Managing Forma Board rendering "look and feel"

juri_pelkonenE4XP5 · ‎05-18-2026

Are there any best practises, tip &tricks and such for keeping the Forma Board AI rendering output consistent?

It seems that the produced rendering style varies quite a lot, and that's not good in a project that has tens of images to render.

I know this feature is not ment to produce high quality and marketing level stuff, but as usual, we like to keep things quick and simple here at SARA, and Forma Board has proven to be an absolute win for our "quick and dirty" phase of project development.

PS: I love it how I can finally demonstrate with ease what our golf resort site will look like during a zombie outbreak induced by an eruption.

samarvir_singh_bhayana · ‎05-21-2026

Hey,

Thanks for the question.

I have been using a prompt template (shared below) with an inspirational image and it has worked well for me.

Workflow

1. Use chatgpt or similar and give Template prompt with your studio's past project image or inspirational image to create a reusable prompt.
2. Use this new Prompt with Precise Pro model.

Please let me know if you have any questions.

Best,
Samarvir singh
Product Lead (Forma Board)

Template prompt -

# =============================================================================

# AEC View Style — TEMPLATE (v2)

# -----------------------------------------------------------------------------

# Mental model: this file is CSS for an AEC view.

#

# AEC model ─► HTML (the WHAT — geometry, programme, semantics)

# This file ─► CSS (the HOW — surfaces, light, atmosphere, image)

#

# CSS does not know whether a `<div>` is a navigation, an article, or a

# card. It only knows colour, font, border, filter. This file follows the

# same discipline: it says nothing about what the model IS (residential,

# hospital, master-plan, concept-phase, planning-submission). It only

# says how any model dropped into it should LOOK.

#

# If a field tempts you to describe the project — its programme, its

# phase, its audience, its location, its scale — that field belongs in a

# different document (see AEC_VISUALISATION.md), not here.

# -----------------------------------------------------------------------------

# Schema : alpha

# Filename : <kebab-name>.yaml

# Consumers : LLM-driven render pipelines, real-time engines, prompt

# composers. The bottom of this file shows the canonical

# YAML → prompt translation.

# =============================================================================

# The preset is identified by its FILENAME (kebab-case, .yaml). No

# `name` or `description` fields — the look documents itself.

# ---------------------------------------------------------------------------

# 1. style (CSS analogue: theme)

# ---------------------------------------------------------------------------

# The rendering register. Top-level visual decision; every other field

# is interpreted in its context.

#

# photoreal full PBR, calibrated light

# conceptual simplified materials, expressive light

# watercolour non-photoreal painterly, hand-drawn feel

# sketch line-driven, minimal colour

# white-model untextured grey + glass

# photomontage composited into real site photo

# line-drawing orthographic / axonometric line work

# technical-illustration construction-style cutaway

# ---------------------------------------------------------------------------

style: <enum>

# ---------------------------------------------------------------------------

# 2. lighting (CSS analogue: color)

# ---------------------------------------------------------------------------

# The light SOURCE. What the sun is doing.

# ---------------------------------------------------------------------------

lighting:

timeOfDay: <enum> # dawn | morning | midday | afternoon |

# golden-hour | dusk | twilight | night

weather: <enum> # clear | partly-cloudy | overcast |

# rain | snow | fog

sunAzimuth: <deg> # 0-360, optional unless analytical

sunAltitude: <deg> # 0-90, optional unless analytical

mood: <free-text> # calm | dramatic | optimistic | sombre |

# warm | clinical | etc.

# ---------------------------------------------------------------------------

# 3. surroundings (CSS analogue: background)

# ---------------------------------------------------------------------------

# What the model is rendered AGAINST. This is the visual backdrop, NOT

# the project's location. "Tropical coast" and "dense urban" are visual

# scenes; "Caribbean cruise port" or "downtown LA office tower" are

# projects and DO NOT belong here.

# ---------------------------------------------------------------------------

surroundings:

scene: <enum> # rural | suburban | dense-urban |

# historic-core | tropical-coastal |

# waterfront | desert | alpine |

# landscape | industrial | abstract-void

vegetation: <free-text> # "mature deciduous", "palm-dominant", "none"

terrain: <free-text> # "flat", "hilly-coastal", "granite outcrop"

water: <free-text> # optional, "tidal river", "fjord", "none"

ground-cover: <free-text> # optional, "stone setts", "moss", "sand"

sky: <free-text> # optional, "blue-violet gradient", "high cumulus"

