

Q: When should I pin model_version?

A: Pin it for production stability and regression testing. Leave it unpinned if you want automatic improvements (but accept drift risk). Bria notes model version can be specified or omitted for default.

Q: How is Lite different from the Standard pipeline?

A: The input combinations, flows, and parameters are identical. The differences are:

• VLM bridge: Uses the open-source FIBO-VLM instead of Gemini 2.5 Flash — lower fidelity on complex inputs

• Image model: Uses the distilled Fibo Lite model — faster inference, lower quality ceiling

• On-prem deployment: Fully local deployment is supported (Standard pipeline cannot run on-prem)

• Missing parameters: resolution, steps_num, and negative_prompt are not available in Lite

Q: When should I choose Lite over Standard?

A: Choose Lite when speed or data privacy is the priority, or when the workflow requires on-prem deployment. For maximum visual fidelity and instruction adherence, use Standard.

Q: What does this endpoint actually do?

A: It runs only the first half of the generation pipeline — the translation step. It takes your input and returns a structured_prompt (JSON blueprint) without generating an image. The Standard pipeline uses Gemini 2.5 Flash as the VLM bridge, giving you the highest-fidelity interpretation of complex or nuanced inputs.

This is the same translation step that happens automatically inside /v2/image/generate — this endpoint just surfaces it so you can inspect, edit, or version the JSON before committing to a render.

Q: What input combinations does it support?

A: The endpoint supports four flows:

• From text — prompt only. Returns a structured prompt derived from your text brief.

• From image — images only (single image). Returns a structured prompt derived from a reference image.

• From image + text — images (single image) + prompt. Returns a structured prompt shaped by both the reference image and text guidance.

• Refine existing — structured_prompt + prompt. Returns an updated structured prompt with targeted changes applied.

Q: Why generate a structured prompt separately rather than letting /v2/image/generate handle it automatically?

A: Three main reasons:

1. Human-in-the-loop review — inspect or edit the JSON before it drives a visual. Useful for compliance checks, brand consistency review, or catching prompt misinterpretations before spending generation time.

2. Programmatic editing and prompt libraries — build UIs where users tweak a JSON spec, or maintain a reusable library of structured prompts for campaigns and brand templates.

3. Hybrid deployment — use Bria's state-of-the-art VLM bridge via API while self-hosting the open-source FIBO image model on your own private cloud.

Q: How do I write inputs that produce better structured prompts?

A: Treat your input like a creative brief. The VLM bridge responds well to specifics:

• Subject — describe it clearly, including materials, colors, and scale

• Composition — camera angle, framing, where the subject sits in the frame

• Lighting — direction, quality (soft/hard), time of day

• Constraints — what must stay unchanged, what to avoid

The more concrete your input, the less the bridge has to guess — and the less you'll need to correct in the resulting JSON.

Q: When should I use the Lite structured prompt endpoint?

A: The Lite equivalent of /v2/structured_prompt/generate. Supports the same four input flows and use cases, but uses the open-source FIBO-VLM bridge instead of Gemini 2.5 Flash — lower fidelity on complex inputs, but faster and fully compatible with on-prem deployment.

Use this endpoint when your generation will run on the Lite pipeline. The resulting structured_prompt should be passed to /v2/image/generate/lite, not the Standard endpoint.

Q: When do I use generate_from_diff?

A: Use it when your product allows users to directly edit structured prompt JSON, and you want Bria to:

• interpret the semantic diff

• output an optimized structured prompt reflecting the user’s change

Bria describes this endpoint for diff-based structured prompt generation. (docs.bria.ai)

Q: What’s the best way to edit structured prompts safely?

A: Encourage:

• small targeted edits (lighting, composition, background)

• not full rewrites

• lock seed while testing “did this change do what I expect?”

This aligns with Bria’s “structured, disentangled control” philosophy. (docs.bria.ai)