# ---------------------------------------------------------------------------

# 4. materials (CSS analogue: per-class background)

# ---------------------------------------------------------------------------

# Surface treatment per slot. Slots are open-ended — add or drop as the

# look demands. Values may be free text, a `{token.ref}` from a sister

# DESIGN.md, or a vendor asset id.

#

# At `style: white-model` this block is typically omitted entirely.

# ---------------------------------------------------------------------------

materials:

facade: <free-text | token-ref>

glazing: <free-text | token-ref>

roof: <free-text | token-ref>

ground: <free-text | token-ref>

accent: <free-text | token-ref>

landscape-soft: <free-text | token-ref>

# additional slots permitted: interior-floor, interior-wall,

# interior-ceiling, water-feature, signage, etc.

# ---------------------------------------------------------------------------

# 5. palette (CSS analogue: filter: hue-rotate /

# saturate / contrast)

# ---------------------------------------------------------------------------

# Image-level colour treatment. Applied AFTER lighting and materials

# resolve. This is the global colour grade.

# ---------------------------------------------------------------------------

palette:

temperature: <enum> # cool | neutral | warm

saturation: <enum> # desaturated | muted | natural | vivid

contrast: <enum> # low | natural | high

dominant-hue: <free-text> # optional, "blue-violet", "amber", "monochrome"

# ---------------------------------------------------------------------------

# 6. atmosphere (CSS analogue: filter: blur /

# backdrop-filter)

# ---------------------------------------------------------------------------

# What's in the AIR between camera and model. Affects depth perception

# and softness without touching surfaces or light source.

# ---------------------------------------------------------------------------

atmosphere:

haze: <enum> # none | light | heavy

depth-fog: <enum> # none | subtle | strong

visibility: <free-text> # optional, "infinite", "300m", "milky"

# ---------------------------------------------------------------------------

# 7. post-processing (CSS analogue: filter: grain / blur /

# drop-shadow / box-shadow)

# ---------------------------------------------------------------------------

# Screen-space effects added on top of the rendered image.

# ---------------------------------------------------------------------------

post-processing:

exposure: <enum> # underexposed | calibrated | overexposed

grain: <enum> # none | subtle | film | heavy

bloom: <enum> # none | subtle | strong

vignette: <enum> # none | subtle | strong

edge-treatment: <enum> # soft | natural | sharp | hand-drawn-line

# =============================================================================

# Prompt assembly (canonical)

# -----------------------------------------------------------------------------

# Pipelines convert this YAML to a natural-language prompt using the rule

# below. The model, camera, and output format come from the pipeline,

# not from this file.

#

# Template:

#

# "Render the supplied model as a {style} image. Place it in

# {surroundings.scene} surroundings — {surroundings.vegetation},

# {surroundings.terrain}{, water if present}{, ground-cover if present}.

# Sky: {surroundings.sky}.

# Light it as {lighting.timeOfDay}, {lighting.weather};

# mood: {lighting.mood}.

# Surfaces: {materials.facade} facades, {materials.glazing} glazing,

# {materials.roof} roofs, {materials.ground} ground, {materials.accent}

# accents, {materials.landscape-soft} planting.

# Colour grade: {palette.temperature}, {palette.saturation},

# {palette.contrast}{, dominant-hue if present}.

# Atmosphere: {atmosphere.haze} haze, {atmosphere.particulate},

# {atmosphere.depth-fog} depth fog.

# Post: {post-processing.exposure} exposure, {post-processing.grain}

# grain, {post-processing.bloom} bloom, {post-processing.vignette}

# vignette, {post-processing.edge-treatment} edges."

#

# A pipeline is free to use a richer template; this is the minimum

# faithful translation.

# =============================================================================

kevin_vong1 · ‎06-11-2026

Hi there! Just checking in to see if the solution above worked for you. If it answered your question, please consider marking it as the accepted solution so it can help other community members find the answer more easily.

Community

Managing Forma Board rendering &quot;look and feel&quot;